The ownership map: a request triage desk

External. The answer lives in the market: competitors, buyers, agencies, retailers, culture. Marketing or BD is accountable and usually responsible. Data is informed, or consulted for one narrow internal fact. "Review competitor X's new range" is external: the analyst has no privileged view of a competitor's shelf.

Internal. The answer lives in the company's own data: orders, sessions, sends, subscriptions, spend. Data is accountable and responsible; marketing is consulted on what decision the answer feeds. "Which search queries bring buyers?" is internal: only a join of search queries to orders can answer it, and marketing cannot easily build that join alone.

Both. The answer needs a market half and an evidence half. Market sizing needs a top-down figure from outside and a bottom-up figure from our own conversion and price data. An ideal customer profile needs BD's knowledge of who buys and our own retention by account type. These are the rows where a single owner is always wrong: give it all to BD and the evidence half is anecdote; give it all to data and the market half is guesswork by people who do not talk to buyers. How to build both halves of a sizing, and sign them separately, is the whole of Market Sizing for Data Teams, Business session 3 ↗.

A common shortcut is to route by who asked: everything from BD goes to the BD analyst, everything from brand goes to the survey person. It fails because the same requester sends all three kinds of ask in one week. Kettle & Co's BD lead sends an internal ask (why are office-tea trials not converting), a split ask (who is our ideal B2B customer) and an external ask dressed as a data request (write our supermarket penetration strategy) within ten working days. Routing by requester hides that; tagging by knowledge shows it.

The tag also survives reorganisations. People change roles; the fact that a competitor review needs knowledge of a competitor does not.

Say yes to everything asks 426 hours of a 120-hour month. The team is not lazy and not slow; it is writing a penetration strategy (60 modelled hours), a market research study (40) and a brand tracker (28) while the campaign readout, the dashboard, the churn dip, the holdout read, the CAC table and the B2B funnel wait behind them. Nine of the twelve asks only data can do miss their date. Zero asks match the RACI, because taking all of an internal ask still means making the business decision that follows it.

Say no to anything not ours uses 67 hours and nothing is late. It looks like the best-run team in the building. It also leaves nine decisions to be made on anecdote: the creator budget, the Meta budget, the trade fair, the Japan research, the supermarket plan, the market size, the ICP, the Amazon launch and the brand tracker all go ahead without the internal half of their evidence. Fifteen of 24 asks match, because the internal and purely external rows are right and every split row is wrong.

Split by the RACI uses 106 hours, nothing is late, nothing is decided on anecdote, and all 24 match. It is not free: it costs 39 hours more than saying no, and those hours are the consulted deliverables. That is the price of "consulted" being a real job.

Press it. Capacity drops by a third to 80 hours and the bench prints, for every rung, which data-owned asks now miss their date. Yes-to-everything starves ten. The split starves six: the campaign readout, the dashboard, the churn dip, the holdout read, the CAC table and the B2B funnel, all in the second half of the month, behind the consulted work. Say-no starves only one, the B2B funnel, and still leaves nine decisions on anecdote.

That result is the honest one and it is worth saying out loud: when capacity drops, the split does not survive unchanged. The answer is not to fall back to "no". It is to renegotiate the consulted deliverables in writing: the brand tracker read moves to next month, the Amazon evidence shrinks to the top five SKUs, and each requester hears it with a new date before the old date passes. Try it on the desk: set two of the split rows to a smaller scope by redirecting them, and read which column pays. Partnership session 4 writes the scripts for that conversation.

What it fixes. The split rows get one owner for the evidence, so market sizing does not arrive as two numbers from two teams that never met. Research method gets one standard: sample sizes, weighting and survey design are owned by people trained in them. Marketing and BD get one door to knock on.

What it costs. Two things, and the bench shows both. First, analysts drift from supplying evidence into owning the conclusion: the person who wrote the sizing model starts presenting the go or no-go, and the BD lead becomes an audience for their own strategy. Second, external research is expensive in hours and lands on the same capacity as the internal asks only data can do. The yes rung on the desk is model A without its own headcount.

Model A works best when the insights function is staffed separately from the analysts who own internal evidence, and when its remit is written as "evidence for the decision" rather than "the decision".

Data informs the bet; the business places it. The marketing lead will be judged on the Meta budget; the BD lead on the supermarket plan. If an analyst writes the plan, three things go wrong at once: the person accountable did not make the decision they are accountable for, the analyst is now defending a strategy instead of the evidence under it, and the next time the evidence disagrees with the plan, the analyst is marking their own homework.

Decision intelligence writers make a version of this point by separating the decision-maker's role from the analyst's; data mesh writers make another by treating ownership as a domain boundary that needs an explicit contract wherever two domains meet. Neither turns the split into a rule. The split is a choice, argued from who carries the outcome, and a team that chooses model A for good reasons is not wrong.

Print this block and adapt it to your own team. The hours are Kettle & Co's modelled hours; yours will differ. A is accountable, R responsible, C consulted, I informed. Where data is C, the deliverable column is the contract.

Print this page   The RACI as a general project tool, including how to run the workshop that fills it, lives in Tech Project PMO, session 5 ↗. This page applies it to the marketing and data boundary only.

One A per row. Two accountable owners is no accountable owner. On split rows, the A sits with the business side and data is R for its half, which is how market sizing reads "BD A, BD R top-down, data R bottom-up".

Every data C has a deliverable and a date. If you cannot name what data hands over, the C is decoration and should be an I.

Review the matrix when capacity changes. The break button showed that the right split at 120 hours starves six asks at 80. The matrix is a plan for a month, not a constitution.