CPG Analytics Is Fifteen Questions, Each With a Written Rule

The decisions that set a specialty food brand's operating margin come down to about fifteen questions, and each one can be answered by a rule written down once and run on current data. Most of them are answered from memory instead, because the analysis behind each one is built from scratch every time it is asked.

Cinderhaven Provisions, a fictional specialty food company whose numbers form a synthetic dataset, invoices about $25M a year to its trade partners. Its founder sits in the Monday operating meeting with three tabs open: a margin spreadsheet updated last Thursday, a UNFI deduction report not yet reconciled, and a Retail Link extract of Walmart scan data. The CFO asks whether the company can afford another retailer launch. The sales lead says the club retailer promotions "felt strong": there were twenty-two of them in 2023 and 2024, at a promotion cost of $67,576, and nobody in the room has a number for what they returned. The ops lead reports that 84% of Walmart shipments arrived on time and complete in 2025 but cannot say which of the misses triggered chargebacks and which the retailer absorbed. Each decision the meeting makes rests on a number somebody remembered rather than calculated.

That is not a data literacy problem. It is an infrastructure problem. The questions are finite, they recur, and they have answers; what is missing is a standing way to produce them.

The question list is shorter than it looks

Specialty food CEOs ask variations on the same set. Should I drop my biggest customer, the one whose deduction tail consumes half the account's margin? Can I afford this retailer launch, or will slotting and free fills burn more cash than the first year returns? Which SKUs should go? Where is the trade spend going? What would a recall cost? Will my product data break at Walmart? Which channel makes money after trade, deductions, and float?

These are not open-ended explorations. They are binary or ranked verdicts: yes or no, this one or that one, above threshold or below. A $25M brand does not need a data science team to model them. It needs the rules written down, which is the process hygiene, and the data piped in clean, which is the data hygiene. Without either half, the answer is a guess.

Gartner's 2024 Chief Data and Analytics Officer Agenda Survey names low data literacy and resistance to change as top inhibitors of data and analytics success, and finds that poor data literacy pushes executives toward decisions made on gut feel. At a specialty food brand between $10M and $50M, the gut-feel decision has a plainer cause. The questions were never formalized, so every answer needs a fresh analytical build, and the build waits behind everything else on the calendar.

Rules replace exploration once the questions are known

The conventional route in CPG is the BI project: scope the question, pull the data, build a dashboard, present findings, shelve the dashboard, and start again in six months when the question comes back. Many business problems are still solved that way, through traditional approaches that take months or years; McKinsey's picture of the data-driven enterprise of 2025 has them resolved in hours, days, or weeks instead of through lengthy, sometimes multiyear, road maps. A $25M food brand cannot absorb the slow version fifteen times over.

The questions are stable, so the rules that answer them can be written down. "Should I drop my biggest customer?" resolves to: take the account's gross revenue, subtract the trade spend commitment, subtract trailing-twelve-month deductions by category, subtract compliance chargebacks, apply the timing cost of the cash conversion cycle, and compare the resulting contribution with what the same capital would earn in the next-best account. That is arithmetic with documented thresholds, not discovery.

The same structure holds across the list. "Can I afford this launch?" is a cash-flow model with inputs from trade spend commitments, slotting terms, projected velocity, and current working capital. "Which SKUs should go?" is a portfolio sort on contribution margin per facing after trade, the same SKU rationalization logic category managers at Kroger and Albertsons already apply to their own shelves. Once the method is encoded, it runs on this week's data instead of being rebuilt.

What a question engine returns

I built the engine against Cinderhaven Provisions, so every rule could be checked end to end before it touched a real brand's books. A CEO picks a question from the list. The engine runs documented rules against the current dataset: not a model, not a prompt. It returns four things: a one-sentence verdict, one chart that shows the shape of the answer, three numbers that anchor the verdict, and a link to the full worked analysis.

"Should I drop my biggest customer?" returns, for one Cinderhaven account: "No, but renegotiate. Over the trailing 12 months the account produces $412K in contribution after a $187K deduction tail, which ranks third. The deduction rate is 2.1x the portfolio average, and 62% of it traces to shortage claims disputable with better ASN data." The chart shows contribution by account with the deductions overlaid. The three numbers: $412K contribution, $187K deductions, 62% traceable to data errors.

A CEO reads that verdict in thirty seconds, and a CFO who wants to audit the logic can read the rules that produced it. The alternative, the BI dashboard, asks the executive to form the question, navigate the interface, and interpret the chart. Where the promotion plan, the scan data, and the deductions sit in three systems with three owners, a dashboard that leaves the user to join them in their head moves the analysis onto the person with the least time for it.

Fifteen questions map to the decisions that move margin

The questions follow the operating decisions:

Customer portfolio: which accounts earn their capital once the deduction tail and compliance cost are netted, and which look profitable only before that. Channel allocation: which channels return the most per dollar deployed after trade and float, a question the data rarely answers directly because it sits in four systems that are not joined at the channel level. SKU economics: which products earn shelf space and which ones the rest of the portfolio subsidizes. Trade effectiveness: where the 15-20% of gross revenue committed to trade lands, and what it produces. Launch viability: whether a new retailer authorization generates net cash or consumes it. Forecast accuracy: whether the demand plan is tight enough to hit OTIF thresholds without building excess inventory. And underneath, the operational questions (recall exposure, data readiness, cash conversion) that decide whether the brand can execute the strategy the CEO just approved.

Thirteen of the fifteen return a verdict from the engine, run against the Cinderhaven dataset (Cinderhaven Provisions: 50 SKUs, six retailers, ten channels), with every rule documented, every threshold sourced, and every verdict auditable. The other two, distributor reconciliation and recall cost, hand the question to their standalone tools. The full list, each item phrased as the question a CEO would ask and paired with its rule, is published in the tool. A branded food company's margin is governed by a finite set of decisions, and a finite set of decisions can be answered by rules.

The place to start is the question last answered from memory: if Monday's meeting called the deduction line "high" or a promotion "strong" without a number, that question already has a rule waiting to be written.

The operating question your Monday meeting cannot answer yet

Tell me which question your Monday meeting cannot answer, and I'll write back with the rule that answers it. No call.