A cross-platform advertisement expenditure joint budget allocation and ROI optimization method and system

Abstract

This invention relates to the fields of digital marketing technology and artificial intelligence optimization technology, and particularly to a method and system for joint budget allocation and ROI optimization of cross-platform advertising expenditures. It addresses the channel contribution assessment bias caused by traditional "last click" attribution through the Shapley value attribution model; it achieves quantitative prediction of inter-channel synergy and crowding-out effects by constructing a joint prediction model using nonlinear saturated response functions and cross-influence coefficients; it overcomes budget inefficiencies caused by independent optimization on a single platform through global optimization solutions using nonlinear programming and sequential quadratic programming algorithms; and it ensures the safe implementation of automated decision-making through safety executors and anomaly circuit breakers. This method elevates advertising budget allocation from a "local, static, and biased" level to a "global, dynamic, and scientific" level, significantly improving the overall advertising ROI by an estimated 15%-30%.

Description

Technical Field

[0001] This invention relates to the fields of digital marketing technology and artificial intelligence optimization technology, and in particular to a method and system for joint budget allocation and ROI optimization of cross-platform advertising expenditures. Background Technology

[0002] In the cross-border e-commerce sector, sellers typically run ads on multiple platforms simultaneously, including Amazon, Google, Facebook, and TikTok, to acquire traffic and orders. Current mainstream ad management software and platform-specific tools have the following limitations:

[0003] Existing tools (such as Amazon's Ads Console and SellerMotor's single-platform optimization) typically allocate budgets and optimize bids only for a single platform or campaign. This "siloed" management ignores the synergistic effects (e.g., brand search ads improving Amazon organic search) and competitive relationships (e.g., budget competition between platforms) between different platforms, failing to achieve optimal allocation of global funds. The user's purchase journey is usually cross-platform and multi-point-of-reach. Most existing systems rely heavily on "last-click" attribution data provided by each platform, severely underestimating the contribution value of other channels (e.g., social media brand advertising), leading to systemic biases in budget decisions based on biased data. Most optimization tools focus on reducing cost-per-click or ACOS of a single campaign, lacking a holistic approach from the perspective of maximizing overall company profits. Furthermore, optimization strategies are often static rules, unable to adapt to dynamic changes in the market bidding environment and product lifecycle. There is often a delay between budget allocation decisions and actual adjustments by the advertising platform, and a lack of closed-loop feedback learning mechanisms prevents rapid strategy adjustments based on real-time performance.

[0004] Therefore, there is an urgent need in this field for a technical solution that can connect cross-platform data, accurately attribute causes, and dynamically and intelligently allocate and optimize budgets based on overall profits. Summary of the Invention

[0005] Therefore, there is a need to provide a method and system for joint budget allocation and ROI optimization of cross-platform advertising expenditures to solve the technical problems involved in the background technology mentioned above.

[0006] To achieve the above objectives, the inventors provide a method for joint budget allocation and ROI optimization of cross-platform advertising expenditures, comprising the following steps:

[0007] S1: Collect advertising campaign data and corresponding sales order data from multiple advertising platforms, reconstruct cross-platform anonymous user behavior sequences and use a hybrid attribution model based on Shapley values ​​to calculate the contribution profit of each advertising touchpoint and obtain calibrated contribution profit data.

[0008] S2: Based on the aforementioned profit contribution data, a nonlinear saturated response function is fitted to each advertising campaign, and the cross-influence coefficient between different advertising channels is calculated through multivariate time series analysis to construct a joint prediction model for predicting total profit contribution under any budget allocation scheme.

[0009] S3: With the goal of maximizing the predicted total profit contribution of all advertising campaigns, the budget allocation problem is formalized into a constrained nonlinear programming problem, and iteratively solved using a sequential quadratic programming algorithm to output the optimal budget allocation scheme.

[0010] S4: The optimal budget allocation scheme is decomposed into step-by-step adjustment instructions and time-based strategy adjustment instructions through a safety executor, and then sent to each advertising platform for execution. At the same time, the core performance indicators are monitored in real time. When the core performance indicators are detected to deteriorate beyond the preset threshold, the abnormal circuit breaker mechanism is triggered.

[0011] Unlike existing technologies, the above-mentioned technical solution has the following advantages: It solves the channel contribution assessment bias problem caused by traditional "last click" attribution through the Shapley value attribution model; it achieves quantitative prediction of inter-channel synergy and crowding-out effects by constructing a joint prediction model using nonlinear saturated response functions and cross-influence coefficients; it overcomes budget friction caused by independent optimization on a single platform by using nonlinear programming and sequential quadratic programming algorithms for global optimization; and it ensures the safe implementation of automated decision-making through safety executors and abnormal circuit breaker mechanisms. This method elevates advertising budget allocation from a "local, static, and biased" level to a "global, dynamic, and scientific" level, significantly improving the overall advertising return on investment, which is expected to increase by 15%-30%.

[0012] Furthermore, in step S1, the specific method for calculating the profit contribution of each ad touchpoint using the hybrid attribution model based on Shapley values ​​is as follows:

[0013] Treating each ad touchpoint as a player in a cooperative game, and order profit as the cooperative gain, the touchpoint is calculated using the following formula. Shapley value weights :

[0014] ;

[0015] in, For the total set of contact points, For the set of contact points The resulting conversion probability; for a total profit of The order, the touchpoints along its path The contribution profit is .

[0016] As described above, the specific calculation method for contribution profit in the hybrid attribution model based on Shapley values ​​is defined. By treating advertising touchpoints as players in a cooperative game, and using the Shapley value formula to calculate the marginal contribution of each touchpoint in all possible user journey combinations, the true value of supporting channels such as brand advertising and social media advertising can be objectively and fairly quantified, avoiding budget misallocation due to underestimating their contribution. This technique ensures that the attribution results are mathematically fair and unique, providing a reliable data foundation for subsequent optimization.

[0017] Furthermore, in step S2, the nonlinear saturated response function is the Michaelis-Menten function:

[0018] ;

[0019] in For the budget, To maximize output potential, The cross-influence coefficient is a half-saturation constant; the cross-influence coefficient is calculated using a vector autoregression model, forming a cross-influence coefficient matrix. , of which elements Indicates channel Budget for channels The impact on output; the joint forecasting model for a given budget allocation vector The method for calculating the predicted total contribution profit is as follows:

[0020] .

[0021] As described above, the specific implementation methods of the nonlinear saturated response function, cross-influence coefficient, and total contribution profit prediction formula in the joint prediction model are defined. The Michaelis-Menten function can accurately characterize the diminishing marginal returns of advertising expenditure; the vector autoregression model can dynamically capture the synergistic promotion or crowding-out competition effects between channels. Through this technical solution, the system can answer complex prediction questions such as "how will the output of channel B change if the budget of channel A is increased," providing a high-precision decision-making basis for global optimization and significantly improving the scientific nature of budget allocation.

[0022] Further, in step S3, the objective function of the constrained nonlinear programming problem is:

[0023] ;

[0024] The constraints include total budget constraints, upper and lower limits of channel budgets, return on investment constraints, and key performance indicator constraints. During the solution process, the sequential quadratic programming algorithm approximates the original nonlinear problem into a quadratic programming subproblem in each iteration, solves the search direction and step size, and updates the budget allocation scheme until the change in the scheme is less than the preset threshold or the maximum number of iterations is reached.

[0025] As described above, the objective function, constraints, and specific implementation methods of the sequential quadratic programming algorithm in the global optimization solution are defined. Using the sum of the total profit contribution from all channels as the objective function ensures global optima rather than local optima from the top-level design. The sequential quadratic programming algorithm can efficiently solve complex optimization problems with nonlinear constraints, meeting the timeliness requirements of minute-level decision-making while ensuring solution accuracy. This technology enables the system to quickly output the optimal budget allocation scheme under dozens of advertising campaigns and multiple constraints.

[0026] Furthermore, step S3 also includes a competition-aware dynamic adjustment sub-step:

[0027] The system monitors the average bid fluctuations of core keywords in the market in real time. When it detects that the competition intensity of a certain channel has surged week-on-week and exceeded the preset threshold, it automatically and temporarily relaxes the return on investment constraint of that channel and then performs global optimization.

[0028] As described above, a competition-aware dynamic adjustment sub-step is introduced. By monitoring the average bid fluctuations of core keywords in the market in real time, the system automatically relaxes the ROI constraints of corresponding channels when a surge in competition is detected. This allows the budget allocation strategy to proactively adapt to changes in the market environment, rather than reacting passively. This technology achieves dynamic self-adaptation of advertising, avoiding missed traffic opportunities or ineffective budget consumption due to intensified market competition, and improving the system's robustness in complex market environments.

[0029] Furthermore, in step S4, the specific method of the step-by-step gradual adjustment is as follows:

[0030] For budget increases, the target budget value is broken down into multiple incremental adjustment stages. An observation period is set after each stage adjustment to monitor the consumption rate and cost per conversion. The next adjustment is executed only after the consumption rate and cost per conversion are normal, until the target budget value is reached.

[0031] As described above, a phased and gradual adjustment method is defined. For budget increases, the adjustment is broken down into multiple progressively increasing stages, with an observation period set after each stage to monitor key metrics. This technique effectively avoids the risk of uncontrolled spending rates or soaring conversion costs caused by a one-time, large budget increase, achieving a smooth transition in budget adjustments and ensuring the stability of the advertising account.

[0032] Furthermore, in step S4, the specific method for adjusting the time-sharing strategy is as follows:

[0033] For bid increases, the strategy should be broken down according to the time period of ad traffic. First, test the increase during the traffic trough period. After verifying that the click-through rate is normal, increase it to the median value during the traffic peak period, and finally adjust it to the target bid value during the subsequent peak period.

[0034] As described above, the specific methods for adjusting the time-based strategy are defined. For bid increases, the strategy is broken down by ad traffic time periods. A test increase is first implemented during off-peak periods, and once verified as effective, a gradual increase is then implemented during peak periods. This technique utilizes the traffic characteristics of different time periods for risk testing, enabling safe bid increases while ensuring campaign effectiveness. It avoids the high cost and low conversion rate issues caused by large bid increases, thus improving the reliability of automated execution.

[0035] Furthermore, in step S4, the specific implementation of the abnormal circuit breaker mechanism is as follows:

[0036] After each instruction is executed, the cost per conversion and impression share are monitored in real time. When the cost per conversion is detected to spike above the threshold percentage within a preset time, the subsequent automatic adjustments of the current advertising campaign are immediately suspended, and the campaign is rolled back to the previous safety settings. At the same time, an alert is sent to the operations staff.

[0037] As described above, the specific implementation method of the abnormal circuit breaker mechanism is defined. After each instruction is executed, the cost per conversion and impression share are monitored in real time. When the metrics deteriorate sharply and exceed the threshold within a short period, automatic adjustment is immediately paused and the system rolls back to the previous safety settings. This technology effectively constructs a safety closed loop of "automatic execution - real-time monitoring - abnormal circuit breaker - automatic rollback," effectively preventing a precipitous deterioration in advertising performance due to model prediction bias or sudden market changes, and ensuring the safe operation of the advertising account.

[0038] Furthermore, it also includes step S5: collecting advertising performance data for the new cycle, evaluating the optimization effect, using the feedback data to update the nonlinear saturated response function and the cross-influence coefficient, and entering the next round of optimization cycle.

[0039] As described above, the continuous learning and iteration steps are defined. By collecting advertising performance data for each new cycle, the feedback data is used to update the nonlinear saturated response function and cross-influence coefficients, forming a complete closed loop of "data-decision-execution-evaluation-update". This technology enables the system to continuously self-optimize as advertising data accumulates, continuously improving model prediction accuracy and optimization performance, and achieving autonomous and scalable advertising operations.

[0040] To achieve the above objectives, the inventors also provide a joint budget allocation and ROI optimization system for cross-platform advertising expenditures, used to implement the aforementioned joint budget allocation and ROI optimization method for cross-platform advertising expenditures, including:

[0041] The cross-platform advertising data attribution module is used to collect advertising campaign data and corresponding sales order data from multiple advertising platforms. By reconstructing the cross-platform user anonymous behavior sequence and using a hybrid attribution model based on Shapley values, the module calculates the contribution profit of each advertising touchpoint and obtains calibrated contribution profit data.

[0042] The joint prediction and simulation module, connected to the cross-platform advertising data attribution module, is used to fit a nonlinear saturated response function for each advertising campaign based on the contribution profit data, and to calculate the cross-influence coefficient between different advertising channels through multivariate time series analysis, thereby constructing a joint prediction model for predicting the total contribution profit under any budget allocation scheme.

[0043] The dynamic budget optimization engine, connected to the joint prediction and simulation module, is used to formalize the budget allocation problem into a constrained nonlinear programming problem with the goal of maximizing the predicted total profit contribution of all advertising campaigns. It then iteratively solves the problem using a sequential quadratic programming algorithm to output the optimal budget allocation scheme.

[0044] The automatic execution and strategy management module connects to the dynamic budget optimization engine and the API of the external advertising platform. It is used to decompose the optimal budget allocation scheme into step-by-step adjustment instructions and time-based strategy adjustment instructions through the safe executor, and send them to each advertising platform for execution. At the same time, it monitors the core performance indicators in real time. When the core performance indicators are detected to deteriorate beyond the preset threshold, the abnormal circuit breaker mechanism is triggered.

[0045] Unlike existing technologies, the above technical solution has the following advantages: it protects the system used to implement the method, including a cross-platform advertising data attribution module, a joint prediction and simulation module, a dynamic budget optimization engine, and an automatic execution and strategy management module. This system solidifies the above method into a modular architecture, forming a complete data flow and decision-making chain between the modules, which can be directly deployed in cross-border e-commerce advertising operation scenarios. This technical effect enables the method to be implemented in a system form, achieving intelligent management of the entire chain from data collection, attribution, prediction, optimization to automated execution, and has clear technical integration and industrial application value. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the steps of a cross-platform advertising expenditure joint budget allocation and ROI optimization method in this embodiment;

[0047] Figure 2This is a structural block diagram of a cross-platform advertising expenditure joint budget allocation and ROI optimization system according to this embodiment;

[0048] Explanation of reference numerals in the attached figures:

[0049] 1. Cross-platform advertising data attribution module; 2. Joint prediction and simulation module; 3. Dynamic budget optimization engine; 4. Automated execution and strategy management module. Detailed Implementation

[0050] To illustrate the possible application scenarios, technical principles, implementable specific solutions, and achievable objectives and effects of this application in detail, the following description, in conjunction with the listed specific embodiments and accompanying drawings, provides a detailed explanation. The embodiments described herein are merely illustrative of the technical solutions of this application and are therefore intended to limit the scope of protection of this application.

[0051] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0052] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0053] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0054] In this application, terms such as “first” and “second” are used only to distinguish one entity or operation from another, and do not necessarily require or imply any actual quantity, hierarchy or order relationship between these entities or operations.

[0055] Without further limitations, the use of terms such as “comprising,” “including,” “having,” or other similar open-ended expressions in this application is intended to cover non-exclusive inclusion, which does not exclude the presence of additional elements in a process, method, or product that includes the stated elements, such that a process, method, or product that includes a list of elements may include not only those defined elements but also other elements not expressly listed, or elements inherent to such a process, method, or product.

[0056] As understood in the Examination Guidelines, in this application, expressions such as "greater than," "less than," and "exceeding" are understood to exclude the stated number; expressions such as "above," "below," and "within" are understood to include the stated number. Furthermore, in the description of the embodiments in this application, "multiple" means two or more (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups" and "multiple times," unless otherwise explicitly specified.

[0057] In the description of the embodiments of this application, the space-related expressions used, such as "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "vertical," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," indicate the orientation or positional relationship based on the orientation or positional relationship shown in the specific embodiments or drawings. They are only for the purpose of describing the specific embodiments of this application or for the reader's understanding, and do not indicate or imply that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0058] Unless otherwise expressly specified or limited, the terms "installation," "connection," "linking," "fixing," and "setting," as used in the description of the embodiments of this application, should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral arrangement; it can be a direct connection or an indirect connection through an intermediate medium; it can be a relationship of two components combined together, an interaction relationship between two components, or a connection within two structures. Those skilled in the art to which this application pertains can understand the specific meaning of the above terms in the embodiments of this application according to the specific circumstances.

[0059] Example 1

[0060] This embodiment combines Figure 1 The overall flow of the method of the present invention is described.

[0061] like Figure 1As shown, the cross-platform advertising expenditure joint budget allocation and ROI optimization method provided by this invention includes the following steps:

[0062] Step S1: Attribution Aggregation Step. Advertising campaign data and corresponding sales order data are collected from multiple advertising platforms. By reconstructing cross-platform anonymous user behavior sequences and employing a hybrid attribution model based on Shapley values, the contribution profit of each advertising touchpoint is calculated, resulting in calibrated contribution profit data.

[0063] Step S2: Model Training and Prediction. Based on the contribution profit data, a nonlinear saturated response function is fitted to each advertising campaign, and the cross-influence coefficients between different advertising channels are calculated through multivariate time series analysis to construct a joint prediction model for predicting total contribution profit under any budget allocation scheme.

[0064] Step S3: Global Optimization Solution. With the goal of maximizing the predicted total profit contribution of all advertising campaigns, the budget allocation problem is formalized as a constrained nonlinear programming problem. The total budget for the next cycle and business constraints are input, and the optimal budget allocation scheme is output through iterative solution using a sequential quadratic programming algorithm.

[0065] Step S4: Safe and automated execution steps. The optimal budget allocation scheme is decomposed into step-by-step adjustment instructions and time-based strategy adjustment instructions through a safe executor, and then sent to each advertising platform for execution. At the same time, core performance indicators are monitored in real time, and an abnormal circuit breaker mechanism is triggered when the indicators deteriorate beyond a preset threshold.

[0066] Step S5: Continuous Learning and Iteration. Collect advertising performance data for the new cycle, evaluate the optimization effect, use the feedback data to update the nonlinear saturated response function and the cross-influence coefficient, and enter the next optimization cycle.

[0067] The following provides a detailed explanation of each step.

[0068] The specific implementation of step S1 (attribution aggregation step) is as follows:

[0069] Data fusion and path reconstruction:

[0070] First, advertising logs and order data from various advertising platforms (such as Amazon, Google, Facebook, TikTok, etc.) are synchronized via API. Advertising logs include impression events and click events. Each event record contains: a hashed user device ID or first-party cookie, a timestamp, an advertising campaign ID, a channel identifier, and the event type (impression or click). Order data includes: order amount, order profit, user device ID, and order time.

[0071] Using hashed user device IDs or first-party cookies as associated keys, cross-platform advertising events are sorted by timestamps to reconstruct a complete anonymous user interaction sequence for each final converted order. This sequence begins with the user's first ad exposure and ends with the final click to complete the conversion, including all cross-platform ad touchpoints in between.

[0072] For example, the reconstructed user interaction sequence might be: user sees a brand ad on Facebook (exposure) → user searches for keywords on Google and clicks on a search ad (click) → user browses a product detail page on Amazon (exposure) → user clicks on an Amazon product promotion ad (click) → completes the purchase.

[0073] Contribution profit calculation based on Shapley value:

[0074] The attribution problem is modeled as a cooperative game. Let... This is the set of all touchpoints in the user interaction sequence corresponding to a given converted order. = Each ad touchpoint is considered a "player," and the order profit... It is considered "cooperative benefits".

[0075] For any subset of touch points ⊆ Define the characteristic function For only through subset The touchpoints in the middle (i.e., the points where the user only touches) The probability that an advertisement (in the context of a web application) can achieve a conversion. This probability can be determined by statistically analyzing historical data that satisfies the condition that "the set of touchpoints is exactly..." The frequency at which a conversion ultimately occurs during the user journey is used to estimate this.

[0076] Contact Shapley value weights The average marginal contribution, representing the contribution across all possible joining orders, is calculated using the following formula:

[0077]

[0078] in, yes Not included any subset, Representing a subset The number of middle contacts Indicates contact point Join a subset The resulting increase in marginal conversion probability.

[0079] After calculating the Shapley value weight for each touchpoint, for a total profit of The order, the first in its path The profit contribution per touchpoint is:

[0080]

[0081] The system aggregates the contribution profit of each touchpoint across all orders, categorized by campaign ID and channel, and outputs a calibrated "Contribution Profit" dataset. This dataset reflects the true value contribution of each campaign after considering cross-platform synergies, serving as the value basis for subsequent optimization steps.

[0082] The specific implementation of step S2 (model training and prediction step) is as follows:

[0083] Create an ad response function:

[0084] Fit historical data to each core advertising campaign unit (such as a product promotion ad group) and establish a budget. The non-linear relationship between output (profit contribution) and [other factors]. This embodiment uses a saturated growth model—the Michaelis-Menten function—for fitting:

[0085]

[0086] in, The amount of budget allocated to this advertising campaign; This represents the maximum output potential of the advertising campaign, that is, the maximum profit that can be achieved when the budget approaches infinity; It is the half-saturation constant, which is the budget amount required to reach half of the maximum output.

[0087] Specific parameters for each advertising campaign are obtained by regressing historical data. and Specifically, we collected data points on the actual profit contribution of the advertising campaign at different budget levels over a past period. The curve is fitted using the nonlinear least squares method, minimizing the following loss function:

[0088]

[0089] Solving for the results and The estimated value.

[0090] Quantify the cross-influence between channels:

[0091] The synergistic and crowding-out effects among different advertising channels are measured using multivariate time series analysis (such as vector autoregression models, VAR). Let there be a total of... Each advertising channel has a daily budget expenditure sequence. and the daily output sequence of each channel (Output equals profit contribution). Construct a VAR(p) model:

[0092]

[0093] in This is a vector containing budget and output variables for each channel. This is the coefficient matrix. Through model estimation, the impact of budget changes in each channel on the output of other channels can be obtained.

[0094] The final cross-influence coefficient matrix is ​​formed. , of which elements Indicates channel For every unit increase in the budget, the channel The incremental effect of output (positive values ​​indicate synergistic promotion, negative values ​​indicate crowding out competition). For example, This represents the impact coefficient of Facebook's brand advertising budget on Amazon's in-site advertising output.

[0095] Perform a "What-if" simulation prediction:

[0096] When the user inputs a hypothetical budget allocation vector At that time, the system calculates the predicted total contribution profit according to the following steps:

[0097] The first step, basic forecasting: Substitute the budget for each channel into its respective response function to obtain the basic output of each channel. .

[0098] The second step is cross-impact adjustment: calculating the additional output or loss of each channel due to changes in the budgets of other channels.

[0099]

[0100] in, Indicates channel Budget for channels The impact coefficient of output.

[0101] Step 3: Summary: The predicted total contribution profit is as follows:

[0102]

[0103] This predictive model can answer complex questions such as "how will the output of channel B change if the budget of channel A is increased", providing a basis for decision-making for global optimization.

[0104] The specific implementation of step S3 (global optimization solution step) is as follows:

[0105] Optimization problem modeling:

[0106] The budget allocation problem is formalized as a constrained nonlinear programming problem.

[0107] Objective function: Maximize the predicted total profit contribution from all advertising campaigns.

[0108]

[0109] in, Total number of advertising campaigns, To be assigned to the The budget for an advertising campaign For the first The response function of an activity. This represents the cross-influence coefficient.

[0110] The main constraints include:

[0111] (1) Overall budget constraint: ,in This is the total advertising budget for the next cycle.

[0112] (2) Upper and lower limits of channel budget: ,in and The first The minimum necessary budget and maximum affordable budget for each advertising campaign can be set according to business strategy.

[0113] (3) Return on investment constraint: ,in = , Preset Lower limit.

[0114] (4) Key performance indicator constraints: ,in = / (conversion rate), This is the preset upper limit for the cost per conversion.

[0115] Solving a sequence of quadratic programming problems:

[0116] The engine employs a Sequential Quadratic Programming (SQP) algorithm for efficient solution. This process is an iterative feedback loop:

[0117] The first step is initialization: using the current period's budget allocation or average allocation as the starting point. .

[0118] The second step is iterative solution: in each iteration... In the algorithm at the current point At this point, the original nonlinear problem is approximated as a more easily solvable quadratic programming (QP) subproblem. The objective function of this subproblem is the second-order Taylor expansion of the original objective function at the current point, with constraints of a first-order linear approximation. Solving this QP subproblem yields the optimal search direction. and step length .

[0119] Step 3: Update and convergence judgment: Update the budget allocation scheme. Repeat the above process until the change in the solution is less than the preset threshold. (For example If the value is 0.01 or the maximum number of iterations is reached (e.g., 100 iterations), then the optimal budget allocation scheme is output. .

[0120] Dynamic adjustment of competitive situation perception:

[0121] The engine integrates a competitive landscape awareness submodule. This submodule monitors the average bid fluctuations of core keywords across various advertising platforms in real time. Specifically, it obtains keyword bid data from each platform via API and calculates the ratio of the average bid for each keyword this week to the average bid for the previous week.

[0122] When the system detects a surge in competition intensity in a channel exceeding a preset threshold (e.g., 15%) week-on-week, it determines that competition in that channel has intensified. At this point, the optimization engine automatically and temporarily relaxes the ROAS constraint for that channel. For example, The constraint has been lowered from 3.0 to 2.5, allowing the engine to allocate more budget to this channel under more lenient conditions to maintain competitiveness. The constraint will be restored to its original value once competitive intensity returns to normal.

[0123] The specific implementation of step S4 (safety automation execution step) is as follows:

[0124] Gradual adjustments to the budget:

[0125] For advertising campaigns that require a significant budget increase, the safety actuator breaks them down into multiple small adjustments and sets an observation period.

[0126] For example, if an advertising campaign's budget needs to be increased from $1,000 to $2,000, a 100% increase, the safety actuator will not adjust the budget all at once, but will instead execute the following step-by-step strategy:

[0127] Step 1: Adjust the budget to $1300 and start timing and observation.

[0128] Observation period (3 hours): The system monitors the consumption rate and cost per conversion (CPA) of the activity in real time.

[0129] If the consumption rate is normal (not exceeding 80% of the budget limit) and the CPA does not exceed the preset threshold, then proceed to step two: adjust the budget to $1600.

[0130] Second observation period (3 hours): Monitor the above indicators again.

[0131] If the indicators are normal, proceed to step three: adjust the budget to the target value of $2000.

[0132] The entire adjustment process was completed smoothly within 12-24 hours, avoiding runaway spending or soaring CPA caused by a one-time large budget increase.

[0133] Adjustment of bidding time-sharing strategy:

[0134] For advertising campaigns that require an increase in bids, the safety actuator performs time-based segmentation based on the advertising traffic time period.

[0135] For example, the bid for a certain keyword needs to be increased from $1.5 to $3.0. The security executor executes the following time-sharing strategy:

[0136] Step 1: During a traffic slump (e.g., 2:00 AM - 4:00 AM), test-increase the bid to $1.8. Monitor the click-through rate changes during this period.

[0137] If the click-through rate is normal (no abnormal drop), increase the bid to $2.3 during the next traffic peak (e.g., 8:00 PM - 10:00 PM) to capture peak traffic.

[0138] After verifying that the conversion rate and CPA are normal at this bid, the bid will be finally adjusted to the target value of $3.0 during the peak period of the following day.

[0139] This time-sharing strategy can effectively test market reactions, verify the effect of adjustments during low-risk periods, and then be fully applied during high-value periods.

[0140] Abnormal circuit breaker mechanism:

[0141] After each instruction is executed, the system monitors key performance indicators in real time, including cost per conversion (CPA) and impression share.

[0142] Set a circuit breaker threshold, for example: CPA spikes by more than 50% within 2 hours. When the metric is detected to deteriorate beyond this threshold, the system immediately triggers the circuit breaker.

[0143] Suspend all subsequent automatic adjustment commands for this advertising campaign;

[0144] Automatically rollback to the previous security settings (budget and bid are restored to the values ​​before the circuit breaker was triggered);

[0145] Send alerts to operations personnel, including information such as circuit breaker time, abnormal indicators, current values ​​and thresholds, and the sequence of operations that have been executed.

[0146] After receiving the alert, the operations staff will intervene to analyze the cause. Once the problem is confirmed to be resolved, the automatic adjustment can be manually restored.

[0147] The specific implementation of step S5 (continuous learning and iterative step) is as follows:

[0148] After a campaign cycle is completed, the system collects actual advertising performance data for the new cycle, including actual budget expenditure, actual profit contribution, and actual conversion rate for each advertising campaign.

[0149] The system compares the actual data with the predicted data in step S2 to evaluate the optimization effect. Evaluation indicators include: prediction accuracy (such as MAPE), ROI improvement, budget utilization, etc.

[0150] Based on the evaluation results, the system updates the model parameters:

[0151] Update the ad response function: Add the actual data from the new period to the historical dataset and refit the Michaelis-Menten function parameters for each ad campaign. and .

[0152] Update cross-influence coefficients: Add the time series data from the new period to the VAR model and re-estimate the cross-influence coefficient matrix. .

[0153] The updated model enters the next optimization cycle, achieving a complete closed loop of "data-decision-execution-evaluation-update". As deployment data continues to accumulate, the model's prediction accuracy and optimization effect continue to improve.

[0154] Example 2

[0155] This embodiment uses an annual promotional event as an example to illustrate the complete execution flow of the method of the present invention.

[0156] Suppose a cross-border e-commerce seller has a total budget of $100,000 during the Prime Day promotion and needs to allocate it among four channels: Amazon in-site advertising, Google search advertising, Facebook brand advertising, and TikTok referral advertising.

[0157] Step S1 (Attribution Aggregation): The system collects advertising data from the pre-sale period and historical sales events, and calculates the profit contribution of each channel using the Shapley value attribution model. Attribution results show that Facebook brand advertising has a significant positive impact—users who viewed Facebook ads saw a 2.5x increase in conversion rates on Amazon.

[0158] Step S2 (Model Training and Prediction): The system fits the response functions of each channel and finds that Amazon in-site advertising has diminishing marginal returns after the budget exceeds $40,000; while TikTok advertising shows exponential growth after the budget reaches $15,000. VAR model analysis shows that Facebook advertising has a positive synergistic effect on Amazon in-site advertising (coefficient +0.08), while Google Shopping advertising is competitive with Amazon in-site advertising (coefficient -0.05).

[0159] Step S3 (Global Optimization): The optimization engine aims to maximize total profit contribution and solves for the optimal budget allocation within a total budget constraint of $100,000. During the pre-sale period, the engine suggests allocating more budget to Facebook and TikTok for brand building. On the day of the promotion, the engine monitors the conversion cost of each channel in real time. When the CPA of Google Shopping ads spikes, a portion of the budget is dynamically reallocated to TikTok and Amazon display ads.

[0160] Step S4 (Safe Automated Execution): The budget adjustment instructions output by the engine are distributed step by step through the safe executor. For example, if TikTok's budget is increased from $10,000 to $25,000, the executor completes this in 3 steps, observing the CPA metric every 2 hours between each step. During the execution process, the system monitors in real time, and no circuit breaker is triggered.

[0161] Step S5 (Continuous Learning and Iteration): After the promotion ends, the system collects actual delivery data, updates the response function parameters and cross-influence coefficients of each channel, and prepares for the next round of promotion optimization.

[0162] Using the methods described above, the seller achieved a 400% increase in total sales and a 35% year-on-year increase in overall ROAS, despite a 250% increase in overall advertising spending during the promotional period.

[0163] Example 3

[0164] See Figure 2The present invention also provides a joint budget allocation and ROI optimization system for cross-platform advertising expenditure, which is used to implement the joint budget allocation and ROI optimization method for cross-platform advertising expenditure, including: a cross-platform advertising data attribution module 1, a joint prediction and simulation module 2, a dynamic budget optimization engine 3, and an automatic execution and strategy management module 4.

[0165] Cross-platform advertising data attribution module 1 is used to collect advertising campaign data and corresponding sales order data from multiple advertising platforms. By reconstructing the cross-platform user anonymous behavior sequence and using a hybrid attribution model based on Shapley values, the contribution profit of each advertising touchpoint is calculated to obtain calibrated contribution profit data.

[0166] The joint prediction and simulation module 2 is connected to the cross-platform advertising data attribution module 1. It is used to fit a nonlinear saturated response function for each advertising campaign based on the contribution profit data, and to calculate the cross-influence coefficient between different advertising channels through multivariate time series analysis, so as to construct a joint prediction model for predicting the total contribution profit under any budget allocation scheme.

[0167] The dynamic budget optimization engine 3, connected to the joint prediction and simulation module 2, is used to formalize the budget allocation problem into a constrained nonlinear programming problem with the goal of maximizing the predicted total contribution profit of all advertising activities, and iteratively solves the problem through a sequential quadratic programming algorithm to output the optimal budget allocation scheme.

[0168] The automatic execution and strategy management module 4 is connected to the dynamic budget optimization engine 3 and the API of the external advertising platform. It is used to decompose the optimal budget allocation scheme into step-by-step adjustment instructions and time-based strategy adjustment instructions through the safe executor, and send them to each advertising platform for execution. At the same time, it monitors the core performance indicators in real time. When the core performance indicators are detected to deteriorate beyond the preset threshold, the abnormal circuit breaker mechanism is triggered.

[0169] Unlike existing technologies, the above technical solution has the following advantages: it protects the system used to implement the method, including a cross-platform advertising data attribution module, a joint prediction and simulation module, a dynamic budget optimization engine, and an automatic execution and strategy management module. This system solidifies the above method into a modular architecture, forming a complete data flow and decision-making chain between the modules, which can be directly deployed in cross-border e-commerce advertising operation scenarios. This technical effect enables the method to be implemented in a system form, achieving intelligent management of the entire chain from data collection, attribution, prediction, optimization to automated execution, and has clear technical integration and industrial application value.

[0170] In summary, the present invention provides a method and system for joint budget allocation and ROI optimization of cross-platform advertising expenditures. Through global optimization, limited advertising budgets can be allocated to channels and activities with the highest marginal returns, which is expected to improve the overall advertising ROI by 15%-30%. Based on hybrid attribution and joint prediction, the decision-making basis is upgraded from one-sided "last click" data to full-link data reflecting the real user journey and channel synergy, fundamentally improving the quality of decision-making. The system can perceive market changes in real time and automatically adjust strategies, keeping advertising delivery within an efficient range and reducing the cost and lag of manual monitoring and frequent adjustments. From data to decision-making, and then to execution and feedback, a complete, self-learning optimization loop is formed, greatly improving the automation level and scalability of advertising operations.

[0171] Finally, it should be noted that although the above embodiments have been described in the text and drawings of this application, this should not limit the scope of patent protection of this application. Any technical solutions that are based on the essential concept of this application and utilize the content described in the text and drawings of this application, resulting in equivalent structural or procedural substitutions or modifications, as well as the direct or indirect application of the technical solutions of the above embodiments to other related technical fields, are all included within the scope of patent protection of this application.

Claims

1. A method for joint budget allocation and ROI optimization of cross-platform advertising expenditures, characterized in that, Includes the following steps: S1: Collect advertising campaign data and corresponding sales order data from multiple advertising platforms, reconstruct cross-platform anonymous user behavior sequences and use a hybrid attribution model based on Shapley values ​​to calculate the contribution profit of each advertising touchpoint and obtain calibrated contribution profit data. S2: Based on the aforementioned profit contribution data, a nonlinear saturated response function is fitted to each advertising campaign, and the cross-influence coefficient between different advertising channels is calculated through multivariate time series analysis to construct a joint prediction model for predicting total profit contribution under any budget allocation scheme. S3: With the goal of maximizing the predicted total profit contribution of all advertising campaigns, the budget allocation problem is formalized into a constrained nonlinear programming problem, and iteratively solved using a sequential quadratic programming algorithm to output the optimal budget allocation scheme. S4: The optimal budget allocation scheme is decomposed into step-by-step adjustment instructions and time-based strategy adjustment instructions through a safety executor, and then sent to each advertising platform for execution. At the same time, the core performance indicators are monitored in real time. When the core performance indicators are detected to deteriorate beyond the preset threshold, the abnormal circuit breaker mechanism is triggered.

2. The method for joint budget allocation and ROI optimization of cross-platform advertising expenditure according to claim 1, characterized in that, In step S1, the specific method for calculating the profit contribution of each ad touchpoint using the hybrid attribution model based on Shapley values ​​is as follows: Treating each ad touchpoint as a player in a cooperative game, and order profit as the cooperative gain, the touchpoint is calculated using the following formula. Shapley value weights : ； in, For the total set of contact points, For the set of contact points The resulting conversion probability; for a total profit of The order, the touchpoints along its path The contribution profit is .

3. The method for joint budget allocation and ROI optimization of cross-platform advertising expenditure according to claim 1, characterized in that, In step S2, the nonlinear saturated response function is the Michaelis-Menten function: ； in For the budget, To maximize output potential, The cross-influence coefficient is a half-saturation constant; the cross-influence coefficient is calculated using a vector autoregression model, forming a cross-influence coefficient matrix. , of which elements Indicates channel Budget for channels The impact on output; the joint forecasting model for a given budget allocation vector The method for calculating the predicted total contribution profit is as follows: 。 4. The method for joint budget allocation and ROI optimization of cross-platform advertising expenditures according to claim 1, characterized in that, In step S3, the objective function of the constrained nonlinear programming problem is: ； The constraints include total budget constraints, upper and lower limits of channel budgets, return on investment constraints, and key performance indicator constraints. During the solution process, the sequential quadratic programming algorithm approximates the original nonlinear problem into a quadratic programming subproblem in each iteration, solves the search direction and step size, and updates the budget allocation scheme until the change in the scheme is less than the preset threshold or the maximum number of iterations is reached.

5. The method for joint budget allocation and ROI optimization of cross-platform advertising expenditure according to claim 1, characterized in that, Step S3 also includes a competition-aware dynamic adjustment sub-step: The system monitors the average bid fluctuations of core keywords in the market in real time. When it detects that the competition intensity of a certain channel has surged week-on-week and exceeded the preset threshold, it automatically and temporarily relaxes the return on investment constraint of that channel and then performs global optimization.

6. The method for joint budget allocation and ROI optimization of cross-platform advertising expenditure according to claim 1, characterized in that, In step S4, the specific method of the step-by-step gradual adjustment is as follows: For budget increases, the target budget value is broken down into multiple incremental adjustment stages. An observation period is set after each stage adjustment to monitor the consumption rate and cost per conversion. The next adjustment is executed only after the consumption rate and cost per conversion are normal, until the target budget value is reached.

7. The method for joint budget allocation and ROI optimization of cross-platform advertising expenditure according to claim 1, characterized in that, In step S4, the specific method for adjusting the time-sharing strategy is as follows: For bid increases, the strategy should be broken down according to the time period of ad traffic. First, test the increase during the traffic trough period. After verifying that the click-through rate is normal, increase it to the median value during the traffic peak period, and finally adjust it to the target bid value during the subsequent peak period.

8. The method for joint budget allocation and ROI optimization of cross-platform advertising expenditure according to claim 1, characterized in that, In step S4, the specific implementation of the abnormal circuit breaker mechanism is as follows: After each instruction is executed, the cost per conversion and impression share are monitored in real time. When the cost per conversion is detected to spike above the threshold percentage within a preset time, the subsequent automatic adjustments of the current advertising campaign are immediately suspended, and the campaign is rolled back to the previous safety settings. At the same time, an alert is sent to the operations staff.

9. The method for joint budget allocation and ROI optimization of cross-platform advertising expenditure according to claim 1, characterized in that, It also includes step S5: collecting advertising performance data for the new cycle, evaluating the optimization effect, using the feedback data to update the nonlinear saturated response function and the cross-influence coefficient, and entering the next round of optimization cycle.

10. A joint budget allocation and ROI optimization system for cross-platform advertising expenditure, used to implement the joint budget allocation and ROI optimization method for cross-platform advertising expenditure as described in any one of claims 1 to 9, characterized in that, include: The cross-platform advertising data attribution module is used to collect advertising campaign data and corresponding sales order data from multiple advertising platforms. By reconstructing the cross-platform user anonymous behavior sequence and using a hybrid attribution model based on Shapley values, the module calculates the contribution profit of each advertising touchpoint and obtains calibrated contribution profit data. The joint prediction and simulation module, connected to the cross-platform advertising data attribution module, is used to fit a nonlinear saturated response function for each advertising campaign based on the contribution profit data, and to calculate the cross-influence coefficient between different advertising channels through multivariate time series analysis, thereby constructing a joint prediction model for predicting the total contribution profit under any budget allocation scheme. The dynamic budget optimization engine, connected to the joint prediction and simulation module, is used to formalize the budget allocation problem into a constrained nonlinear programming problem with the goal of maximizing the predicted total profit contribution of all advertising campaigns. It then iteratively solves the problem using a sequential quadratic programming algorithm to output the optimal budget allocation scheme. The automatic execution and strategy management module connects to the dynamic budget optimization engine and the API of the external advertising platform. It is used to decompose the optimal budget allocation scheme into step-by-step adjustment instructions and time-based strategy adjustment instructions through the safe executor, and send them to each advertising platform for execution. At the same time, it monitors the core performance indicators in real time. When the core performance indicators are detected to deteriorate beyond the preset threshold, the abnormal circuit breaker mechanism is triggered.

Images