A method for effect quantification and optimization of scenario-based marketing planning

Abstract

The application relates to the technical field of big data analysis, and discloses a scenario-based marketing plan effect quantification optimization method, which comprises the following steps: step 1, scene modeling and activity data back mining; step 2, scene matching degree calculation and user layering; step 3, contact contribution decomposition and effect attribution; step 4, optimization strategy generation and small flow verification; and step 5, effect monitoring and model iteration. The scenario-based marketing plan effect quantification optimization method converts an abstract planning scene into a measurable index through scene element structural decomposition and matching degree grading, the time window is accurate to an hour segment, the space fence is based on POI to demarcate a radius, the trigger condition is converted into a numerical threshold, each exposure record is given a matching degree score of 0 to 1.0, the average matching degree of the whole activity is 0.58, the conversion rate of the 1.0 group is 56.3%, the conversion rate of the 0 group is only 11.2%, the ROI is increased by 0.18 on average when the matching degree is increased by 0.1, and data support is provided for the effectiveness judgment of scene design.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, specifically to a method for quantifying and optimizing the effectiveness of scenario-based marketing planning. Background Technology

[0002] In offline scenario-based marketing in industries such as fast-moving consumer goods and catering, planning teams typically distribute coupons in specific locations such as around office buildings and commercial districts, taking into account factors such as weather and time of day, to guide users to redeem them in stores. Taking a certain herbal tea brand's "High Temperature Cooling Coupon" campaign in the summer of 2024 as an example, the plan set conditions such as 11:30 am to 1:30 pm on weekdays, within 500 meters of office buildings, and a temperature of no less than 32°C. The budget of 98,000 yuan was allocated to five channels: WeChat Moments ads, official account push, mini-program pop-ups, store posters, and community sharing. During the 7-day campaign, it reached more than 80,000 people.

[0003] After the campaign, the operations staff followed standard practice to compile the data: exporting impressions, clicks, coupon redemptions, and transaction counts from various platforms, calculating the overall redemption rate, and then dividing it by the budget to obtain the ROI. This evaluation method, which only looks at the overall funnel metrics, suffers from three deficiencies. First, whether the exposure occurred within the scenario set by the campaign was not quantified. In this campaign, only 23.7% of the actual exposure records matched all three criteria (time, location, and target audience), while another 11.7% of the exposures were completely off-target. However, the final redemption revenue of 153,000 and ROI of 1.56 were the result of a mixed calculation of these high-match and low-match exposures, failing to distinguish the contribution of the scenario setting itself to conversion. Second, the return on investment of each channel touchpoint could not be broken down individually. From seeing the ad to redeeming in-store, users experienced an average of 2.4 touchpoints. How much did the WeChat Moments ad contribute, and how much did the official account message contribute? Conventional attribution methods cannot answer this question. In practice, WeChat Moments consumed 43% of the budget, while WeChat Official Accounts accounted for only 17%. However, post-event review revealed that WeChat Official Account messages achieved a higher redemption rate, indicating a lack of a mechanism to identify these high-efficiency touchpoints during the campaign rather than afterward. Third, there was a lack of dynamic adjustment mechanisms during the campaign. When redemption rates dropped in some stores, the operations team couldn't determine whether the issue stemmed from time slot selection, the radius of the redemption area, or the coupon value. They could only summarize the situation after the campaign ended, missing the window for optimizing resource allocation within the campaign period.

[0004] The root cause of these deficiencies lies in the fact that existing evaluation methods separate scenario planning from performance measurement: the planning phase defines detailed scenario elements, while the performance evaluation phase only focuses on the final conversion funnel. There is no quantitative correspondence between the two, nor is a closed loop formed from performance data feedback to planning parameters. Therefore, a quantitative optimization method for scenario-based marketing planning performance is needed that can quantify scenario matching, decompose touchpoint contributions, automatically generate optimization strategies, and perform controlled verification. Summary of the Invention

[0005] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method for quantifying and optimizing the effectiveness of scenario-based marketing planning. This method offers advantages such as providing data support for judging the effectiveness of scenario design and solves the problem of existing evaluation methods separating scenario planning from effect measurement.

[0006] (II) Technical Solution To achieve the goal of providing data support for judging the effectiveness of scenario design, the present invention provides the following technical solution: a method for quantitatively optimizing the effect of scenario-based marketing planning, including step 1: scenario modeling and activity data collection, step 2: scenario matching degree calculation and user segmentation, step 3: touchpoint contribution decomposition and effect attribution, step 4: optimization strategy generation and small-traffic verification, and step 5: effect monitoring and model iteration. Step 1: scenario modeling and activity data collection includes step 101: structured decomposition of scenario elements, step 102: access to full-link behavioral data, and step 103: multi-source ID integration and abnormal traffic filtering. Step 2: Scenario matching degree calculation and user segmentation includes step 201 scenario reach matching degree classification, step 202 user conversion depth marking and step 203 scenario sensitivity modeling and user grouping. Step 3: Touchpoint contribution decomposition and effect attribution includes step 301 conversion path touchpoint sequence restoration, step 302 Shapley value attribution calculation and step 303 scene matching degree-ROI cross analysis. Step 4: Optimization strategy generation and low-volume verification includes step 401 high-sensitivity feature rule extraction, step 402 budget reallocation simulation, and step 403 controlled deployment and effect comparison. Step 5: Effect monitoring and model iteration includes Step 501 real-time data dashboard deployment, Step 502 bi-weekly strategy review cycle, and Step 503 scenario strategy template accumulation.

[0007] Preferably, step 101 involves the structured decomposition of scene elements; Read the marketing plan and break down the scenario description into structured fields according to five dimensions: "time window, spatial fence, trigger conditions, target audience profile, and channel touchpoint matrix". The time window is accurate to the hour (e.g., 11:30-13:30 on weekdays). The spatial fence is defined by the radius based on Gaode POI data (500 meters around office buildings). The trigger conditions are converted into quantifiable thresholds (temperature ≥32℃, air quality index <100). The target audience profile is limited to a combination of age, gender, and consumption preference tags. The channel touchpoints record the planned ad placements and push methods. We analyzed 152 fast-moving consumer goods (FMCG) planning proposals accumulated over the past two years using the same standards to create a scenario element library, with the field missing rate controlled within 6%.

[0008] Preferably, step 102 involves full-link behavioral data access; During the campaign, data from the previous day was retrieved daily at midnight from ByteDance's advertising platform, Tencent Ads, its own mini-programs, store POS systems, and community SCRM systems. The collected fields included: user OpenID / device number, timestamp, exposure creative ID, click coordinates, coupon code, redemption time, redemption store code, payment amount, and LBS reporting coordinates. Each record was stamped with a "data collection timestamp" upon integration. The maximum number of records that could be retrieved in a single API call was set to 8,000, and any excess records were retrieved in batches.

[0009] Preferably, the process uses the MD5 hash of the mobile phone number and DeviceID as the primary keys, and unifies the user identifiers of different systems into a unified user ID through a historical ID mapping table. Multiple exposures generated by the same user within 2 seconds are deduplicated, and only the first one is retained. Records with clicks less than 0.4 seconds after exposure and no subsequent behavior are filtered and marked as suspected bot traffic. After rule filtering, the isolated forest algorithm is used to perform secondary anomaly detection on the click frequency and coupon redemption interval distribution, and 317 records of fraudulent traffic generated by 13 accounts are removed.

[0010] Preferably, step 201 involves classifying the scenario reach matching degree. For each exposure record, calculate the degree of matching between its actual triggering conditions and the planned scenario. Time matching: whether the exposure time is within the planned time window ±1 hour; Space matching: whether the user's LBS is within the fence radius when the exposure occurs; Audience matching: whether the device tag matches the target audience profile. All three matching items score 1.0 points, matching two items score 0.6 points, matching one item score 0.3 points, and matching none items score 0 points. Summary of the score distribution of all exposure records: 1.0 points accounted for 23.7%, 0.6 points accounted for 38.5%, 0.3 points accounted for 26.1%, 0 points accounted for 11.7%, and the average score for the whole event was 0.58.

[0011] Preferably, step 202 involves user conversion depth tagging; The conversion depth was divided into L1 (exposure), L2 (click / view landing page), L3 (coupon / add to cart), and L4 (redeem / payment) based on the behavioral funnel. The highest depth reached by each user was calculated, and the conversion rate of each layer was recorded. The funnel for this event was as follows: L1→L2 conversion rate 2.7%, L2→L3 conversion rate 21.4%, and L3→L4 conversion rate 42.8%. Among them, the L3→L4 conversion rate of users with a scenario matching degree of 1.0 reached 56.3%, while that of users with a scenario matching degree of 0 was only 11.2%.

[0012] Preferably, step 203 involves scene sensitivity modeling and user segmentation; 325,000 user behavior samples were extracted from 10 historical campaigns of the same product category. Using scenario matching degree, channel type, frequency of activity in the past 30 days, consumption power stratification, and time period as features, a LightGBM binary classification model was constructed to predict the probability of users completing L4 conversion. The model outputs a scenario sensitivity score of 0-1. The model has an AUC of 0.78 and a KS value of 0.41 on the test set. Users with a score ≥0.7 are classified as highly sensitive, 0.4-0.7 as moderately sensitive, and <0.4 as low sensitive.

[0013] Preferably, step 301 involves restoring the path contact sequence; For users who have completed L4 conversion, all their touchpoint events are extracted in reverse chronological order to construct a path sequence. The touchpoints are categorized into 5 types: WeChat Moments ads, official account template messages, mini program pop-ups, store poster QR codes, and community sharing. Multiple triggers of the same touchpoint by the same user within 30 minutes are merged into one, resulting in 7,831 complete conversion paths with an average path length of 2.4 touchpoints. Step 302: Calculation of Shapley value attribution; The Shapley value method of cooperative game theory was used to calculate the marginal contribution of each touchpoint in the conversion. The maximum alliance size was set to 4, and the path with more than 4 touchpoints was truncated to the last 4. The average contribution percentage of the Shapley value of each touchpoint was as follows: WeChat Moments ads 35.6%, WeChat Official Account messages 23.1%, Mini Program pop-ups 18.9%, store posters 15.2%, and community sharing 7.2%. Combined with cost data: total investment of 98,000 yuan, WeChat Moments 42,000 yuan, WeChat Official Accounts 17,000 yuan, Mini Programs 15,000 yuan, store materials 14,000 yuan, and community operations 10,000 yuan. The ROI of each touchpoint was calculated as follows: WeChat Moments 1.33, WeChat Official Accounts 2.06, Mini Programs 1.87, stores 1.41, and community 0.88. WeChat Official Accounts and Mini Programs were identified as high-efficiency touchpoints. Step 303: Scene Matching Degree - ROI Cross-analysis; Exposure records were grouped into four levels based on matching degree, and the conversion ROI of users in each group was calculated. The ROI of the 1.0 group was 2.35, the 0.6 group was 1.68, the 0.3 group was 0.97, and the 0 group was 0.52. For every 0.1 increase in matching degree, the ROI increased by an average of 0.18. This relationship curve provides a basis for subsequent budget allocation.

[0014] Preferably, step 401 involves extracting highly sensitive feature rules; Using a decision tree (maximum depth 5, minimum leaf nodes 200), rule mining was performed on the behavior and environmental characteristics of highly sensitive users, extracting three effective rules: The coupon redemption rate for WeChat official account pushes between 10:00 AM and 12:00 PM on weekdays is 19% higher than that for pushes in the evening; the redemption rate for mini-program coupons popped up within 200 meters of a store via LBS is 23% higher than that beyond 500 meters; users who have previously purchased similar products are less sensitive to discounts than users who have made different purchases, requiring more aggressive coupons (¥8 off for purchases over ¥20 vs. ¥3 off for purchases over ¥15). Based on this, scenario optimization suggestions were generated: concentrate WeChat official account push resources on weekday mornings; reduce the LBS fence radius of mini-program pop-ups from 500 meters to 200 meters; and distribute more aggressive coupons to users who have made different purchases. Step 402: Budget reallocation simulation; A random forest model is trained based on current activity data to predict redemption revenue under different budget allocation schemes. The input variables are the budget for each touchpoint, the allocation ratio of time periods, and the coupon denomination. The simulation increases the total budget from 98,000 to 105,000, reduces the consumption during inefficient time periods in WeChat Moments (20:00-22:00 in the evening), transfers the released 6,000 budget to the morning time period of the official account and the pop-up window of the mini program, and adds 1,000 for high-value coupon subsidies. Step 403: Controlled online deployment and effect comparison; Eighteen stores were selected as the experimental group and 22 stores as the control group. The experimental group applied the optimization strategy (redistribution of touchpoint time periods, narrowing of the fence, and differentiation of coupons), while the control group used the original plan. The implementation period was 14 days. After the event, the average redemption rate of the experimental group was 17.6% higher than that of the control group. The t-test showed p=0.032. The ROI of the experimental group was 1.81, and that of the control group was 1.51, which confirmed the effectiveness of the optimization. Subsequently, all stores switched to the new strategy.

[0015] Preferably, step 501 involves the deployment of a real-time data dashboard; A DataV-based visualization dashboard was built, updating the redemption amount, average scenario matching degree, and real-time ROI values ​​for each touchpoint every hour. Two warning lines were set: an SMS reminder was triggered when the redemption rate fell below the predicted lower limit by two standard deviations; and the dashboard automatically suggested suspending spending on that channel if the ROI of a certain touchpoint was below 0.8 for two consecutive days. The dashboard also displayed changes in the percentage of highly sensitive users, which gradually increased from 19.4% to 24.1% during the optimization period. Step 502: Bi-weekly strategy review cycle; The entire process is restarted every 15 days, merging newly accumulated data into the historical sample pool, and retraining the scene sensitivity model and attribution model. After the first round of review, the average scene matching score increased from 0.58 to 0.66, the proportion of highly sensitive users increased to 24.8%, and the overall ROI increased from 1.56 to 1.73. After the second round of review, the ROI further increased to 1.79, and the matching score stabilized at around 0.67. Each iteration outputs an optimization record table, including parameter changes and effect increments. Step 503: Scenario strategy template accumulation; Validated optimization rules are archived in the strategy library according to the "scenario type - industry - season" dimension. For example, the strategy template for the "summer high temperature - beverage - office building" scenario is recorded as: increase the weight of the WeChat official account's morning push to 70%, LBS fence 200 meters, and offer a coupon of 8 yuan off for purchases of 20 yuan or more for users of different categories. When a new plan is launched, the strategy library automatically matches the most similar historical scenario and recommends the initial parameter combination, which is expected to save 30% of the cold start trial budget. The strategy library is manually reviewed quarterly to remove rules that have become invalid due to seasonal changes or changes in consumer habits.

[0016] (III) Beneficial Effects Compared with existing technologies, this invention provides a method for quantifying and optimizing the effectiveness of scenario-based marketing planning, which has the following beneficial effects: 1. The method for quantifying and optimizing the effectiveness of this scenario-based marketing campaign transforms abstract planning scenarios into measurable indicators through the structured decomposition of scenario elements and matching degree grading. The time window is accurate to the hour, the spatial fence is defined based on the POI radius, the triggering conditions are converted into numerical thresholds, and each exposure record is assigned a matching degree score from 0 to 1.0. The average matching degree of the entire campaign was 0.58, the redemption conversion rate of the 1.0 group was 56.3%, and the 0 group was only 11.2%. For every 0.1 increase in matching degree, the ROI increased by an average of 0.18, providing data support for judging the effectiveness of scenario design.

[0017] 2. The method for quantifying and optimizing the effectiveness of this scenario-based marketing plan uses the Shapley value method to decompose the contribution of multi-touchpoint conversion paths, solving the problem of conversion attribution. On 7831 complete conversion paths with an average path length of 2.4 touchpoints, the marginal contribution of each touchpoint was calculated one by one, identifying that the ROI of the official account was 2.06, the mini program was 1.87, and the Moments that consumed the most budget was only 1.33, providing an actionable basis for budget allocation.

[0018] 3. The method for quantifying and optimizing the effectiveness of this scenario-based marketing plan is based on a scenario sensitivity model and decision tree rule mining. It automatically generates optimization strategies. The LightGBM model, with an AUC of 0.78 and a KS of 0.41, divides users into three groups: high, medium, and low sensitivity. The decision tree extracts executable rules from these groups, such as "the coupon redemption rate is 19% higher on weekday mornings than in the evening" and "the redemption rate increases by 23% when the fence is reduced to 200 meters." This makes optimization suggestions no longer dependent on human experience and guesswork.

[0019] 4. The method for quantifying and optimizing the effectiveness of this scenario-based marketing plan, through random forest budget simulation and controlled experiments, completes effect prediction and risk control before the strategy goes live. The simulated redemption revenue is predicted to increase from 153,000 to 181,000, with a root mean square error of 0.83 million after 50% cross-validation. The actual test results of 18 experimental stores and 22 control stores verify that the redemption rate increased by 17.6%. The ROI of the experimental group was 1.81 compared with that of the control group of 1.51, and the difference was statistically significant, ensuring the reliability of the full switch.

[0020] 5. The method for quantifying and optimizing the effectiveness of this scenario-based marketing plan establishes a bi-weekly review cycle and a strategy template accumulation mechanism to form a closed loop of continuous optimization. The data dashboard monitors the redemption rate and touchpoint ROI in real time. The model is updated with new data every 15 days. After two rounds of iteration, the overall ROI increased from 1.56 to 1.79, and the average matching degree increased from 0.58 to 0.67. Validated rules are archived in the strategy library according to scenario type. When a new plan is launched, it can be matched with historical templates, which is expected to save 30% of the cold start trial budget. The strategy library is manually reviewed every quarter to ensure timeliness. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] This solution provides a technical solution, specifically, a method for quantifying and optimizing the effectiveness of scenario-based marketing planning, including the following steps: Step 1: Scenario modeling and activity data collection; Step 101: Structural decomposition of scene elements; Read the marketing plan and break down the scenario description into structured fields according to five dimensions: "time window, spatial fence, trigger conditions, target audience profile, and channel touchpoint matrix". The time window is accurate to the hour (e.g., 11:30-13:30 on weekdays). The spatial fence is defined by the radius based on Gaode POI data (500 meters around office buildings). The trigger conditions are converted into quantifiable thresholds (temperature ≥32℃, air quality index <100). The target audience profile is limited to a combination of age, gender, and consumption preference tags. The channel touchpoints record the planned ad placements and push methods. We analyzed 152 fast-moving consumer goods (FMCG) planning proposals accumulated over the past two years using the same standards to create a scenario element library, with the field missing rate controlled within 6%. Step 102: End-to-end behavioral data access; During the campaign, data from the previous day was retrieved daily at midnight from ByteDance's advertising platform, Tencent Ads, its own mini-programs, store POS systems, and community SCRM systems. The collected fields included: user OpenID / device ID, timestamp, exposure creative ID, click coordinates, coupon code, redemption time, redemption store code, payment amount, and LBS reporting coordinates. Each record was stamped with a "data collection timestamp" upon integration. The maximum number of records that could be retrieved in a single API call was set to 8,000, and any excess records were retrieved in batches. Taking a certain herbal tea brand's summer "cooling coupon" campaign as an example, a total of 623,000 raw data records were retrieved during the 7-day campaign period. After cleaning in step 103, 581,000 valid records were retained. Step 103: Multi-source ID integration and abnormal traffic filtering; Using the MD5 hash of the phone number and DeviceID as the primary keys, user identifiers from different systems are unified into a single user ID through a historical ID mapping table. Multiple exposures generated by the same user within 2 seconds are deduplicated, and only the first one is retained. Records with clicks less than 0.4 seconds after exposure and no subsequent behavior are filtered and marked as suspected bot traffic. After rule filtering, the isolated forest algorithm is used to perform secondary anomaly detection on the click frequency and coupon redemption interval distribution, and 317 records of fraudulent traffic generated by 13 accounts are removed. The final analysis included 557,000 valid behavioral log entries, with 82,000 unique user IDs. Step 2: Calculate scene matching degree and segment users; Step 201: Scene reach matching degree classification; For each exposure record, calculate the degree of matching between its actual triggering conditions and the planned scenario. Time matching: whether the exposure time is within the planned time window ±1 hour; Space matching: whether the user's LBS is within the fence radius when the exposure occurs; Audience matching: whether the device tag matches the target audience profile. All three matching items score 1.0 points, matching two items score 0.6 points, matching one item score 0.3 points, and matching none items score 0 points. Summary of the score distribution of all exposure records: 1.0 points accounted for 23.7%, 0.6 points accounted for 38.5%, 0.3 points accounted for 26.1%, 0 points accounted for 11.7%, and the average score for the whole event was 0.58; Step 202: User conversion depth tagging; The conversion depth was divided into L1 (exposure), L2 (click / view landing page), L3 (coupon redemption / add to cart), and L4 (redemption / payment) based on the behavioral funnel. The highest depth reached by each user was calculated, and the conversion rate at each level was recorded. The funnel for this campaign was as follows: L1→L2 conversion rate 2.7%, L2→L3 conversion rate 21.4%, and L3→L4 conversion rate 42.8%. Among them, the L3→L4 conversion rate of users with a scenario matching degree of 1.0 reached 56.3%, while that of users with a matching degree of 0 was only 11.2%. Step 203: Scene sensitivity modeling and user segmentation; 325,000 user behavior samples were extracted from 10 historical campaigns of the same product category. Using scenario matching degree, channel type, frequency of activity in the past 30 days, consumption power stratification, and time period as features, a LightGBM binary classification model was constructed to predict the probability of users completing L4 conversion. The model outputs a scenario sensitivity score of 0-1. The model has an AUC of 0.78 and a KS value of 0.41 on the test set. Users with a score ≥0.7 are classified as highly sensitive, 0.4-0.7 as moderately sensitive, and <0.4 as low sensitive. In this event, the highly sensitive user group accounted for 19.4%, with an average redemption amount of 12.6 yuan per person; the moderately sensitive group accounted for 45.3%, with an average redemption amount of 4.1 yuan per person; and the lowly sensitive group accounted for 35.3%, with an average redemption amount of 0.8 yuan per person. Step 3: Touchpoint contribution decomposition and effect attribution; Step 301: Reconstruct the contact sequence of the conversion path; For users who have completed L4 conversion, all their touchpoint events are extracted in reverse chronological order to construct a path sequence. The touchpoints are categorized into 5 types: WeChat Moments ads, official account template messages, mini program pop-ups, store poster QR codes, and community sharing. Multiple triggers of the same touchpoint by the same user within 30 minutes are merged into one, resulting in 7,831 complete conversion paths with an average path length of 2.4 touchpoints. Step 302: Calculation of Shapley value attribution; The Shapley value method of cooperative game theory was used to calculate the marginal contribution of each touchpoint in the conversion. The maximum alliance size was set to 4, and the path with more than 4 touchpoints was truncated to the last 4. The average contribution percentage of the Shapley value of each touchpoint was as follows: WeChat Moments ads 35.6%, WeChat Official Account messages 23.1%, Mini Program pop-ups 18.9%, store posters 15.2%, and community sharing 7.2%. Combined with cost data: total investment of 98,000 yuan, WeChat Moments 42,000 yuan, WeChat Official Accounts 17,000 yuan, Mini Programs 15,000 yuan, store materials 14,000 yuan, and community operations 10,000 yuan. The ROI of each touchpoint was calculated as follows: WeChat Moments 1.33, WeChat Official Accounts 2.06, Mini Programs 1.87, stores 1.41, and community 0.88. WeChat Official Accounts and Mini Programs were identified as high-efficiency touchpoints. Step 303: Scene Matching Degree - ROI Cross-analysis; Exposure records were grouped into four levels based on matching degree, and the conversion ROI of users in each group was calculated. The ROI of the 1.0 group was 2.35, the 0.6 group was 1.68, the 0.3 group was 0.97, and the 0 group was 0.52. For every 0.1 increase in matching degree, the ROI increased by an average of 0.18. This relationship curve provides a basis for subsequent budget allocation. Step 4: Optimize strategy generation and small-scale verification; Step 401: Extraction of highly sensitive feature rules; Using a decision tree (maximum depth 5, minimum leaf nodes 200), rule mining was performed on the behavior and environmental characteristics of highly sensitive users, extracting three effective rules: The coupon redemption rate for WeChat official account pushes between 10:00 AM and 12:00 PM on weekdays is 19% higher than that for pushes in the evening; the redemption rate for mini-program coupons popped up within 200 meters of a store via LBS is 23% higher than that beyond 500 meters; users who have previously purchased similar products are less sensitive to discounts than users who have made different purchases, requiring more aggressive coupons (¥8 off for purchases over ¥20 vs. ¥3 off for purchases over ¥15). Based on this, scenario optimization suggestions were generated: concentrate WeChat official account push resources on weekday mornings; reduce the LBS fence radius of mini-program pop-ups from 500 meters to 200 meters; and distribute more aggressive coupons to users who have made different purchases. Step 402: Budget reallocation simulation; A random forest model is trained based on current activity data to predict redemption revenue under different budget allocation schemes. The input variables are the budget for each touchpoint, the allocation ratio of time periods, and the coupon denomination. The simulation increases the total budget from 98,000 to 105,000, reduces the consumption during inefficient time periods in WeChat Moments (20:00-22:00 in the evening), transfers the released 6,000 budget to the morning time period of the official account and the pop-up window of the mini program, and adds 1,000 for high-value coupon subsidies. The simulated output predicted that the write-off revenue increased from 153,000 to 181,000, and the ROI increased from 1.56 to 1.72. After 50% cross-validation, the root mean square error of the prediction was 8,300 yuan. Step 403: Controlled online deployment and effect comparison; Eighteen stores were selected as the experimental group and 22 stores as the control group. The experimental group applied the optimization strategy (redistribution of touchpoint time periods, narrowing of the boundary, and differentiation of coupons), while the control group used the original plan. The implementation period was 14 days. After the event, the average redemption rate of the experimental group was 17.6% higher than that of the control group (t-test p=0.032). The ROI of the experimental group was 1.81, and that of the control group was 1.51, confirming the effectiveness of the optimization. Subsequently, all stores switched to the new strategy. Step 5: Performance monitoring and model iteration; Step 501: Deploy the real-time data dashboard; A DataV-based visualization dashboard was built, updating the redemption amount, average scenario matching degree, and real-time ROI values ​​for each touchpoint every hour. Two warning lines were set: an SMS reminder was triggered when the redemption rate fell below the predicted lower limit by two standard deviations; and the dashboard automatically suggested suspending spending on that channel if the ROI of a certain touchpoint was below 0.8 for two consecutive days. The dashboard also displayed changes in the percentage of highly sensitive users, which gradually increased from 19.4% to 24.1% during the optimization period. Step 502: Bi-weekly strategy review cycle; The entire process is restarted every 15 days, merging newly accumulated data into the historical sample pool, and retraining the scene sensitivity model and attribution model. After the first round of review, the average scene matching score increased from 0.58 to 0.66, the proportion of highly sensitive users increased to 24.8%, and the overall ROI increased from 1.56 to 1.73. After the second round of review, the ROI further increased to 1.79, and the matching score stabilized at around 0.67. Each iteration outputs an optimization record table, including parameter changes and effect increments. Step 503: Scenario strategy template accumulation; Validated optimization rules are archived in the strategy library according to the "scenario type - industry - season" dimension. For example, the strategy template for the "summer high temperature - beverage - office building" scenario is recorded as: increase the weight of the WeChat official account's morning push to 70%, LBS fence 200 meters, and offer a coupon of 8 yuan off for purchases of 20 yuan or more for users of different categories. When a new plan is launched, the strategy library automatically matches the most similar historical scenario and recommends the initial parameter combination, which is expected to save 30% of the cold start trial budget. The strategy library is manually reviewed every quarter to remove rules that have become invalid due to seasonal changes or changes in consumer habits. Furthermore, by structurally decomposing scene elements and classifying matching scores, the abstract planning scene is transformed into measurable indicators. The time window is accurate to the hour, the spatial fence is defined based on the POI radius, the triggering conditions are converted into numerical thresholds, and each exposure record is assigned a matching score from 0 to 1.0. The average matching score for the entire event was 0.58, the redemption conversion rate of the 1.0 group was 56.3%, and the 0 group was only 11.2%. For every 0.1 increase in matching score, the ROI increased by an average of 0.18, providing data support for judging the effectiveness of scene design. Furthermore, the Shapley value method was used to decompose the contribution of multi-touchpoint conversion paths, solving the conversion attribution problem. On 7831 complete conversion paths with an average path length of 2.4 touchpoints, the marginal contribution of each touchpoint was calculated one by one, identifying that the ROI of the official account was 2.06, the mini program was 1.87, and the Moments that consumed the most budget was only 1.33, providing an actionable basis for budget allocation. Furthermore, based on the scene sensitivity model and decision tree rule mining, optimization strategies are automatically generated. The LightGBM model divides users into three groups of high, medium and low sensitivity with an AUC of 0.78 and a KS of 0.41. The decision tree extracts executable rules such as "the coupon redemption rate of WeChat official account push is 19% higher in the morning than in the evening" and "the redemption rate increases by 23% when the fence is reduced to 200 meters", so that optimization suggestions no longer rely on human experience and guesswork. Furthermore, through random forest budget simulation and controlled experiments, effect prediction and risk control were completed before the strategy went live. The simulated redemption revenue was predicted to increase from 153,000 to 181,000, with a root mean square error of 0.83 million after 50% cross-validation. The actual test results of 18 experimental stores and 22 control stores verified that the redemption rate increased by 17.6%. The ROI of the experimental group was 1.81 compared with that of the control group of 1.51, and the difference was statistically significant, ensuring the reliability of the full switch. Furthermore, a bi-weekly review cycle and strategy template accumulation mechanism were established to form a closed loop for continuous optimization. Data dashboards monitor the reimbursement rate and touchpoint ROI in real time. The model is updated with new data every 15 days. After two rounds of iteration, the overall ROI increased from 1.56 to 1.79, and the average matching degree increased from 0.58 to 0.67. Validated rules are archived in the strategy library according to scenario type. When a new plan is launched, it can be matched with historical templates, which is expected to save 30% of the cold start trial and error budget. The strategy library is manually reviewed every quarter to ensure timeliness. Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for quantifying and optimizing the effectiveness of scenario-based marketing planning, comprising: Step 1: scenario modeling and activity data collection; Step 2: scenario matching degree calculation and user segmentation; Step 3: touchpoint contribution decomposition and effect attribution; Step 4: optimization strategy generation and small-scale validation; and Step 5: effect monitoring and model iteration, characterized in that: Step 1: Scene modeling and activity data collection includes step 101, structured decomposition of scene elements, step 102, access to full-link behavior data, and step 103, multi-source ID integration and abnormal traffic filtering. Step 2: Scenario matching degree calculation and user segmentation includes step 201 scenario reach matching degree classification, step 202 user conversion depth marking and step 203 scenario sensitivity modeling and user grouping. Step 3: Touchpoint contribution decomposition and effect attribution includes step 301 conversion path touchpoint sequence restoration, step 302 Shapley value attribution calculation and step 303 scene matching degree-ROI cross analysis. Step 4: Optimization strategy generation and low-volume verification includes step 401 high-sensitivity feature rule extraction, step 402 budget reallocation simulation, and step 403 controlled deployment and effect comparison. Step 5: Effect monitoring and model iteration includes Step 501 real-time data dashboard deployment, Step 502 bi-weekly strategy review cycle, and Step 503 scenario strategy template accumulation.

2. The method for quantifying and optimizing the effectiveness of scenario-based marketing planning according to claim 1, characterized in that: Step 101 involves the structured decomposition of scene elements. Read the marketing plan and break down the scenario description into structured fields according to five dimensions: "time window, spatial fence, trigger conditions, target audience profile, and channel touchpoint matrix". The time window is accurate to the hour (e.g., 11:30-13:30 on weekdays). The spatial fence is defined by the radius based on Gaode POI data (500 meters around office buildings). The trigger conditions are converted into quantifiable thresholds (temperature ≥32℃, air quality index <100). The target audience profile is limited to a combination of age, gender, and consumption preference tags. The channel touchpoints record the planned ad placements and push methods. We analyzed 152 fast-moving consumer goods (FMCG) planning proposals accumulated over the past two years using the same standards to create a scenario element library, with the field missing rate controlled within 6%.

3. The method for quantifying and optimizing the effectiveness of scenario-based marketing planning according to claim 1, characterized in that: Step 102 involves full-link behavioral data access. During the campaign, data from the previous day was retrieved daily at midnight from ByteDance's advertising platform, Tencent Ads, its own mini-programs, store POS systems, and community SCRM systems. The collected fields included: user OpenID / device ID, timestamp, exposure creative ID, click coordinates, coupon code, redemption time, redemption store code, payment amount, and LBS reporting coordinates. Each record was stamped with a "data collection timestamp" upon integration. The maximum number of records that could be retrieved in a single API call was set to 8,000, and any excess records were retrieved in batches.

4. The method for quantifying and optimizing the effectiveness of scenario-based marketing planning according to claim 1, characterized in that: The system uses the MD5 hash of the phone number and DeviceID as the primary keys, and a historical ID mapping table to unify user IDs from different systems into a unified user ID. Multiple exposures generated by the same user within 2 seconds are deduplicated, and only the first one is retained. Records with clicks less than 0.4 seconds after exposure and no subsequent behavior are filtered and marked as suspected bot traffic. After rule filtering, the isolated forest algorithm is used to perform secondary anomaly detection on the distribution of click frequency and coupon redemption interval, and 317 records of fraudulent traffic generated by 13 accounts are removed.

5. The method for quantifying and optimizing the effectiveness of scenario-based marketing planning according to claim 1, characterized in that: Step 201: Scene reach matching degree classification; For each exposure record, calculate the degree of matching between its actual triggering conditions and the planned scenario. Time matching: whether the exposure time is within the planned time window ±1 hour; Space matching: whether the user's LBS is within the fence radius when the exposure occurs; Audience matching: whether the device tag matches the target audience profile. All three matching items score 1.0 points, matching two items score 0.6 points, matching one item score 0.3 points, and matching none items score 0 points. Summary of the score distribution of all exposure records: 1.0 points accounted for 23.7%, 0.6 points accounted for 38.5%, 0.3 points accounted for 26.1%, 0 points accounted for 11.7%, and the average score for the whole event was 0.

58.

6. The method for quantifying and optimizing the effectiveness of scenario-based marketing planning according to claim 1, characterized in that: Step 202: User conversion depth marking; The conversion depth was divided into L1 (exposure), L2 (click / view landing page), L3 (coupon / add to cart), and L4 (redeem / payment) based on the behavioral funnel. The highest depth reached by each user was calculated, and the conversion rate of each layer was recorded. The funnel for this event was as follows: L1→L2 conversion rate 2.7%, L2→L3 conversion rate 21.4%, and L3→L4 conversion rate 42.8%. Among them, the L3→L4 conversion rate of users with a scenario matching degree of 1.0 reached 56.3%, while that of users with a scenario matching degree of 0 was only 11.2%.

7. The method for quantifying and optimizing the effectiveness of scenario-based marketing planning according to claim 1, characterized in that: Step 203 involves scenario sensitivity modeling and user segmentation. 325,000 user behavior samples were extracted from 10 historical campaigns of the same product category. Using scenario matching degree, channel type, frequency of activity in the past 30 days, consumption power stratification, and time period as features, a LightGBM binary classification model was constructed to predict the probability of users completing L4 conversion. The model outputs a scenario sensitivity score of 0-1. The model has an AUC of 0.78 and a KS value of 0.41 on the test set. Users with a score ≥0.7 are classified as highly sensitive, 0.4-0.7 as moderately sensitive, and <0.4 as low sensitive.

8. The method for quantifying and optimizing the effectiveness of scenario-based marketing planning according to claim 1, characterized in that: Step 301 involves restoring the conversion path contact sequence. For users who have completed L4 conversion, all their touchpoint events are extracted in reverse chronological order to construct a path sequence. The touchpoints are categorized into 5 types: WeChat Moments ads, official account template messages, mini program pop-ups, store poster QR codes, and community sharing. Multiple triggers of the same touchpoint by the same user within 30 minutes are merged into one, resulting in 7,831 complete conversion paths with an average path length of 2.4 touchpoints. Step 302: Calculation of Shapley value attribution; The Shapley value method of cooperative game theory was used to calculate the marginal contribution of each touchpoint in the conversion. The maximum alliance size was set to 4, and the path with more than 4 touchpoints was truncated to the last 4. The average contribution percentage of the Shapley value of each touchpoint was as follows: WeChat Moments ads 35.6%, WeChat Official Account messages 23.1%, Mini Program pop-ups 18.9%, store posters 15.2%, and community sharing 7.2%. Combined with cost data: total investment of 98,000 yuan, WeChat Moments 42,000 yuan, WeChat Official Accounts 17,000 yuan, Mini Programs 15,000 yuan, store materials 14,000 yuan, and community operations 10,000 yuan. The ROI of each touchpoint was calculated as follows: WeChat Moments 1.33, WeChat Official Accounts 2.06, Mini Programs 1.87, stores 1.41, and community 0.

88. WeChat Official Accounts and Mini Programs were identified as high-efficiency touchpoints. Step 303: Scene Matching Degree - ROI Cross-analysis; Exposure records were grouped into four levels based on matching degree, and the conversion ROI of users in each group was calculated. The ROI of the 1.0 group was 2.35, the 0.6 group was 1.68, the 0.3 group was 0.97, and the 0 group was 0.

52. For every 0.1 increase in matching degree, the ROI increased by an average of 0.

18. This relationship curve provides a basis for subsequent budget allocation.

9. The method for quantifying and optimizing the effectiveness of scenario-based marketing planning according to claim 1, characterized in that: Step 401 involves the extraction of highly sensitive feature rules. Using a decision tree (maximum depth 5, minimum leaf nodes 200), rule mining was performed on the behavior and environmental characteristics of highly sensitive users, extracting three effective rules: The coupon redemption rate for WeChat official account pushes between 10:00 AM and 12:00 PM on weekdays is 19% higher than that for pushes in the evening; the redemption rate for mini-program coupons popped up within 200 meters of a store via LBS is 23% higher than that beyond 500 meters; users who have previously purchased similar products are less sensitive to discounts than users who have made different purchases, requiring more aggressive coupons (¥8 off for purchases over ¥20 vs. ¥3 off for purchases over ¥15). Based on this, scenario optimization suggestions were generated: concentrate WeChat official account push resources on weekday mornings; reduce the LBS fence radius of mini-program pop-ups from 500 meters to 200 meters; and distribute more aggressive coupons to users who have made different purchases. Step 402: Budget reallocation simulation; A random forest model is trained based on current activity data to predict redemption revenue under different budget allocation schemes. The input variables are the budget for each touchpoint, the allocation ratio of time periods, and the coupon denomination. The simulation increases the total budget from 98,000 to 105,000, reduces the consumption during inefficient time periods in WeChat Moments (20:00-22:00 in the evening), transfers the released 6,000 budget to the morning time period of the official account and the pop-up window of the mini program, and adds 1,000 for high-value coupon subsidies. Step 403: Controlled online deployment and effect comparison; Eighteen stores were selected as the experimental group and 22 stores as the control group. The experimental group applied the optimization strategy (redistribution of touchpoint time periods, narrowing of the fence, and differentiation of coupons), while the control group used the original plan. The implementation period was 14 days. After the event, the average redemption rate of the experimental group was 17.6% higher than that of the control group. The t-test showed p=0.

032. The ROI of the experimental group was 1.81, and that of the control group was 1.51, which confirmed the effectiveness of the optimization. Subsequently, all stores switched to the new strategy.

10. The method for quantifying and optimizing the effectiveness of scenario-based marketing planning according to claim 1, characterized in that: Step 501: Deployment of real-time data dashboard; Build a visualization dashboard based on DataV, which refreshes the reimbursement amount, average scenario matching degree, and real-time ROI value of each touchpoint every hour. Set two warning lines: trigger SMS reminder when the reimbursement rate is lower than the prediction lower limit by 2 standard deviations. If the ROI of a certain touchpoint is below 0.8 for two consecutive days, it will automatically suggest pausing spending on that channel. The dashboard will also display the change in the percentage of highly sensitive users. During the optimization period, this value gradually increased from 19.4% to 24.1%. Step 502: Bi-weekly strategy review cycle; The entire process is restarted every 15 days, merging newly accumulated data into the historical sample pool, and retraining the scene sensitivity model and attribution model. After the first round of review, the average scene matching score increased from 0.58 to 0.66, the proportion of highly sensitive users increased to 24.8%, and the overall ROI increased from 1.56 to 1.

73. After the second round of review, the ROI further increased to 1.79, and the matching score stabilized at around 0.

67. Each iteration outputs an optimization record table, including parameter changes and effect increments. Step 503: Scenario strategy template accumulation; Validated optimization rules are archived in the strategy library according to the "scenario type - industry - season" dimension. For example, the strategy template for the "summer high temperature - beverage - office building" scenario is recorded as: increase the weight of WeChat official account push in the morning to 70%, LBS fence 200 meters, and offer a coupon of 8 yuan off for purchases of 20 yuan or more for users of different categories. When a new plan is launched, the strategy library automatically matches the most similar historical scenario and recommends the initial parameter combination, which is expected to save 30% of the cold start trial budget. The strategy library is manually reviewed quarterly to remove rules that have become invalid due to seasonal changes or changes in consumer habits.