It's not a display template, it's a collaborative mechanism to turn insight into shelf action.

01 Facts & context

Pepsi discloses the Perfect Store system and its door-to-door tools

2026-10-20, Pepsi publishes articles on the Perfect Store system for retailers in its news and media articles. According to the article, the system combines the experience of shopkeepers with product insights, porter analysis, technical tools and front-line sales representatives to help retailers understand the trends of shoppers and product classes, design strategies, land in the market and measure impact. The article quotes Miguel Suadi, Senior Vice-President of Pepsi International Business, who says that retailing is becoming more complex, with buyers crossing more channels, consumer scenes and door-to-door flows, and needs to link customers to shoppers' insights, advanced analysis and front-line team experience as an ecology.

The competencies listed in the articles include product foresight, shopkeeper research, store clustering, product optimization, spatial planning, shop intelligence and product growth insight. Technically, reference is made to s-DNA shop portraits, sales team automation tools, PepsiConnett for shop owners, shelf image recognition tool Pep IRIS, and the AI capabilities being invested. The article also gives a case direction to the British market: as demand for protein products grows, the team works with clients to build Protein Bay, turning the vision of the class into a clearer shop experience. The article also refers to exchanges between Pepsi and industry institutions such as IGD. The above-mentioned sources have been disclosed and no additional validation of their effects is provided here.

02 GSP analysis

Change the "perfect door" from a uniform template to a business-based combination of capabilities

The most noteworthy operational judgement of this disclosure is not a single tool, but an adjustment to the way the term “perfect store” is defined. Traditional practices tend to make perfect shops a standard set of displays to be issued by Headquarters, where they are executed, inspected and rated, and eventually turned into compliance competitions. The article repeatedly emphasizes that “it is not an ideal model that applies everywhere”, but rather answers “what makes this retail environment more effective for shoppers”. This effectively changed the delivery from a standard atlas to a decision-making package tailored to the type of store, consumption scene and channel, the standard being only one output. The real difficulty is that the standard drawings can be produced centrally and distributed at once, while the decision-making packages require branders to redistribute judgement between headquarters and the region: which types of roles will be harmonized at Headquarters and which displays will be given to a front-line team familiar with the local flow.

Example scenario: a chain of convenience stores introducing the same set of product recommendations at the same time as a large mall, which may require compressing of single items around breakfast, commuters and nighttime scenes and increasing the visibility of goods that are taken or taken away; Large malls may be more likely to be home-based and weekend hoarding logical group blocks. If branders prepare only one set of options, they will be treated as headquarters missions by the business teams on both sides. Conversely, the same set of insighting capabilities can produce different movements if the industry role is defined first and the product width, location and promotional rhythm are determined. The trade-off here is that customization increases relevance, but increases programme design and communication costs, so that marginal costs must be controlled by a conglomerates and reusable modules. In a different light, the real risk of customization is not a cost increase, but rather an out-of-control standard — one for each region, which does not allow for horizontal comparisons or the deposition of effective practices back to Headquarters.

The qualification is whether the differences between the stores are sufficient to change purchasing decisions. If the tasks of buyers in the same industry are highly similar, over-customization creates executory noise; If the channel is fragmented significantly, the uniform template will fail faster than the cost of customization. Branders first have to answer “the level of difference” before deciding whether to customize to the city, the business world or the single store. A workable test is the selection of two types of outlets for small-scale comparisons, one for the implementation of uniform distribution programmes and one for industry role-adjusted comparisons of marketing and stock-outs over the same cycle; If differences are not significant, custom-made privileges should be withdrawn and processes simplified, rather than the programme version should continue to be folded.

03 GSP analysis

Values come from a closed circle of insight to execution, not a single-point tool.

The article quotes Paul Nash, the UK ' s leader in the class, who said that the real value lies in connecting different capacities from building a shelf to long-term growth in the common support category. The sentence points to the commercial logic of such systems: single-sale data reporting, single-sale shelf identification, single-sale display services are prone to competition; Insight, planning, execution and measurement are closed, and the client buys a traceable business result, not a single document or a single photograph. For branders, this means a shift in the income structure from one-time project fees to ongoing services, as well as assuming responsibility for directional judgement, not just delivery.

The key to a closed ring is that each ring has a clear input output. Buyers study the output of the demand scenario, the product cluster output of the group rules, the optimized output of the product is a single item and space trade-off, the image recognition output is an execution deviation, the sales team automates the output of the task list, and the measurement chain returns the results to the assumptions. Any missing link would degenerate into either “pretty reporting, no change in shelf” or “execution is hard and no one can verify the direction”. Example of scenario: a regional beverage class found that coffee was growing in the traffic hub, but image recognition showed that the freezer gold layer was occupied by high-sugar beverages; If there is no planning link to turn insight into a display rule, the results of identification can only be a list of issues; Nor can the rules prove effective without measuring the impact of the adjustment. There is also a link that is often overlooked: who has the right to change the shelf. If the planned output is not accompanied by the minimal actions and duty bearers that can be performed by the door, the closed ring will still break at the last metre.

Therefore, the assessment of such a system should not only ask how advanced the tools are, but also whether the interface between the various links is responsible, whether there is a uniform sign and type of product language, and whether the “supposition-action-verification” can be completed within a cycle. Connectivity is in itself an infrastructure that needs to be built, not a by-product of nature. A specific test would be to take a sample of the individual shop case: from the initial insight hypothesis, to ask around who was responsible for each step, what records had been left, and whether the changes in the final sales could be traced back to the original assumption; Any intermediate or non-retroactive link is the point of connection that needs to be prioritized for the next round.

04 GSP analysis

The changes in the road for shoppers require the shop to re-assemble its product according to the scene.

The article notes that products suitable for large supermarkets may meet different needs in convenience stores, cafés or travel kiosks; Buyers move between more channels, scenes and patterns, and behaviour changes fast enough to require a more fine view of the particle size and a faster response. This is a direct implication for the operations of the store: the basic unit of goods management is shifting from a “shelter” to a “shopping mission in the scene”. In switching from business to business, buyers bring with them a set of mission constraints — time budget, number of peers, carrying conditions, follow-on journeys, while the shelf structure answers only “what brand to buy” and does not answer “Why to buy at this moment”. Therefore, the starting point for the shop restructuring class is not to move the front line but to identify the type of tasks that the shop has high frequency and then to determine which should be mission-oriented and which should still be organized according to brands and specifications.

Example of hypothesis: A shopper buys breakfast on commutes, focusing on speed of pick-up, combinations and the visibility of cold cabinets; Home purchases on weekends are concerned with specifications, taste selection and replenishment capacity. The role of the class is difficult to perceive if the shop continues to be displayed on the supplier ' s branded block and the shopkeeper needs to consolidate the solution on his own. The Protein Bay case, to which Pepsi referred, essentially aims to raise the demand for proteins from a single distributed product, to form an identifiable landscape block, to reduce the cost of searching, and to bring about a greater concentration of material education. It is important to note that the benefits of the site block are not merely a single item within the block, but should be seen in the proximity of the neighbouring group and the overall guest list, otherwise it would be easy to concentrate the previously dispersed sales in one place and draw a false and successful conclusion.

The scene shows the boundaries. Site blocks can take up space, increase the number of porter renovations and may conflict with established branding rules. Conditions are determined as to whether the demand for the scene is sufficiently concentrated, whether it can be easily maintained by the shop staff, and whether it can lead to neighbouring groups rather than simply internal transfers. Branders should define, in conjunction with retailers, the entry and exit criteria for the scene, for example by setting a trial period for time slots, door-to-door types or sales contributions, so as to avoid a reversal of a transformation into a long-term burden. More specifically, the pre-reform recording of the area ' s time distribution and stock shortages, and the post-reform comparison of the net increase observed by the non-renovation store; If the increase is mainly due to the replacement of a single item in the same category and the maintenance work hours increase significantly, then the amount due is either seasonal and temporal.

05 GSP analysis

One-line team moving from executor to intuitive interpreter and giver.

The article states explicitly that AI helps teams to assess more information, identify more detailed opportunities, anticipate changes in the category of goods, but human skills such as judgement, communication and relationship building are therefore more important; The tool reduces the time available to organize technical material and frees up space for people to explain the meaning of information, to explain it to clients and to make common choices. This judgement has a substantial impact on the organizational division of labour: first-line sales are no longer just about visiting missions and displays, but rather about the responsibility of translating the results of the analysis into the market language — an explanation of why the store should be restructured, what it is expected to see, and who it should look for in the event of an anomaly.

Example of scenario: image recognition reveals a high rate of stock shortage in a shop shelf and the system can generate replenishment tasks; However, the reason for the lack of goods was logistics, back-silo management and over-presentation, which required on-the-spot judgement on the part of frontline personnel. If the organization only evaluates photo coverage and mission completion rates, personnel tend to close the task quickly, rather than addressing root causes. A more rational division of labour would be: the analytical team would be responsible for modelling and grouping rules, the client team would be responsible for aligning product-class strategies with retailers, the front line team would be responsible for door-to-door validation and exceptional reporting, and the three-party would be responsible for product-type results rather than for process indicators. The implementation could be done with a “justification of exceptions” field and a periodic review to convert the HF exception into a revised rule or input for grouping adjustments.

The key trade-off here is standardization and autonomy. The more standardized the task, the easier it is to scale; But the more exceptions, the greater the need for a line of discretion. Branders should identify actions that need to be harmonized and which can be adjusted by the first line within the delegation of authority, and establish an exceptional feedback channel to allow the flow of shop knowledge back into group rules and programme design, otherwise first-line experience remains at the individual level and cannot be transformed into organizational capacity. To test the effectiveness of the mechanism, two things can be seen: whether the first line of proposed exceptions are classified and trigger the updating of the rules, and whether newcomers can use their records to quickly re-enter their old employees; If neither is possible, it means that empowerment remains a slogan.

06 GSP analysis

Inputs should be based on data base and group rules, followed by an AI expansion

The article mentioned that s DNA used information such as sales, shop lists, local interest points, demographics and passenger flows to build shop paintings, and that AI input was used to assess more information and identify more detailed opportunities. An easy to ignore business issue is the sequencing of these capabilities. If the signage is not uniform, the list of outlets is not adequate, sales data cannot be gathered by the door, and the advanced model can only produce seemingly fine conclusions on unreliable bottoms, and the wrong subgroups will pass on to distribution, displays and appraisals.

Thus, a more secure sequence of inputs would be to harmonize shop owners ' data and class languages before establishing interpretable rules for shop grouping and then to select a few high-value scenarios for planning and implementation on a pilot basis before expanding the scope of AI application. Example of a scenario: a brander first piloted a breakfast scene in two types of shops, a traffic hub and a community convenience store, grouping existing sales and passenger data, and manually validating the scheme; Automation of image recognition and task access is followed by confirmation that sales and stock-outs are improved and traceable. In this way, each technical input is matched by a clear operational problem rather than a tool to be bought for use. The clustering rules themselves should also be maintained in order to allow the client team to explain to the retailer why a particular store was placed in a particular group, otherwise joint business plans would be difficult to advance.

The alternative explanation for the need for vigilance is that growth may come from promotional, market expansion or seasonal factors, rather than from insight into the system itself. The pilot design could easily be attributed to new instruments if there was no matching shop, no baseline period and no distinction between natural growth. Investing in decision-making should be based on the premise of “demonstration of causality” as an extension, rather than on “tools on the line”. The pragmatic approach is to answer three questions before each expansion round: whether operational issues are still in place, whether the data base meets the threshold of availability, and whether the previous round of increases can withstand the matching test; Either option should be discontinued during the pilot phase instead of continuing with the supplementary budget.

07 Suggested actions

Order of implementation: six-step advance from class vision to door-to-door validation

As a first step, a retailer and a class of goods are selected as common subjects for testing, with clear targets for the growth of each counterpart and class of goods, avoiding the decentralization of responsibilities by opening up multiple groups of goods simultaneously. The trade-off between this step is that the smaller the scope of the pilot, the easier it is to control the variable, but also the greater reliance on individual counterpart input; Therefore, the subjects should choose the categories of products where there is a marked difference in the store, where retailers have a common will, rather than the most beautiful ones. The second step is to use the available data to complete the shop grouping, to distinguish between business circles, trade patterns and shopping tasks, and to produce interpretable group rules rather than black-box labels. The rules need to allow first-line personnel to repeat them in a few words, for example, “the traffic hub is dominated by buying or leaving”, otherwise the subsequent door-to-door programme cannot be understood and adapted.

The third step is to work with retailers to develop visions and role definitions for product categories and to identify scenarios that need to be strengthened and monolithic products that require trade-offs. The trade-off here is that the more complete the vision, the better the orientation of the door shop programme, the higher the cost of consensus; The vision should be reduced to two or three proven scenarios, rather than listing all opportunities. The fourth step is to transform the vision into a door shop-level programme that includes spatial planning, displaying rules, promotional rhythms and replenishment requirements, and to indicate which actions are uniform and which allow the door to adjust. The marking of the alignment of competencies is a key design that determines whether the programme is a template or an implementable framework.

Step five, implemented by a first-line team at the pilot shop, implements deviations through image recognition or manual verification, processes exceptions on a weekly basis and records reasons. Example of the scenario: A pilot shop should have a breakfast set-up set up according to the scheme, but the owner changed to a side rack due to the wiring restrictions, with the exception of the log record, which reads “Strict entry, stand-off blocking access”, which is more valuable than a simple determination of “non-compliance”. The sixth step is to set a fixed cycle double disk, to compare the difference between the pilot shop and the control shop in terms of stock-outs, visibility, sales and feedback from buyers, and to decide to expand, adjust or stop. The duty bearers of the six steps should be clear: the quality team is responsible for vision and trade-offs, the analysis team is responsible for clustering and measurement, the client team is accountable for retailer alignment, and the front-line team is responsible for door-to-door certification. Each step of the output should be transferable, e.g., cluster table, scenario card, exception log and review conclusions; Without these transfers, the project would have been degraded into meetings and reporting.

08 Suggested actions

A small number of first-line indicators and a comparison of experiments to determine whether they are worth expanding

It is not appropriate to measure such a system by looking only at sales. It is proposed to set three tiers of indicators: programme coverage, stock-outs and display compliance at the executive level; (b) The search time, visibility and transformation of purchases of the target scene goods at the shopkeeper level; The operational level is based on product-based marketing, customer order structures and repurchases. The three tiers of indicators should be tied to the same pilot shops and the same window of time, so as not to allow teams to speak separately. The selection of the first line of indicators should be subject to trade-offs: coverage can quickly reflect whether the programme is located, but it can be easily induced to “cover for coverage”; Looking for time and visibility is closer to those experienced by shoppers and more costly to collect, so it is appropriate to use a small number of prior-stage indicators for early judgement and then to use business indicators for confirmation.

Experimental design, using a pilot shop versus a matching shop, with matching dimensions including industry, business circles and baseline sales; For scenarios where contrasts cannot be set, at least the baseline period is maintained and disruptive factors such as promotion and warehousing are recorded for the same period. Example of the scenario: a region selects a pilot breakfast set of twenty stores and a similar set of twenty for eight weeks of continuous observation, while recording external factors such as promotion slots and weather; The difference cannot be fully attributed to the programme if the pilot shop improves visibility and target class marketing, but the control group also improves marketing. The test may be based on whether the prior indicator has been synchronized within the same week and whether the business indicator has been followed up late, with priority given to skepticism of the implementation bias rather than the programme itself.

Indicators should also include the cost side: programme design hours, door-to-door adaptation inputs, and additional hours of visits on the first line. The expansion has operational significance only if the improvements are sufficient to cover these costs. Quarterly stop judgement: if there is no significant improvement in two consecutive cycles, then it should go back to the goods-based vision or cluster rules to find reasons instead of continuing to add tools. The trade-off lies in the fact that the speed of expansion often conflicts with the intensity of evidence; It is suggested that small-scale, short-cycle control experiments be used to build up evidence and to decide whether to increase inputs, rather than to mask the inadequacy of measurements with a larger number of outlets.

09 Suggested actions

Applicable conditions and adjustment signal: under what circumstances should it be done and under what circumstances should it be stopped

Such systems are better suited to scenes where channels are fragmented, shop differences affect purchasing decisions, retailers are willing to share data and share responsibility for goods. If retailers are highly concentrated, decision-making power in the category of goods is limited at headquarters and there is little room for implementation in the shop, the value of the customized door shop programme will decline, and the emphasis should shift to headquarters-level strategy and supply chain synergies. If branders themselves have weak owner data and high mobility of front-line teams, trajectories can magnify implementation bias. In determining applicability, three questions can be asked: whether the differences in shopping tasks between the shops are sufficient to change display decisions, whether retailers are willing to share the risks in the selection of goods, and whether there is time to judge on the first line rather than simply to make cards.

The adjustment signals included: unexplainable door group rules, persistent high exceptions, frequent turnover of retailers, incomparability of pilot improvements and lack of judgemental space for first-line visits to be filled with tasks. When these signals occur, the range should be reduced and returned to a category of goods, type of door, for re-certification instead of adding more AI functionality or expanding the number of doors. Assumptions: If the “door adjusts itself” share of the exception logs continues to rise, this may indicate that the uniform action is too well designed, when the door is relaxed and the scheme card is simplified, rather than strengthening the auditing penalty.

The final criterion is not whether the system is “advanced”, but whether it will make retailers more willing to share the trade-offs, give frontline personnel a better capacity to explain and adjust, and make it easier for shoppers to perform purchase tasks in specific shops. Only when these three things happen at the same time can insighting the input translate into door-to-door operations. The terms of the suspension should also be agreed in advance, for example, if two consecutive round-trip cycles fail to distinguish between the pilot and the contrast, or where cost side inputs continue to exceed the benefits of improvement, the expansion should be suspended and the vision and cluster rules re-suppositioned.