Seller automated engine architecture and methodology for optimized pricing strategies in automated real-time iterative reverse auctions over the internet and the like for the purchase and sale of goods and services

Abstract

An improved seller automated engine architecture methodology particularly (though not exclusively) for use in automated real-time iterative reverse auctions over the Internet and the like for the purchase and sale of goods and services, providing a choice of architectural implementations while enabling price optimization on market share-directed considerations, specific sales target-directed implementations, seller utility derivative-following implementations, model optimizer implementations and explorations, mathematical optimization-oriented and rules-based implementations.

Description

where

• • u_R is the revenue utility
  • u_M is the profit utility
  • u_S is the sales volume utility
  • w_R is the revenue utility weight
  • w_M is the profit utility weight, and
  • w_S is the sales volume utility weight, etc.

These weights are typically configured by the seller. Some sellers may want to assign equal weights to targets such as revenue, profit, sales volume etc. Others may value one more than others; yet others may change the weights over time and as market conditions evolve. For example, for a retailer with seasonal apparel, sales volume target may be much more important than revenue or profit at the end of the season. If a product is selling below expectations, the seller can increase the sales volume weight substantially over the weights of other targets. If not configured by the seller, equal values for the weights may be used. The configurable weights provide the ability for sellers to assign emphasis (or de-emphasis) to each of their targets.

In the later-described target-directed implementation of FIGS. 14 and 15 herein, the utilities are functions of respective “distances” to targets, the “distance” being the quantitative measure of how far away the seller is from the specified target. For example, if the target revenue for Q1 is $10 m, and on February 1, the current revenue is $3 m, then the “distance” for the revenue is a function of ($3 m/$10 m).

Different forms of utility functions may be used by each seller subject to its own unique objectives. The utility function provides the ability for the seller to specify the pace/aggressiveness with which to approach the target. Some examples will now be presented, through the system of the invention allows the use of any utility function desired.

Linear Utility Function

The linear utility function is computed as the achieved revenue, profit and sales volume etc so far, normalized by respective targets in the specified time interval. For example, the revenue utility is (at time “t”),

$\begin{array}{cc}{u}_R=\frac{R_t}{R_T}& \left(1\right)\end{array}$
where R_T is the pre-specified target revenue over time interval T, and R_t is the current revenue achieved at time “t≦T”.
This utility can be further weighted with respect to other utilities by using seller configuration as shown in the previous section.

The linear utility can also be represented as a function of the “distance” to the respective targets. For example, the distance of revenue from the revenue target and a simple corresponding linear utility can be defined as

$\begin{array}{cc}{d}_R=\frac{R_t}{R_T}-1u_R=1+d_R.& \left(2\right)\end{array}$
The weighted distance is then
d=w_Rd_R+w_Md_M+w_Sd_S.  (3)
The distance “d” is a quantitative, yet intuitive measure of how far the seller is from his targets. For example when the current revenue (R_t) is 0, the revenue distance d_R=−1; and so long as the revenue is less than the revenue target, the revenue distance is negative indicating that the target is not yet achieved. When R_t=R_T (i.e., the seller has achieved the target), the revenue distance=0. As the seller exceeds the target, the corresponding “d” becomes positive. The total linear utility is thus,
u=1+d.  (4)
This is a simple form of the linear utility function, with a more general form being
u=1+ad.  (5)
This simple form of the linear utility function is illustrated in FIG. 8. With such a linear utility function, the seller approaches the target with a constant slope indicating that the gain in utility from reducing the distance by one unit is constant no matter how far the seller is from the target. The slope parameter “a” indicates how aggressively the seller approaches the target.

When a new auction request is received, the expected utility from winning the auction is

$\begin{array}{cc}u=w_R\frac{R_t+Q\rho \left(p\right)p}{R_T}+w_M\frac{M_t+Q\rho \left(p\right)\left(p-C\right)}{M_T}+w_S\frac{S_t+Q\rho \left(p\right)}{S_T},& \left(6\right)\end{array}$
where

• • w_R is: Revenue utility weights.
  • w_M Profit utility weights.
  • w_S Sales volume utility weights.
  • Q Number of units
  • p Unit quote price be seller. This is the variable being optimized.
  • ρ(p) Historic bid-response function, typically a representation of recent historic market price “p” computed using a seller preferred duration of history.
  • C Unit cost for seller as configured by seller.
  • R_t Achieved revenue so far.
  • R_T Revenue target as configured by seller, where T is the total time to achieve the target.
  • M_t Achieved profit so far.
  • M_T Profit target as configured by seller where T is the total time to achieve the target.
  • S_t Achieved sales volume so far.
  • S_T Sales volume target as configured by seller where T is the total time to achieve the target.
    The optimal price is computed by maximizing this utility equation.

The linear utility function can be considered as a risk-neutral way to reach the target. This means that no matter how far away the seller is from the targets, the additional gain in winning the next auction request does not change. This form of utility function is typically a reference function and is easy to use and provides diversification to changing market conditions.

Piecewise Linear Utility Function

The piecewise linear utility function allows the seller to change his or its utility gain at arbitrarily chosen discrete distances to the target. This enables further optimization for the more sophisticated seller. A sample piecewise linear utility curve is shown in FIG. 9(a).

In this example, using the previous definition of “distance”, the piecewise linear utility function can be defined as
u=1+d,  (7)
if d≦0; and as
u=1+ad,  (8)
if d>0.
“a” is a factor to measure aggressiveness after the target has been reached. If “a” is chosen to be <1, it is less aggressive; and if a>1, it is more aggressive. In general, the total utility is a weighted combination of several (two, in the above example) piecewise linear utilities. In FIG. 9(a), parameter “a” is the additional utility gain when the achievement doubles the target, assuming the utility is 1 when the target is just met. In the figure, a=0.4. The seller configures this value as desired.

It may be noted that the figure is just an example where the slope changes at d=0. This, however, could be any value of d, and in general, the slope could change at multiple values of d. FIG. 9(b) illustrates a piecewise linear utility function with three pieces, each with a different slope.

Exponential Utility Function

An exponential utility function has the form
u=A−Be^−ad  (9)
where A and B are scaling parameters, a is a factor to measure risk aversion and d is the normalized distance to the target. When d<0, it is below the target; when d=0, it is at the target; when, d>0 it is above the target. A sample exponential utility curve corresponding to the above equation is shown in FIG. 10(a).

This exponential utility has the several properties that may represent desired seller behavior. For example:

• • 1. When distance to the target is large, the additional gain in utility (represented by the slope of the curve) is also large. This enables the seller to approach the targets more aggressively when he is further away from the targets. The degree of aggressiveness is entirely seller selectable thus giving the seller countless configuration options to meet his own objectives.
  • 2. As the distance to the target goes to 0 from a negative number, the slope decreases, thus reducing the additional utility gain from winning the next auction. This represents a sellers state where things are going well, and the seller starts to become less aggressive.
  • 3. After the target is met, the slope further decreases. This reflects the type of seller who has very little incentive to exceed the given target.
    Parameter “a” is a factor to measure the aggressiveness when the seller pursues the target, and it is configured by the seller.

As an illustration, suppose the additional utility gain from doubling the target is α, (shown in FIG. 10(a)), parameters A, B and a are determined by solving the following equations:
A−B=1
A−Be^a=0
A−Be^−a=1+α.  (10)
The equations reflect three situations: 1) when d=−1, u=0; 2) when d=0; u=1; 3) when d=1, u=1+α.

The above-defined exponential utility function has the unique property that the smaller a, the more aggressive it is when pursuing the target before reaching the same target. When a gets bigger and bigger, it gets closer and closer to the linear utility function, and it becomes less and less aggressive below the target. It is this ability to go from highly exponential to linear that makes this approach very attractive.

The expected utility with one additional bid request is
u=w_R(A_R−B_Rexp(−a_Rd_R))+w_M(A_M−B_Mexp(−a_Md_M))+w_S(A_S−B_Sexp(−a_Sd_S)).  11)
Here A, B and a are exponential utility function parameters, and d_R, d_M and d_S are normalized distances to targets. They are defined as

$\begin{array}{cc}\begin{array}{c}{d}_R=\frac{R_t+Q\rho \left(p\right)p}{R_T}-1\\ d_M=\frac{M_t+Q\rho \left(p\right)\left(p-C\right)}{M_T}-1\\ d_S=\frac{S_t+Q\rho \left(p\right)}{S_T}-1.\end{array}& \left(12\right)\end{array}$
We can also define a weighted exponential distance to targets as
d=W_Rexp(−a_Rd_R)+W_Mexp(−a_Md_M)+W_Sexp(−a_Sd_S).  (13)

As before, the optimal price is computed by maximizing the weighted utility function. The maximization of the weighted exponential utility function is equivalent to the minimization of the weighted exponential distance to targets. When each request is received, the target-directed approach will try to minimize the distance to the target.

Another embodiment of the exponential utility type function is illustrated in FIG. 10(b).
u=−A+Be^ad.  (14)
This function has the unique property that it approaches the target less aggressively when the distance to the target is large, and gets more aggressive as the distance to the target decreases. Such properties may represent a desired seller behavior; for example:

• • 1. When the distance to the target is large, the additional gain in utility (represented by the slope of the curve) is small. This enables the seller to approach the targets slowly when he or it is further away from the targets. As before, the degree of aggressiveness is entirely seller selectable, thus giving the seller countless configuration options to meet his or its own objectives.
  • 2. As the distance to the target goes to 0 from a negative number, the slope increases, thus increasing the additional utility gained from winning the next bid. This represents a sellers state where the seller starts to become more aggressive.
    After the target is met, the slope further increases; reflecting that type of seller who has extremely high incentive to exceed the given target.

In addition, a piecewise exponential utility function may also be used as shown in FIG. 11 (a), which enables the seller to adjust aggressiveness as desired.

Hybrid Utility Function

As is evident from this discussion, the novel implementation of the invention allows any utility function or a combination of heterogeneous utility functions to be used (such as a piecewise exponential, a combination of piecewise linear or polynomial, and exponential, etc.) to represent each seller's unique objectives as the distance from the target varies. The selection of the utility functions can also be changed by the seller via configuration.

The hybrid utility function provides a powerful ability to vary the aggressiveness of the optimizer based on the distance to the target. The hybrid function consists of multiple pieces of linear and exponential utility functions as desired by the seller. As the distance to the target varies, different function forms can be used based on the seller's objectives and risk averseness. A sample hybrid utility function is illustrated in FIG. 11 (b).

The wide range of utility functions as well as the ability to run multiple optimizers with different utility functions gives the seller full coverage of the massive parameter space of all combinations of his or its objectives.

Target Allocation using Qualitative Demand Curve

The target-directed implementation of FIGS. 14 and 15 also provides the seller with the ability to be time and demand sensitive in pursuing the target. In case of a retailer, at the start of a quarter, the seller is likely far away from the quarterly targets. This is expected and the seller need not be overly aggressive in pricing products. As time progresses, e.g., the end of the quarter nears, if the seller is still very far from the target, then it may be prudent to pursue the target more aggressively; i.e., the additional gain in winning the subsequent auction would be considered high. There are numerous other factors that are time dependent that affect the desired seller's utility. For example, the seller may have explicit sub-goals for the holiday season; the seller may be aware of changing buying patterns over weekends vs weekdays; there may be an upcoming event such as a super-bowl or a snowstorm; an advertising campaign that may lead to increased/decrease demand and thus require specific targets; in a 24×7 environment, the demand may be different during day vs night, and even during various hours of the day. A qualitative demand curve configuration can provide the seller with the ability to capture the varying relative demand over time and further enhances the effectiveness of the utility function. The seller is asked optionally to configure a qualitative demand curve for the target period. The seller's configuration could be based on a graphical display input; such as the sample qualitative demand curve for a weekly target illustrated in FIG. 12. The demand curve provides a mechanism to measure and track the intermediate progress of the seller, as opposed to traditional techniques based on demand forecasting. In the figure, the demand gradually increases over the week and through day 6, while decreasing back down on day 7. The weekly target is then subdivided into daily targets based on the relative demand curve. This process may be further used to derive night/day or even hourly targets during which the demand is relatively uniform. Effectively, one can compute the area under the curve as a relative number and then appropriately map it to the quantitative sales targets for that specific duration.

The target-directed implementation finds the optimal price by maximizing the utility gain with respect to the distance to the target for the current sub-period. As time progresses from one sub-period to the next, the sales targets may be readjusted based on performance and seller configuration. If the seller does not achieve the weekly revenue targets for the first week of the month, for example, the remaining unmet target from the week may be rolled over in several ways, if so desired, such as to the next week, uniformly or non-uniformly distributed across the remaining week of the month, year etc. This is completely dependent on the seller configuration. If the seller chooses to distribute the unmet targets to the remaining sub-periods, the distance to the new (higher) target effectively increases thus making the optimizer more aggressive. As the end of the year approaches, all the unmet targets from before may be left to fulfill in the last month, if desired, resulting in potentially aggressively moving towards the target. The ability to use the demand curve to sub-divide targets provides intermediate check points, effectively minimizing of the seller's risk. If, however, the seller does not want to do this, all the unmet targets can be pushed to the end. The seller may also change the utility functions based on time to map against the demand curve. For example, they may choose to pursue the sub-targets less aggressively early in the quarter, while becoming more aggressive later in the quarter. Thus, by using the demand curve with target allocations, and weights, the target-directed approach can satisfy a wide variety of seller objectives.

Historic Bid-Response Functions

One of the parameters in the utility functions is ρ(p), referred to herein as the historic bid-response function, characterizing the recent historic market behavior in terms of price and participants in the context of a real-time [ARTIST] environment. The historic bid-response function is used to select a nominal reference price while taking into account the price variations of recent vintage. Such reference price allows the optimizer to start closer to the current market conditions. The target-directed optimizer converges to the market price based on the goal of minimizing the distance to the targets, even if sufficient historical data is not available to compute a sufficient bid response function. While meaningful parameters for the historic bid-response function help faster convergence, they are not necessary for correctness.

The parameters of the historic bid-response function can be computed statistically by fitting a curve to the available auction historical data, based on minimizing the squared errors or using maximum-likelihood estimates as later described. Further enhancement can be made by more heavily weighing more recent historical data during the estimation.

The historic bid-response function parameters may be computed either at the SAEJ or at the RAC. RAC may provide a service to all the SAEJs by periodically transmitting the parameters of well-known and/or common historic bid-response functions. Individual SAEJs may use the given bid-response parameters or may use other bid-response curves by based on seller configuration and/or historical data gathered by the individual SAEJs.

As noted earlier, the historic bid-response function is used as a nominal proxy for the recent market behavior. It allows the implementation quickly to compute an optimal price close to the market price. In case the computed bid-response parameters are not available, or if the market price has deviated from the last computation of bid-response parameters, the distance to the target enables the implementation to converge to the true market price.

Two earlier mentioned common methods for computing bid-response parameters that are uniquely applied in this solution will now be described—minimizing squared errors, and maximum likelihood estimations.

Minimizing the Squared Errors

The historic bid-response parameters are estimated by minimizing the squares of the error terms from the bid-response curve to the actual wins and losses of each auction opportunity within the historic window configured. For auction “1” the indicator variable W_i is assigned the value of 1 if the auction is won, and W_i is assigned the value of 0 if the auction is lost. Each auction i also has a weight W_i which expresses the confidence in the accuracy or relevance of the point. Thus, to determine the best estimates of the parameter values, the bid-response should be as close to W_i as possible. This can be achieved by solving the unconstrained optimization problem

$\begin{array}{cc}\underset{a,b,c,d}{\mathrm{Minimize}}\sum _i{w_i\left[\rho \left(p_i,Q_i,p_{\mathrm{ci}}|a,b,c,d\right)-W_i\right]}^2.& \left(16\right)\end{array}$
One statistic measuring the goodness-of-fit is the sum of squared errors (SSE) itself; that is,

$\begin{array}{cc}\mathrm{SSE}=\sum _i{w_i\left[\rho \left(p_i,Q_i,p_{\mathrm{ci}}\right)-W_i\right]}^2& \left(17\right)\end{array}$
Maximum-Likelihood Estimation

This implementation for computing the bid-response parameter values involves choosing parameter values that closely mimic the pattern of wins and losses resembling the actual historic outcomes.

Assuming all auctions are independent, the probability of realizing the actual outcome for the auction based on a particular bid i is
L_i(a,b,c,d)=W_iρ(p_i,Q_i,p_ci|a,b,c,d)+(1−W_i)[1−ρ(p_i,Q_i, p_ci|a,b,c,d)].  (18)
The parameters, therefore, can be chosen in such a way that they maximize the probability of realizing the actual outcome over all observations. This can be achieved by solving the equation

$\begin{array}{cc}\underset{a,b,c,d}{\mathrm{Maximize}}\prod _i{L}_i\left(a,b,c,d\right).& \left(19\right)\end{array}$
Further enhancement is made to the above embodiment based on natural logarithms of the likelihood, by solving the equation

$\begin{array}{cc}\underset{a,b,c,d}{\mathrm{Maximize}}\sum _i\ln \left[L_i\left(a,b,c,d\right)\right]& \left(20\right)\end{array}$

It is now in order to return to the previously discussed exemplary implementation of FIGS. 14 and 15.

The Logit Bid-Request Function

In the preferred embodiment of the invention, the ideal bid-response function, indeed, is the Logit function due to its unique characteristics. The general form of the Logit function is:

$\begin{array}{cc}\rho \left(p\right)=\frac{1}{1+e^{a+\mathrm{bp}+\mathrm{cQ}+{\mathrm{dp}}_c}},& \left(15\right)\end{array}$
where
p Unit price quoted by the seller.
ρ(p) Historic bid-response function at price p
Q Order size (number of units)
p_c Recent historic average unit price in the market,
and where a, b, c, d are scaling parameters for the Logit function. These parameters are used to measure the effects of various factors:

• • a Parameter related to non-price factors
  • b parameter related to price factors
  • c Parameter for the order size
  • d Parameter for the recent historic average market price
    A typical Logit bid-response function is shown in FIG. 12.

Variants of the above Logit function and other functions can also be used such as:

Power function:

$\rho \left(p\right)=\frac{\alpha }{\alpha +{\mathrm{rp}}^{\gamma }}$

• • Gompertz function: ρ(p)=a+c exp(−exp(−b(p−d)))
  • Von-Bertalanffy function: ρ(p)=(a^b−(a^b−c)e^{−bd(p−p0)})^1/b
  • Janoschek function: ρ(p)=1−exp(−bp^c)
    Exemplary Target-Directed Implementation
    (FIGS. 14 & 15)

A typical flowchart implementation for the illustrative sales target-directed implementation of FIGS. 14 and 15 follows:

Flow Chart For FIGS. 14 and 15

• • 1. Initialization
    • Set utility weights based on seller configuration at 32, FIG. 14.
    • Choose utility function at 30 and aggressiveness factors based on seller configuration.
    • Set the seller configured sales targets and duration of the targets at 31.
    • Use seller-configured qualitative demand curve created as shown in FIG. 13 to divide the target period into T sub-periods at 31D in FIG. 14.
    • Allocate targets to sub-periods at 31E, in proportion to such sub-period demands.
  • 2. Repeat for Each Sub-Period
    • Repeat for each RFB for a new auction (as configured by seller) Set historic bid-response function parameters at 40, FIG. 15, if updated values are available from RAC (41) or local computation
    • Compute at 42 achieved revenue, profit and sales volume so far Compute expected utility at 43 with this request against targets for this sub-period
    • Compute the optimal price at 44 by maximizing the expected utility and return the optimal price at 45.
  • 3. At the end of each sub-period, automatically readjust targets for the remaining sub-periods, basing on the over or under achievements in this period, subject to seller configuration.

It may be noted that, as with the rules-engine implementation of exemplary FIG. 7, above, presented, this implementation does not impose any restrictions on the price that is returned to the RAC. The price is returned via the integrator which may be configured in several ways such as, the following.

For RFB, the optimizer is invoked at the start of every new auction, and the price resulting from the optimizer computation, can be used as a temporary computed optimized floor (in case of the retail vertical) for the next bidding rounds in the same auction. This is an optimized sub-constraint clearly residing within the overall constraints configured by the seller. In this case, the integrator starts with a configured (or computed) ceiling price and iteratively drops the price to the temporary computed optimized floor or until the auction ends (whichever comes first). For RFQ, the options are simpler such as return the list price configured by the seller or return the optimal price at 45 as computed by the optimizer.

It should be understood that these are just exemplary embodiments of how the combination of optimizers and integrators can be used for quotes and auctions. There are, of course, numerous other embodiments based on the fundamental innovation of this invention in its seller automated engine computing prices/bids in real-time and based on a seller's widely varying set of objectives.

This sales target-directed implementation, however, makes two assumptions:

• • 1. For faster convergence, typically, recent historic data over a desirable time interval or desirable number of auctions is used to generate a bid-response function. Further filters could be applied such as seller class, time of day etc to further refine this data based on seller's requirements, and
  • 2. For tracking a sellers intermediate progress based on demand, the qualitative demand curve of FIG. 12 is configured by the seller so that the target can be broken into sub-periods. The objective here is not to have the actual quantitative charts with time, but rather a qualitative positioning with respect to time. As an example, it may be desirable to know if a product such as a camera is sold more on Friday, Saturday, and Sunday in relationship to other weekdays. On the other hand, a bank CD may be more likely to be transacted during weekdays than over the weekend. As discussed earlier, the qualitative demand curve can be obtained from seller input or from historical data.
    Such inputs for implementation may include:
  • 1. Sellers targets such as revenue, profit and sales volume targets for each product. The targets may be quarterly, monthly, weekly or daily, or any other interval selected by the seller. In the example of sales targets, revenue and profit are in monetary units, and sales volume is the number of product units the seller expects to sell.
  • 2. Bid-response parameters (for a reference market price for faster convergence). The bid-response parameters can be sent to the SAEJ by the RAC, or computed by SAEJ as described earlier.
  • 3. Qualitative demand curves for each product for tracking progress can be configured or extracted from historical buyer activities as described earlier, particularly where only relative demand is needed.
  • 4. Seller's utility function (or functions if multiple chosen) and associated utility parameters, including aggressiveness factors, as configured by the seller.
  • 5. Relative utility weights. Relative importance of revenue, profit and sales volume to the seller. The weights are real numbers between 0 and 1, and they should sum to one.
  • 6. Product unit cost.
  • 7. Appropriate price constraints. The lower limit of the price could be the cost in case of a camera; on the other hand in case of a bank CD, upper limit on interest is the constraining issue rather than the lower limit.

As described earlier, furthermore, the target directed implementation is well suited for the sales target optimizer where the seller is trying to achieve annual, quarterly, monthly etc, targets. The target intervals can also overlap with two optimizers operating over the desired time periods and with appropriate targets, respectively. The target-directed implementation can also be used for events price optimization. In this case, the seller may want to boost sales during a week/weekend or any arbitrary time interval. The seller configures the optimizer with the appropriate targets, duration, weights etc, and the optimizer computes an optimal price from the targets. The integrator then decides which price (or a new price resulting from integrating the several optimal prices) to use.

Another example of effective use of this implementation is for supply break optimizations. As described earlier, the seller can specify the supply break information, and these form the targets entered into the supplier break optimizer 11. When the number of items sold so far is sufficiently close to the supply break (as specified by the seller), the supply break optimizer is triggered and processes requests until the supply break is met or the supply break time expires (whichever comes first). The price integrator PI then uses the results of the individual optimizers, along with appropriate constraints on pricing, and other input from the seller, and then generates the bid price at 7 to be returned to the RAC.

As may be evident from the above discussion, the target-directed implementation has several key advantages that enables its use for a variety of optimization purposes, with the ability to adapt to a wide range of evolving market conditions and seller objectives. Some of these advantages are.

• • 1. It is target-directed and significantly enhances, in real-time and in an automated fashion, the offline and heretofore manual, cumbersome sales process of many organizations. This results in efficiency improvement of several orders of magnitude
  • 2. It allows the seller to set sales targets (using revenue, profit, sales volume and other similar metrics) and optimally, in real-time, manage product pricing to maximize the likelihood of reaching the targets in ever-changing market conditions.
  • 3. Multiple independent targets can also be set. For example, in case of CDs, a bank may set deposit goals of $10 million, 3 months duration at lending rate of 8%, spread of 3% to be raised in 5 days, and $20 million, 6 months duration at 8.5% lending rate, spread of 2.5% to be raised in 4 days, etc. The implementation may be used by independent and/or dependent and/or any combination to achieve all of the targets simultaneously.
  • 4. Can be used by multiple (similar or different types) optimizers such as sales target optimizers, supply break optimizers, event optimizers etc, each of which is tailored to seller objectives, thus enabling the seller to achieve a large variety of objectives.
  • 5. Both independent and dependent targets can be set.
  • 6. This approach maximizes a weighted combination of multiple targets enabling the seller to make trade-offs among different targets in real-time on a dynamic basis. This is unlike traditional systems where such processing consists of highly inefficient manual processes.
  • 7. Seller has numerous utility function options to choose from to meet its unique objectives.
  • 8. By assigning desired weights to the sales volume target, the seller can manage inventory efficiently.
  • 9. By assigning desired weights to the revenue target, the seller can manage revenue efficiently.
  • 10. By assigning desired weights to the profit target, the seller can manage profit efficiently.
  • 11. No prediction of future demand is required at all.
    Exemplary Market Share-Directed Implementation
    (FIG. 16)

The market share-directed implementation of the invention previously described computes price by considering the percentage of the auctions won (market share) in the previous pre-defined number of requests. As an example, consider that 10 auctions were won out of 100 requests received. This could be computed as a unique number per so many requests received for example in a slot of 100 auctions, or could be a rolling number such as number of auctions won in last 100 auctions. If the market share is below the desired target, the price will be appropriately corrected automatically and in real-time until the desired objective is matched. If the number of auctions won is higher than the desired market share, the machine will automatically make appropriate corrections using this feedback loop to cut the winning percentage down, if so desired by the seller (or vice versa). The key idea is to make the winning independent of the time of the day, or day of the week, or if it was a snow day, etc, and ensure that the winning percentage number effectively adjusts itself as per the market demand without requiring any quantitative forecast.

A typical implementation is given below, and illustrated in FIG. 16, though there are a number of variations that may be chosen to implement such, including rolling computation of the market share.

Flow Chart For FIG. 16

1. Initialization

• • Set initial price at 31 (could be either based on seller configuration or based on pre-specified criteria on historic market price).
  • Set market share target as configured by the seller at 30.
  • Set scaling factor as configured by seller at 32. The scaling factor determines how aggressively the seller wants to pursue the market share target.
    2. Repeat after every N requests
  • a. Track and compute market share for these N requests at 34.
  • b. Compute absolute “distance” between achieved and target market share at 35.
  • c. IF market share<target,
    • set new price=old price−scaling factor*distance, at 37.
  • Otherwise,
    • set new price=old price+scaling factor*distance, at 38.
  • END
  • d. Set the new price as the price for next N requests at T3.

As with the target-directed implementation, previously discussed, the result of this market-share-directed implementation may be used in a variety of ways.

This market-share-directed implementation makes no assumptions about the demand, buyer and competitor characteristics. The key input is the target market share. In addition, the product cost and the appropriate constraints on the unit price are also required inputs. One additional feature provided by the RAC is the ability to provide a constraint in terms of not to exceed a configured market share, thus enabling liquidity. The RAC market share will override the seller-configured market share in case the latter exceeds the former. As an example, the RAC could constrain any one entity to no more than 25% market share.

There are several advantages for this market-share implementation:

• • 1. It allows the seller to track and respond to the changing market conditions rapidly.
  • 2. The implementation is independent of time, and entirely a function of market demand within a pre-specified interval used as a measure (e.g. annualized market share). This can be further subdivided (eg. on a quarterly basis) with its own unique market-share target. This will provide the ability to the seller to have quarterly targets in addition to an annualized blended target. As an example, if it is a holiday and not much trade is being conducted, the demand on a bank CD is likely to be lower; and as those requests slowly arrive, the system will respond in a measured manner attempting to win a certain percentage of auctions regardless of the slowness of the request arrivals.
  • 3. It helps the seller maintain, increase or decrease market share automatically and in real-time, in a highly competitive environment.
  • 4. Ability to keep pace with the market automatically if demand slows down or speeds up.
  • 5. Independent of other competitor's promotions etc if so desired.
  • 6. Ability to diversify and always participate in the market, thus reducing the risk of tactical decisions adversely affecting the goal in a significant manner.
  • 7. Persistence of price promotion and price avoidance are schematically represented as time cycles in FIG. 17. For a retail vertical, generally a seller does not want to establish a product price any lower than necessary in a fiercely competitive environment. In circumstances such as suppliers break or aggressive inventory reduction where the seller may have to dip a bit lower to meet the targets; it may be highly desirable to act in a manner which minimizes the persistence of such reduced price. As an example, if a seller found a new viable price at which it can sell 100 cameras, it may try to win 100 consecutive auctions and deal with the resulting likelihood of a residual price persistence. That is, for one auction, the price is lowered and the auction is won. Then, subsequently, the price is raised for some further auctions. To avoid a low persistent market price, a duty cycle of periodic price raisings may be instituted as shown at “Price persistence avoidance during promotion”, FIG. 17. Thus, the same function can be executed with a duty cycle of, for example, 10%, implying winning 1 out of 10 auctions. This will result in a stretched number of auctions across which 100 cameras were sold, but, the price persistence is now minimized. This can be implemented in a number of different ways. One way is to use the Integrator to put on the brakes such that it does not try to win any auctions for ‘N’ number of slots after winning ‘M’, generating the desired duty cycle conceptually shown in FIG. 17. Both ‘N’ and ‘M’ are seller configurable.
    Exemplary Utility Derivative-Following Implementation
    (FIG. 18)

The utility derivative-following implementation, as the name suggests, is based on improving the seller's utility over time. In the utility derivative-following approach, in the preferred embodiment, the utility can be formulated as a function of distance to the target as previously described in the target-directed implementation. In alternative embodiments, any other utility function may be used as desired by the seller.

The utility derivative-following implementation, however, compares the utility gained (or lost) from change in price over two previous intervals. It then adjusts its price by looking at the amount of utility earned on the previous interval as a result of the previous interval price change. If the price change produced a higher utility, then the strategy makes a similar change in price for the next period. If the previous change produced lower utility, then the strategy makes an opposing price change for the next period. If the utility is target based, this approach will try to follow the market and pursue the targets simultaneously.

A typical implementation is shown below and illustrated in FIG. 18.

Flow Chart For FIG. 18

• • 1. Initialization
    • Set initial price at 31. This initial price could be configured by the seller based on his understanding of the market price, or it may also be obtained from historical data if available.
    • Set initial utility at 39. The initial utility is calculated based on the seller's configured desired utility function. In case of a target based utility function, the calculation is based on the distance to the target.
  • 2. Repeat after every N requests. (The initial price is fixed at 40 for N requests, where N is configured by the seller). After every N requests at T4, the utility gain/loss is computed and the price is adjusted based on this.
    • Track and compute utility gain at 41 for these N requests.
  • IF utility gain>utility gain of previous N requests and IF current price>previous price, THEN
    • add a positive random amount within constraints to the current price (This amount could be pre-specified by the seller or computed as a small fraction of the current price)
  • Otherwise,
    • subtract a positive random amount at 43 within constraints from the current price
  • END
  • IF utility gain<utility gain of previous N requests,
    THEN
  • IF current price>previous (old) price,
    • subtract a random positive amount at 43 within constraints from the current price
  • Otherwise,
    • add a random positive amount at 42 within constraints to the current price
      Set the newly computed price at T4 as the price for the next N requests.
  • The inputs for this implementation are:
  • Initial price as described in the implementation. The closer the initial price is to the market price, the quicker the implementation converges.
  • Revenue, profit and sales volume targets, if using a utility based on these targets. As noted before, any utility can be used.
  • Sellers utility function and associated parameters.
  • Utility weights.
  • Product unit cost
  • Appropriate constraints on unit price

Since the price changes in this implementation are based on utility, and the utility is related to the specified targets, the implementation can be also considered as target-directed, but with a few further unique advantages.

• • 1. It further reduces computation by not requiring historic bid-response; and
  • 2. It is responsive to the changing market conditions while working toward to the preset targets.
    Exemplary Model Optimizer With
    Exploration Implementation

This implementation has some similar properties to the target-directed implementation before described. It uses the concept of maximizing utility to compute an optimal price for initial use. This is called the “baseline optimal price”. It further improves this baseline optimal price by using random (within constraints) deviations from the baseline optimal price to periodically explore the price space. Once a more optimal price is found, the implementation uses that as the baseline optimal price until an even more optimal price is found. This process continues as the system continues to hunt for the optimal price in an ever changing marketplace. The deviations from the baseline optimal could in fact be related to the distance to the target, or any other number as desired by the seller. The ability to explore the price space by using deviations from the baseline optimal price allows the seller to lead the market where a more optimal price is found.

A typical implementation is described in the flowchart below and is illustrated in FIGS. 19 and 20.

Flow Chart For FIGS. 19 and 20

• • 1. Initialization
    • Set parameters for bid-response function.
    • Decide at 20 on the utility function to use, and set utility parameters.
    • Set relative weights for utilities at 50 in FIG. 19.
  • 2. Repeat after every N requests (52)
    • Set at 61, FIG. 20, historic bid-response function parameters if updated values are available from RAC or local computation
    • Compute optimal price based on maximizing utility and set this as the “baseline optimal price” for the next new auction requests.
    • As requests for new auctions are received, return the “baseline optimal price above”, unless this is the random request is chosen for the exploration experiment whose duration is M requests.
    • If current request is the random request chosen for exploration experiment
      • Compute the distance to the target at 62, FIG. 20, based on the utility function desired.
      • For this request, adjust the price up or down randomly at 63 (within constraints) by an amount proportional to the distance. Call this the “experimental optimal price” (EOP), for use for next M requests
    • After M requests, re-compute utility at 64.
    • If the average utility gain per request using the experimental price EOP after M requests is more than the average utility gain per request, using the baseline optimal price, use the experimental optimal price EOP as the new baseline optimal price at 65 and T7. Otherwise, keep the baseline optimal price unchanged.

Further process requests from the RAC are processed at 59 and the optimal price is returned at 85.

This implementation for FIGS. 19 and 20 makes assumptions similar to the before-described target-directed implementation, as follows:

• • Historic bid-response function parameters for faster convergence.
  • Revenue, profit and sales volume targets.
  • Seller's utility function and associated parameters.
  • Utility weights.
  • Product unit cost
  • Appropriate constraints on price

This implementation being utility based and the utility being related to the targets, it is also thus target-directed, permitting the seller to have the ability to lead the market by setting the product price by using random deviations up or down from the baseline price. The implementation can be further enhanced, moreover, by using machine learning techniques that provide intelligent ways of picking when to experiment; how much to raise/lower the price, and for how long.

Useful Applications of the Invention

While it is clear from the above discussions that the SAEJ of the invention can satisfy the various and ever-changing objectives of sellers in a wide range of verticals, the SAEJ architecture provides the additional ability to the seller of enabling quantitative optimization of prices based on more complex dependent and independent objectives, in turn involving one or more verticals. With prior art or existing non-SAEJ based technology relying on manual or off-line processing, it is not possible to optimize in real-time and iteratively, thus limiting the scope of applications to silos of a small set of individual products, or verticals using manual or batch processing.

It is now in order to describe some of the most useful applications of the SAEJ of the invention that not only reinforce its applicability for a wide range of verticals, but also describe how this architecture provides the ability to optimize across multiple verticals, thereby allowing the seller to achieve overall objectives globally across all the desired verticals.

As earlier mentioned, moreover, the banking community as an example of an industry that can particularly benefit from the adoption of the techniques of the invention.

All the previously described pricing implementations of the invention for the retail industry, indeed, can also directly be applied to the bank deposit and loan vertical.

Banking Certificates of Deposits

Banks rely on consumer deposits (in the form of CDs, High Yield Savings, Checking etc) as a source of funds for their loan commitments. They offer CD products with various maturities—as 3 months, 6 months, 9 months, 1 year, etc. In the case of CDs, they generate profit from the difference (spread) between interest payments on CDs (calculated as the annual percentage yield or APY) and the interest earned on loans.

One of the key challenges for banks is to collect sufficient CD deposits of appropriate terms and within appropriate timeframes so that they can meet their corresponding loan commitments. Banks must also determine the APY offered for each CD product based on various factors, such as short term and long term interest rates, loan commitments, market rates for comparable CDs, required spread to support overhead and other financial market conditions. Currently, the deposit needs are calculated manually and in an ad-hoc manner based on current loan commitments. In addition, banks have no means to use higher APYs (equivalent to lowering the price in retail) to attract requisite consumer deposits within a short time-frame, without significant advertising spending. These solutions require significant advanced planning and preparation, and as a result are unable to respond to needs in real-time as they arise.

The SAEJ architecture and implementations described herein, however, can easily be applied to bank CDs where SAEJ calculates, in real-time, the optimal APY for each CD product based on bank objectives, and market conditions. In a preferred embodiment, the before-described target-directed implementation may be used; but in alternate embodiments, the market-share implementation, the rules-based implementation, the utility derivative-following implementation, the model optimizer with exploration implementation, or a combination of these implementations may also be used.

As before stated the objectives of the bank in meeting loan commitments and achieving profitability etc, can be mapped into quantitative targets for the price optimizers techniques of the invention.

• • Loan commitments can be used to specify deposit targets. For example, if the bank has a loan commitment for $10 m for 1 year, it must collect the loan amount in the form of deposits (or borrow from other institutions such as FHLB, which is a less desirable option due to strict regulations, expensive rates and collateral requirements) so that sufficient funds are available in time to fund the loan for its term. These loan commitments can be used to specify deposit “targets” for the corresponding CD products (of appropriate term), and the deposit targets must be fulfilled within a pre-specified time.
  • Profit targets specified by the bank are a function of the blended average spread between lending rate (or a specified number of points above a benchmark rate) and interest paid on deposits. Profit “P” can be defined as a function of A(r_L−r_A), where A is the amount deposited, r_A is the interest rate that the bank pays to the consumer, and r_L is the interest rate the bank earns by lending or investing the same amount of money. This profit target can be configured by the seller over a pre-specified time interval.
    Thus, for each CD product, the bank configures deposit targets. For each deposit target the following factors are involved:
  • a. Time to fulfill the deposit target. This may be defined as the time when the target deposit is needed by the bank.
  • b. The duration of deposit is specified with the CD product selection
  • c. Profit target.
  • d. Appropriate constraints on APY
  • e. Bank lending interest rate (or a reference rate).
  • f. Bank utility function and relative weights.

For each RFB for a new auction, the following steps are involved:

• • 1. The price optimizer (PO) computes the optimal APY, referred to as the optimal computed APY. The optimal computed APY can be used as the temporary ceiling for the remaining auction rounds.
  • 2. The price integrator (PI) starts with a floor APY (which may be configured by the seller as part of constraints on APY, or be chosen to be a low number within such constraints), and then increments the APY subject to other bids, iteratively in successive round bids, until it wins the auction or reaches the optimal computed APY, whichever comes first. The increment size may be constant (for example, subject to highest bid in the last round), or proportional to the weighted “distance” to targets. The further away the bid from the target, the incremental size will be larger; closer to the target, the size will be smaller. The size can also be inversely proportional to the remaining fulfillment time. In addition, the number of iterations for the auction may be imposed from the controller. The distance to the target may be linear, exponential, piecewise linear, piecewise exponential or hybrid as described herein.

It should be noted that multiple such targets can be configured. A bank, for example, could have a need to raise $10M in deposits in the next 24 hours, under terms of a one year duration, a lending rate of 8.5% with a minimum spread of 3%. Other such targets (also called “buckets”) might be a desire to raise $3M in the next 3 days, with a 6 month duration, for a lending rate of 8% and with a minimum spread of 2.75%. Both these and other such requests can be set-up simultaneously. Each of these targets furthermore, can also be prioritized or a relative weight assigned thereto, if so desired, enabling each new request to be responded to with corresponding aggressiveness subject to the current status of the target. This is in sharp contrast to the prevailing manual practice at banks to speculate and then set the interest rate (and invariably overshoot or undershoot), advertise for much longer duration, and wait for the response. The resulting uncertainty and advertising expenditure in such a process is undesirable. In the mean-time, a mutual commitment cannot be made to the customer, resulting in very little stickiness of associated business.

The Bank Credit Card Problem

Banks issuing credit cards prefer subscribers with specific types of credit scores. Some banks aim for those customers with stellar or fairly decent credit scores. Others prefer sub-prime credits scores; and still others opt for everyone in between. Each strategy has its advantages and disadvantages.

While a portfolio of best credit score subscribers is very safe, the return tends to be low as most such subscribers pay their bills on time, leaving little opportunity for the bank to earn interest and fees. On the other hand, sub-prime subscribers are high risk due to high default rates but they may provide a much higher rate of return. Ideally, banks like to have a portfolio which minimizes the credit risk yet provides a decent rate of return over economic cycles of different hues. Unfortunately, however, their manual process forces them typically to evaluate the portfolio at the end of each time interval (as an example, a quarter), and then estimate the blended credit score across the entire portfolio and the blended interest rate. Not only is this information too late and after the fact, their inability to correct the situation in the short run or to the requisite precision compounds the problem.

Such a problem admirably lends itself for solution, to the use of multiple engines of the invention, one for each desired category of credit score, and an optimizer dedicated to perform front end arbitration and optimization in order to select a specific engine to participate in the subsequent auction.

Consider, as an example, a bank wanting to offer credit cards to subscribers with 4 different categories of credit scores under circumstances of an illustrative total line of credit targets as shown below, with the intent to maintain a blended credit score across all the targets of between 600 and 700 at any point in time, as shown in FIG. 21. In this illustration, there is shown:

• • For credit score>800, the total target credit line amount is $50M
  • For 700<Credit score≦800, the total target credit line amount is $100M
  • For 600<Credit score≦700, the total target credit line amount is $200M
  • For 500<Credit score≦600, the total target credit line amount is $100M ‘For 400<Credit score≦500, the total target credit line amount is $50M

In this case, the PMU 4 consists of a front-end optimizer PO at 80, as well as several SAEJ credit-optimizers (SAEJ1 through SAEJ5), one for each target specified by the seller.

As each request for a new auction arrives, the front-end optimizer 80 examines its credit score in real-time to ensure that it is within the acceptable range (above 400 in this case). Assuming the request is within the range, the optimizer uses the following computation to determine the incremental blended average, if this auction is to be won.

The blended credit score is calculated as

$\begin{array}{cc}C=\frac{\sum _{i=1}^N{F}_i{C}_i}{\sum _{i=1}^N{F}_i},& \left(21\right)\end{array}$
where, in the above formula, C is the blended credit score and Ci is the average credit score for target i. Fi is the amount of credit issued so far for target i, and N is the total number of targets.

If the computed blended credit score is expected to remain within the constraints outlined by the bank, the front-end optimizer signals the corresponding credit-optimizer to participate in the bid process. Alternatively, if the blended average constraints are going to be violated, the front-end optimizer will not signal any credit-optimizer to participate in the bid process. Thus, this optimization across various targets, in real-time and on a dynamic basis, guarantees the bank all the time a consistent blended credit score (CS) quality of its entire credit card portfolio at any time. The bank, furthermore, also has provided for the blended interest rate per line of credit target and across the portfolio—all available at the same time. This is illustrated in FIG. 22 where again as in FIG. 21, multiple SAEJ optimizers (1 through 5) are employed to achieve the blended credit score (CS).

Beyond the blended credit score optimization across the portfolio outlined above, in an alternate embodiment, each SAEJ engine may also optimize its blended interest rate to the bank-specified level, using any of the techniques discussed herein. In this scenario, the bank will have the guarantee of a portfolio optimized for not only specific blended credit scores; but also further optimized to a desired interest rate level for each total credit-line target.

This same concept can also be applied to auto loans and other verticals in financial and similar situations.

Optimized Bank Deposits and Dependent Optimized Loans

Another feature in using the techniques of the present invention, again using the bank vertical as an example, and which was not previously feasible, is herein also enabled. Consider a case in which a bank has set-up and configured multiple optimizers to collect deposits, such that a first is dedicated for 3-month deposits as at 1¹ in FIG. 23, a second at 2¹ for 6 months, a third at 3¹ for 1 year, and a fourth at 4¹ for 2 years and above. Each type of deposit has its own hard constraint on the interest rate to be paid for such deposits such as shown at 1¹¹, 2¹¹ and 3¹¹, respectively. In addition, each one may have a weighted blended interest rate target computed such that, for example, an illustrative $10,000 at 5.25% carries twice as much weight as $5,000 at 5%, when computing the average.

As another important application of the invention, consider that a bank has also adopted and set up the multiple optimizers of the invention for auto loans, categorized, for example, according to the applicants credit scores that it is to process, each, however, subject to its own unique constraints and targets, such as blended interest rate, and spread with respect to some standard benchmark such as Treasury or LIBOR or others. A blended credit-score target can be achieved across these auto loan optimizer-engines 1¹¹, 2¹¹ and 3¹¹, and the same for blended interest rates.

In this situation, a price integrator PI of the invention may readily be used, as in FIG. 23. The integrator connects the above-described deposit engines 1¹ through 4¹ to the front-end optimizer, so-labeled. The over all system may have the following illustrative constraints, with the loans being made only from the deposits collected.

The aggregate percentage contribution from each deposit engine towards loans, however, may be different—auto loans, for example, tend to be for longer terms. The bank could, as an example, specify that deposits are to be disbursed such that 94% of the deposits collected by the 2-year plus engine 4¹, are to be used towards such auto loans; 50% for the 1-year deposits 3¹; 10%, for the 6-months 2¹; and 5% for 3 months 1¹, and so on.

Deposits, furthermore, may be collected by bidding in real-time for prospective depositors according to configured constraints. These deposits may then be funneled, appropriately proportioned, via the integrator to the front-end optimizer of the loan engines. As a request for an auto loan arrives, the front-end optimizer examines its credit score in real-time to ensure that it is within the acceptable range. Assuming the request is within such range, the optimizer uses the following equation to determine the incremental blended credit score average across the portfolio, if this auction is won. The blended credit score is calculated as

$\begin{array}{cc}C=\frac{\sum _{i=1}^N{F}_i{C}_i}{\sum _{i=1}^N{F}_i},& \left(22\right)\end{array}$
where C is the blended credit score and Ci is the blended credit score for engine i; Fi is the amount of loan issued so far for engine i; and N is the total number of loan engines.

If the computed blended credit score is expected to remain within the constraints outlined by the bank, the front-end optimizer signals the matching loan engine to participate in the auction, assuming funds are available. Alternatively, if the blended credit score constraint is going to be violated, this optimizer will not signal any loan engine to participate in the bid process. The selected loan engine then participates in the auction with its unique constraints and targets as configured by the bank. Thus, not only are each engine targets optimized, but blended targets across various engines are also optimized and in real-time and on a dynamic basis. This guarantees the bank a consistent blended credit score quality across its entire auto loan portfolio at any time, all the time. Furthermore, the resulting blended interest rate across such loan engines can also be computed, providing additional insight into the loan quality. Additionally, the front end optimizer also optimizes for the blended portfolio interest rate by computing in real-time the allowable lower limit interest rate constraint as each loan auction request arrives. This dynamically computed allowable constraint is then automatically provided to the selected engine.

As is evident from FIG. 23, moreover, this implementation provides a powerful mechanism for banks to collect deposits in real-time and subsequently to make auto loans in real-time while maintaining the appropriate constraints on both deposits and on loans, such that not only is the deposit base substantially expanded, but the corresponding loan business is appropriately matched—all the while maintaining an optimized loan quality both at in individual engine level and also across the entire loan portfolio. This architectural configuration using implementations described earlier can provide banks with an unprecedented opportunity to grow in a very short time. Clearly, moreover, many variations of this concept, adjusted for vertical specific parameters, can be used in various verticals.

While the important application of the technique of the invention to banking services has been stressed, application to other services and goods bring similar advantages that will now be summarily tabulated.

Summary of the Unique Advantages of the SAEJ Solution of the Invention

• • Provides the ability to sellers simultaneously to process in real-time, a number of multiple interacting variables such as revenue, profit, market share, inventory, supplier-break, competitor events, market demand changes and others, and automatically to optimize, in a dynamically changing environment, and make iterative bids within specified constraints in real-time on demand—all without the need for manually tracking the variables.
  • Once configured, requires no manual intervention from the seller (unless the seller so desires), even in the presence of dynamically changing market conditions.
  • Supports multiple instances of multiple types of price optimizers that can be configured based on the seller's objectives as well as market conditions, demand, competitive behavior, inventory, supplier-break etc. The price optimizers compute when to update the price and by how much, based on real-time data from the marketplace and seller configured parameters, and without requiring any further manual intervention from the seller.
  • Ability to set sales targets (using revenue, profit, sales volume and other similar metrics) for pre-determined time frames such as daily, weekly, monthly, quarterly, annually and so on. Furthermore, optimized product pricing is derived in real-time to reach the targets. The need to speculate the price that the market will bear is also eliminated.
  • If a target is missed, the ability is provided to allocate the missed target in several ways, as desired, such as:
    • move the difference to the immediate next time slot;
    • distribute it equally over multiple times slots;
    • distribute it unequally across multiple time slots to match the customer buying pattern, and so on
  • Ability to adapt to dynamic interplay among sales targets and provide requisite respective emphasis to each sales target, and correspondingly compute an optimum price.
  • Ability to select, among nearly infinite choices of rates of approaches to the targets, an approach unique to the seller's objectives, such as:
    • make all out effort early in the quarter to minimize any risk of missing the quarter;
    • pace slowly initially and pick up momentum as it get closer to the quarter;
    • sell at a steady pace, participating in the market price variations throughout the specific duration resulting in improved diversification;
    • Any combination of these choices can also be selected as a piecewise function.
  • If the sales target is met earlier in the desired interval than anticipated, ability to choose from multiple approaches such as slow down to keep the revenue smooth and predictable; be more aggressive for the rest of the interval; accelerate to substantially enhance the profit as the overheads largely remain the same; and any combination of the above in the form of a piecewise function.
  • Ability to specify market-share target, and, dynamically, in real-time, maintain, increase or decrease market share in a highly competitive environment, without any manual or off-line processing. This makes the process independent of demand as a function of time, and provides the ability to manage growth and resulting market share in numerous ways, including, grow at a pace faster than market, to gain minimum sufficient volume critical to their survivability, or to be a major player, or due to near-term equity valuation considerations, or combinations thereof; or to grow at market pace or higher so as not to allow the competitors an edge; grow only at the market pace, possibly due to various constraints including lack of growth capital, supply of raw material, lack of trained manpower, or the fear of inviting an anti-trust ruling
  • Ability to discover and implement in real-time an optimizer strategy subject to the current distance to the respective targets, and/or unique constraints and time left in the specified duration. This can lead to a sacrifice of profit in favor of revenue, or vice-versa; or the sacrifice of both profit and revenue in order to reduce the inventory rapidly at the tail end of the useful life, especially those of seasonal products; or to a sacrifice of some targets for the desired market share
  • Ability to achieve quickly a volume-based discount from its suppliers/manufacturers automatically and in real-time without requiring any manual intervention. When a seller is close to meeting a supplier-break target the ability is provided to automatically accelerate the product sales by running a very short term promotion in order to reach the target and acquire the much larger retroactive or proactive profit, or both.
  • Ability to run promotions independently of, and/or in anticipation of, competitive promotions in real-time and with the:
    • Ability to automatically track competitor promotions based on current market price, for hundreds or thousands of sellers in an on-line fashion;
    • Ability instantaneously to respond to competitor promotions, if so desired, without any manual intervention;
    • Ability to select which promotions for response based on seller class, geographical location etc.; and
    • Ability to do all of the above without any manual intervention when a competitor promotion does occur.
  • Ability to maintain and track those consumer and reseller desires to purchase products at certain price points determined by them, and offer such prices to this captive market that is willing to purchase at these price points without being forced to advertise such prices to entities outside of the target customers, such as competitors or other customers, and without requiring the customers to register on a seller specific system, such as the seller's web site. This seller offering is made on a real-time basis and is an exceptional tool to enable sellers to generate very short promotions, without significantly alerting the market space. (Traditional mechanisms are ineffectual under these scenarios).
  • The price optimizer together with the Integrator of the invention allows the seller to achieve all of the above goals automatically and simultaneously. This is accomplished by adjusting the priority/weightage of each goal based on prevailing market conditions and the time left to meet the time-sensitive targets as compared to the overall duration and the current distance to the other targets. For certain market conditions or time periods in the sales cycle such as towards beginning of the quarter for example, profit may be more important than revenue; while near middle of the sales quarter, meeting a revenue target may be much more important than profit, and near the end, inventory reduction may have higher priority than either profit or revenue, as per the seller's objectives.
  • Optimizers along with the Integrator herein allow the seller to achieve all of the above goals simultaneously and in real-time without manual intervention; each time the price has to be adjusted to achieve the goals in the presence of dynamically changing market conditions.
  • The optimizer accomplishes the above goals simultaneously in real-time, moreover, without waiting for offline tools to gather market data and adjust pricing over a period such as a week or month or more.
  • The optimizers achieve all of the above goals simultaneously, furthermore, without being required to predict future market conditions. Only qualitative input is required for the demand curve, and if this is not available, the market-share optimizer is used.
  • By constant interaction with the auction/reverse-auction controller, and by generating real-time responses, the system achieves all of the above goals simultaneously, through dynamically updating the offered pricing in real-time to the benefit of the seller, based on its own unique objectives.
  • For sellers such as banks, the ability is provided to specify deposit requirements in the context of lending rates, spread, duration, and blended average deposit rates, and the optimizers of the invention will optimize the associated set of variables in real-time.
  • Sellers such as credit card issuers, moreover, have the ability to specify engines of desired credit score ranges and corresponding credit limits and interest rate ranges, as well as a blended credit score and blended interest rate, and issue credit cards while continuously maintaining all these constraints in real time without any manual or off-line processing.
  • The sellers of such, additionally have the ability to carve out deposit requirements with different constraints, as multiple engines optimize for each one in real-time, and they also simultaneously optimize across them in real-time if so desired. This same concept also applies to other exemplary areas such as credit card engines, each supporting a different range of credit scores etc.
  • Ability to specify a set of engines to collect deposits and then enable the usage of the proceeds, in part or in whole, by a set of dependent engines focused on credit card verticals or auto loan verticals or both, each such set and their respective subsets being allowed to have unique constraints and targets, such as a desirable range of credit score, blended credit score of the portfolio, and blended interest rate of the portfolio etc.

While the automatic price optimizers, price integrators and SAEJ engines containing the same, as well as the underlying mathematical basis, have heretofore been described for use in automatic reverse auction applications, they are of broader utility wherever automatic price optimization calculations may be desired. Such other applications, indeed, may involve manual set-up as, for example, particularly in banking, financial, insurance and related applications.

Further modifications will occur to those skilled in this art and such are considered to fall within the spirit and scope of the invention as defined in the appended claims.

Claims

1. An automatic seller and price quotation system having, in combination, a seller automatic engine storing seller-specific seller, specific predetermined market strategy, business objectives, and market condition information and responsive to buyer requests in order to automatically to create a response to the price quotation based upon the buyer request and within the respective guidelines of seller-specific predetermined information stored in that seller automatic seller engine;

a processor for at least one of initiating an iterative real-time automatic seller engine competition for bettering price quotations within said respective guidelines until a best price quotation is received, transmitting a request for quote to the seller engine, and transmitting a request for price watch to the seller engine;

the seller automatic engine being adapted to process multiple price optimizers through one or more architectural implementations within the seller automatic engine, at least two of the price optimizers configured for different goals, the architectural implementations selected from the group consisting of a parallel processing architecture, a pipeline architecture, a hub and spoke architecture, and a hybrid architecture.

2. The system of claim 1 wherein each price optimizer is configured for an architecture selected from the group consisting of market-share directed implementation, specific sales target-directed implementation, seller utility derivative-following implementation, selected model optimizer implementation with exploration features, mathematical optimization oriented implementation, and a rule-based implementation.

3. A system as claimed in claim 1 wherein the multiple price optimizers are provided to the seller engine through a price management unit for enabling price optimization for one or more of several goals or business objectives of the seller including an objective selected from the group consisting of global price, events, and supplier's breakpoint.

4. The system of claim 3 wherein the price optimization is effected in light of inputted market data, historical data and seller targets.

5. The system of claim 3 wherein all optimized prices within the seller engine are integrated to produce a seller price quote.

6. The system of claim 5 wherein price management is implemented within the seller engine by mathematical logic optimization that addresses each at least one of price optimization and optimized price integration.

7. The system of claim 6 wherein price optimization within the seller engine is effected for one or more of several goals or business objectives of the seller, including global price, events, and suppliers breakpoint.

8. The system of claim 7 wherein price optimizing within the seller engine is effected by an implementation selected from the group consisting of target-directed implementation, market-share implementation, model optimization with exploration implementation, utility derivative-following implementation, and rules-engine implementation.

9. The system of claim 8 wherein a utility derivative-following price bid implementation function is used, and the price is adjusted by the amount of utility earned in a previous period as a result of a price change.

10. The system of claim 8 wherein, when a price request is received, an expected contribution to the seller utility from that request is calculated such that the optimal price will be the price that maximizes such contribution.

11. The system of claim 10 wherein the utility function is selected from the group consisting of substantially of exponential, piecewise exponential, linear, piecewise linear, and hybrid, and wherein each price optimizer is configurable to be one of a sales target optimizer, a market share optimizer, an event optimizer, a supply-break optimizer, a price watch optimizer, or an other optimizer, and each price optimizer is configurable for implementation as a market share-directed implementation, a sales target-directed implementation, a utility derivative-following implementation, or a rules engine implementation.

12. The system of claim 11 wherein an estimation of historic bid-response function parameters is enabled by minimizing squared errors through equations

13. The system of claim 3, wherein optimization for one or more of several goals or business objectives of the seller includes an objective selected from the group consisting of global price, events, and supplier's breakpoint.

14. The system of claim 7, wherein the several goals or business objectives of the seller include at least one of global price, events, and suppliers breakpoint.