How CPG brands use causal AI to plan trade and retail-media investment across Amazon, Walmart Connect, and Kroger in one model, not three.
Most CPG brands I work with plan retail media one retailer at a time. The Amazon Ads team builds one plan, the Walmart Connect team builds another, and Kroger Precision Marketing gets what's left, each judged on its own platform's numbers. The money at stake keeps growing: of the growth in US retail media spend in 2026, 89% will go to Amazon and Walmart, according to EMARKETER's January 2026 retail media forecast.
Most leadership teams read that as a question of which retailer to back. The real problem is that brands plan trade and retail media spend for each retailer in a separate model, so no one sees how a dollar at one changes sales at the others. Below is the workflow I use to plan all three at once, and how to scale it across a portfolio.
Most of the AI work in CPG today falls into three jobs: forecasting demand, producing content and consumer research faster, and deciding where commercial investment goes. The first two save time and cost. The third moves growth, because it changes where trade and media dollars go across retailers.
Only 39% of organizations say AI has added anything to their operating profit, according to McKinsey's November 2025 State of AI survey.
The cross-retailer planning workflow takes weekly spend and sales for each retailer, with trade, price, and distribution alongside, and runs them through one causal model. It produces a single plan that moves budget across Amazon Ads, Walmart Connect, and Kroger Precision Marketing at once.
The workflow has four moves: look past each retailer's own reporting, model all three retailers together, shift budget in one plan, and count the halo that crosses retailer lines.
Each retail media network is a walled garden that measures its own campaigns with its own closed-loop data. Amazon Ads reports the sales its ads touched on Amazon, Walmart Connect does the same at Walmart, and Kroger Precision Marketing reports on Kroger shoppers. None of them sees the others, so their claimed sales overlap and add up to more than the brand actually gained.
Nielsen's 2025 Annual Marketing Report found that 85% of marketers are confident they can measure ROI across traditional and digital channels together, yet only 32% actually do.
The fix is one model that measures every retailer on the same basis, so a brand can maximize total marketing ROI across every retailer. Planning needs that same model, because three separate reports cannot tell you where to move money between retailers.
Put Amazon, Walmart, and Kroger in the same causal model, and a brand can judge every dollar by the total sales it causes, net of price, promotion, distribution, and seasonality. That single view lets a brand compare the three on the same terms and move money between them in one plan.
A marketing mix model has guided media budgets for decades, but it struggles here for a structural reason. Brands raise spend at all three retailers in the same promotional windows, so the three spend lines rise and fall together, and a correlation-based regression cannot tell which one caused the sales. Statisticians call this multicollinearity; in practice, the model divides credit between retailers by past spending habits.
A causal model separates those effects by estimating what would have happened without each investment. It returns a response curve for each retailer: the incremental revenue the next dollar produces, and the spend level where that return starts to fall. To see why correlation breaks down and how causal models fix it, read our comparison of causal AI vs traditional analytics or our breakdown of POEM365 vs MMM vs MTA.
The figures below are illustrative, not drawn from a DATA POEM client: a $50M annual budget for retail media and trade promotion across three retailers. The table shows spend before and after one planning cycle, and the return on the next dollar at each point.
Investment | Before ($M) | Before (%) | Return on next $1 (before) | Change ($M) | After ($M) | After (%) | Return on next $1 (after) |
|---|---|---|---|---|---|---|---|
Amazon Ads | 20 | 40% | $0.75 | -4 | 16 | 32% | $1.00 |
Walmart Connect | 14 | 28% | $1.20 | +2 | 16 | 32% | $1.05 |
Kroger Precision Marketing | 6 | 12% | $1.55 | +3 | 9 | 18% | $1.05 |
Trade promotion, all three retailers | 10 | 20% | $0.90 | -1 | 9 | 18% | $1.00 |
Total | 50 | 100% | 0 | 50 | 100% |
The plan moves $5M, or 10% of the budget: $4M out of Amazon Ads and $1M out of trade promotion, where the next dollar returned $0.75 and $0.90. It goes to Kroger Precision Marketing (+$3M) and Walmart Connect (+$2M) until every next dollar returns $1.00 to $1.05.
In this illustration, averaging each line's return across the dollars moved, the same $50M produces $1.7M more incremental revenue. Three separate plans would miss that shift, because each retailer team defends its own budget.
A cross-retailer halo is the extra sales a brand gets at one retailer because it advertised at another. A shopper who sees a brand's ad on Amazon may buy it at Kroger that weekend, and a Walmart Connect campaign can lift the same brand on Kroger's shelf.
Each retailer's closed-loop data credits only its own shelf, so the halo is invisible to each retailer's own reporting. It appears only in a model that holds all three retailers and offline sales together, which is why the model counts it before recommending a cut at any one retailer.
The gap is scope: a marketing-mix tool models media alone, while AI built for the full commercial picture models media, trade promotion, pricing, and distribution together. The difference matters because in CPG the largest levers often sit outside media: trade spend, shelf price, and the stores that carry the product.
This is the ground DATA POEM’s Enterprise Decision AI covers: one causal AI model for the decisions that marketing, sales, and finance each make separately today. Read how Enterprise Decision AI works in practice across those functions.
For a CPG brand, those decisions run on three planning cycles. Strategic decisions, made each year or quarter, cover portfolio growth planning, new product launches, and where to invest across brands and markets. Functional decisions, made each quarter, cover unified marketing planning, demand forecasting, and how a promotion on one brand lifts or takes sales from another. Tactical decisions, made each week or month, cover moving media money mid-campaign, measuring the sales retail media and search spend actually adds, and demand planning from 4 to 52 weeks out. The cross-retailer plan above sits in the functional and tactical cycles, and the same model carries it up to the portfolio.
Inside POEM365, DATA POEM’s Enterprise Decision AI platform, four Decision Agents each handle one part of cross-retailer allocation: Insights, Forecasting, Planning, and Optimization. Together, they take a brand from understanding what caused last quarter’s sales at each retailer to accurately forecasting where next quarter’s budget should go.
POEM365 is our Large Causal Model (LCM), built on FOUNT, DATA POEM's Large Causal Architecture (LCA), and it delivers Enterprise Decision AI. Each agent has one job in allocation:
This is what agentic AI for CPG should mean in practice: agents that recommend allocation from a causal model, with every recommendation traceable to its drivers.
For most CPG brands, the best place for a first AI project is the decision with the most money and the least proof behind it. That decision is how to split trade and retail media spend between the top three retailers. Once one budget shift there has proved itself, bring pricing, distribution, and the rest of the portfolio into the model.
The starting sequence runs in five steps, from one brand to a tested reallocation.
Planning maturity shows up in the numbers. BCG's February 2026 supply chain planning study found forecast accuracy at the highest-maturity planners is more than 25 percentage points higher than at the lowest.
You don't need to build a new model for every brand in the portfolio: fine-tune one pre-trained causal model on each brand's data instead. Each brand gets its own Enterprise Growth Model, and the portfolio shares one causal foundation, so every brand's results are read on the same terms.
Separate brand models also miss portfolio effects. A price promotion on one brand can pull sales from a sister brand on the same Kroger shelf, and only a model that holds both brands sees the net result.
Add brands in order of retail overlap: brands that share the same three retailers go first, because shared shelf effects are largest there.
CPG companies use AI for consumer insights to summarize reviews, social conversation, and survey responses at scale, and to group shoppers by what they buy and where. The insight becomes commercial when the brand tests it against sales and sees which consumer shift actually moved revenue at each retailer.
AI tools helpful for CPG analytics connect retailer, trade, and sales data in one place and explain the cause behind each recommendation. Test a shortlist on three points: which decisions it supports, whether finance can audit its assumptions, and how it handles data from Amazon, Walmart, and Kroger.
Retail media ROI is measured alongside other marketing channels by putting every channel, on and off retailer sites, into one causal model that estimates incremental sales on the same basis. Validate that model with a holdout test, pausing spend in a few regions at one retailer, to confirm its read before you move budget.
The key trends in AI for retail and CPG are agentic AI that acts within set limits, retailer data shared through clean rooms, and a shift from correlation-based models to causal ones. Across AI in the CPG industry, the common thread is accountability: finance teams want to see why an AI recommendation should move money.
The payback comes from planning retail media, trade, pricing, and distribution at Amazon, Walmart, and Kroger together. The brands that plan every retailer in one model see where each dollar works and can move it on evidence.
Growth Planning shows where the next dollar should go across every retailer, brand, and commercial lever in the plan.
POEM365, from DATA POEM, puts every retailer, brand, and lever in one causal model, so CPG leadership teams can plan the whole business from a single read.
Start with one brand and its three biggest retailers, and talk to the DATA POEM team about running a cross-retailer plan on your data.
Head of Growth Marketing
Let's talk about how we can help you grow your business.
A CMO's playbook for reallocating an existing marketing budget by causal contribution, defending it to the CFO, and holding it in a downturn.
See how Fortune 500 retailers use AI for demand forecasting, personalization, and supply chain, and where most retail AI investment still falls short of the P&L.
A comparison of the top marketing attribution tools 2026, discussing attribution method, incrementality support, best-for use case and pricing signal.
Unified marketing measurement usually means combining MMM, attribution, and incrementality. But that's still fragmentation. Here's why we say so, and what consolidating everything into one causal model changes.