Building a CEO/CFO-Ready Reporting Infrastructure

Finance closes on statutory boundaries. Commercial measures on the boundaries of a buying relationship. Supply chain measures on physical ones. Each frame is accurate, and accurate because it is shaped by what that function has to defend.

Executive questions cross all of them. Answering one means deciding which frame governs at each crossing point: which calendar, which hierarchy, which unit. That is not analysis and not modelling. It is an allocation of who accepts the distortion, and it belongs to no function — which is why it does not get made.

What actually happened

Polymath’s founder ran corporate strategy and global growth analytics for a private consumer packaged goods business of roughly $2–3 billion in revenue that had gone public on the TSX, across all three of its segments, and owned the business health reporting that went to the board.

The three segments do not share economics. The first segment is business-to-business: the company ships to large retailers, and what matters is shipments, sell-through at the shelf, returns and trade spend, on a retail calendar. The second is a licensing and content business, partly business-to-business and partly direct: rights sold into distribution windows, licensing revenue recognized against delivery and minimum guarantees, audience counted in consumption rather than units. The third is direct-to-consumer and digital: daily, in-app, with bookings that are not revenue.

So revenue meant three things. A customer was a retail buying organization, a distribution partner, or an individual consumer. Growth was a shelf position, a renewal, or a retention curve. Each definition was correct inside its segment and each was defended by people who were right.

Category work sharpened it further. Planning a category with a major mass retailer means working in that retailer’s frame — their category definition, their fiscal calendar, their point-of-sale week. It is the only frame in which a shelf decision can be made. Internally the same products also lived in a brand hierarchy built for creative and licensing decisions, and in a revenue ledger built for legal entities and a fiscal month.

Then the board asks one question across all of it: where is growth coming from, and will it hold. After the pandemic that question had teeth — inventory had to be worked down in one segment while another was scaling, and the five-year revenue, margin and portfolio roadmap depended on which answer you believed.

The team unified the data, rebuilt the BI layer and automated demand planning. Those were the right things to do and they did not settle the question. What settled it was duller: for each recurring executive question, deciding which hierarchy and calendar governed, writing it down, and putting the difference to the segment’s own figure beside the enterprise number.

Consolidating the data made all three answers available at once. Availability is not decidability.

The first thing Polymath looks for in a client’s board pack is whether anyone has written down which calendar and which hierarchy governs, because until that exists the conversion is being made silently by whoever assembles the page.

Why it happens

A function’s reporting frame is not a preference. It is set by what that function is accountable for, and it is correct inside that accountability.

Finance draws its boundaries where the law does: legal entity, period, recognition rule, a policy rate held constant so comparison means something. Those boundaries have to survive an auditor. Commercial draws them around the buying relationship — the customer as a negotiating entity rather than a bill-to address — because that is what a sales force is paid on. Supply chain draws them where the goods physically are: the site, the shipped unit, the replenishment week. Workforce cost is drawn on payroll calendars that, across roughly twenty-one countries, land on the fiscal close in almost no month.

Every one of those frames is internally consistent. None converts into another without a loss, and the loss is not error. It is the price of accountability.

The frames do not disagree because someone is wrong. They disagree because each is right about a different thing.

Now put a cross-functional question to them. What happened to margin in this category, and will it repeat? Answering it means choosing a customer hierarchy, a product hierarchy, a calendar, a currency treatment and a unit. Five crossing points, and at each one a frame gives way. Choosing is not an analytical act. It determines which function’s number will sit next to the enterprise number looking wrong, and which function spends the quarter explaining the difference. Nobody volunteers, and nobody can be overruled on the merits, because each is right in its own terms. So the choice is not made. It is absorbed.

It gets absorbed by whoever assembles the pack, usually the most junior person in the chain, working to a deadline and making the conversion silently — and, never having written it down, slightly differently each cycle. The executive team receives a figure with an undocumented conversion inside it, unable to tell whether a movement is performance or a mapping change.

The strongest objection

The strongest objection is practical: we have a warehouse, a semantic layer and a BI team, and they resolve this.

Partly true, and worth conceding properly. A well-built semantic layer kills the trivial disputes — two teams filtering differently, or one working from a stale extract. That is real value, and most enterprises have not captured it.

But a pipeline can carry a conversion rule; it cannot choose one. Deciding that the enterprise margin view runs on the retailer’s calendar, and that finance’s month therefore reconciles to it rather than the reverse, determines who spends the quarter explaining a difference. No engineer can take that decision, and none should be asked to.

The better objection is the mature one: conformed dimensions and master data management are precisely this, and we already run that programme. Agreed, and conformance is the right instrument. The difficulty is that it is performed once, at design, while the frames keep moving — a new retailer hierarchy, a revised recognition policy, an acquisition with its own product tree. Without a standing owner and a trigger to re-approve, the mapping table survives while the agreement behind it quietly expires.

The method

We run six steps. The order matters, and the first does not look like data work.

1. We start from the questions, not the measures. We write down the ten questions your board and executive team actually ask, in the words they use, with verbs in them. Not a KPI inventory. The trade-off we accept: ten is too few and important measures will sit outside the list — correctly, because a complete list produces a reporting estate rather than an executive layer. Skip it and you govern whatever was easiest to certify.
2. We name the crossing points in each question. We list where the question passes between frames: customer hierarchy, product hierarchy, calendar, currency, entity, unit of measure. Most have four or five. The trade-off we accept: it is half a day of unglamorous work that produces no output anyone can present. Skip it and the conversions stay invisible, and an invisible conversion cannot be governed or corrected.
3. We decide which frame governs each crossing point, and we name the function that accepts the distortion. Out loud, in a room with the executives who own the frames. We do not ask which frame is right; we ask which one the enterprise answer uses, and who therefore carries a permanent difference. We settle the smallest set of definitions that will carry the ten questions, and we leave each segment’s economics alone elsewhere. The trade-off we accept: an unpleasant meeting that produces a loser at every crossing point. Skip it and the choice defaults to whoever builds the slide, which is where it sits today.
4. We put the bridge on the page, not in an appendix. Every enterprise figure we publish carries the reconciling difference to the functional figure it will be compared against, stated beside it: this is shipments on our month, that is consumption on theirs, here is the gap. The trade-off we accept: the pack gets denser and the first few are argued over. Skip it and every meeting re-derives the difference, and executives learn to distrust a number rather than read it.
5. We give the rules a trigger, not a cadence. A quarterly review will not catch the change that broke the rule. We bind re-approval to events: a new retailer hierarchy, a territory redraw, a recognition policy change, an entity restructure, an acquisition. Whoever makes the change raises the flag. The trade-off we accept: it obliges functions never consulted before. Skip it and the rules drift out of agreement while still looking maintained, which is worse than having none.
6. We stop functions answering enterprise questions in their own frame. Functional packs keep their frames, and should, because those frames run the function. What we change is that the cross-functional question is answered once, from the governed conversion, and the functional versions of it are withdrawn. The trade-off we accept: functions lose a narrative device they like. Skip it and the executive layer becomes a seventh opinion rather than the deciding one.

None of that requires a platform decision and none of it is expensive. What we need is one executive with authority to make several functions accept a conversion they did not choose, and the discipline to write down what was decided. That written set is the architecture. Everything your Record-to-Report cycle later produces is assembled from it, which is why we build it once rather than re-derive it every close.

Evidence and measures

The exhibit we build first is not a data model. It is this:

Five measures worth tracking over ninety days. None requires a new system:

— For each of the top ten executive questions, whether a written rule names the governing frame at every crossing point. The opening count is usually zero.

— Figures in the current board pack carrying a stated bridge to the functional number they will be compared against.

— Customer, product and calendar hierarchies in active reporting use. Count them before rationalizing anything; the count is the argument.

— Days from a cross-functional executive question to an answer both functions will sign, measured from the meeting rather than the ticket.

— Times in a quarter a published executive figure is restated because a hierarchy or calendar changed rather than because performance did. The least flattering measure here.

If those five move, the reporting layer has become a governance artifact. If only the count of certified datasets moved, it has not.

How Polymath solves this

We open with a working session, not a build. We take the ten questions your board asks that cross segments and, for each, we establish which hierarchy, calendar and unit governs the enterprise answer. Where your segments earn money in genuinely different ways, conformance everywhere is neither achievable nor wanted, so we treat this as Foundational Architecture and answer the architectural question instead: where the seam belongs. What we hand over is a small common definition set with its edges defended.

Your segment finance and commercial leads sit in that room, because the decision is theirs to lose. Once the rules exist we implement them in the reporting layer you already run — the conversion, the bridge printed beside the number, the trigger that forces re-approval when a hierarchy changes — and your analysts produce the following board pack from them, with us alongside for one cycle and then out.

The reuse sits in the seam. A new segment or an acquisition is mapped into a definition set that already exists rather than negotiating a fresh one, which is most of what makes reporting integration slow and expensive. Your question set also becomes the specification for whatever you build next — planning, segment disclosure, and eventually any system asked a cross-segment question with nobody in the middle to make the conversion silently. A segment or an acquisition added later is mapped into definitions that already hold, so the expensive part — deciding who accepts the distortion — has been paid for once and not again.

What it costs to do nothing

Cross-functional questions are the ones with money attached, and the slowest to answer. A category decision that waits a cycle for reconciliation is not delayed, it is forfeited: the shelf is set, the promotional window has closed, the production slot has gone elsewhere. Then the executive team adapts. Once the pack has been wrong for frame reasons, executives stop acting on it and commission their own analysis, rebuilding analytical capacity inside each function in incompatible frames. The estate grows while its authority shrinks.

For a listed company the last cost is external. Guidance, board reporting and segment disclosure rest on the same conversions. When a figure moves because a hierarchy changed rather than because the business did, the explanation is owed to people under no obligation to find that distinction interesting.

Certifying more data will not settle a cross-functional question. Somebody has to decide which frame wins, and write it where the next person can find it. That written decision is where we begin.