The forwardable business case for governed self-service analytics

What an executive needs from a business case

A CEO or CFO approving this needs four things, and not much else: what it costs, what happens if nothing changes, what will be different, and what they personally have to decide.

Leave out the return-on-investment multiple, and any figure borrowed from a vendor or a survey you cannot open. A CFO who finds one number they cannot check stops trusting the rest. Your own counts, from your own request queue, are smaller and much harder to argue with.

Count the requests before you write a word

The case rests on one piece of evidence: a log of the data requests you actually received. Keep it for two normal weeks, not the week of a board meeting. A spreadsheet with eight columns is enough:

In a larger company, the data team is not the only source: finance’s spreadsheet expert and the operations analysts answer requests too. Spend a week asking them to note theirs, or the log undercounts.

Repeats are the column that makes the case. A question you have answered three times is a question you could have defined once. A new question is not waste; it is the job. So when you summarise the log, lead with the repeats, not with the total.

The wait answers the CEO’s own question. From the two date columns, report the average and the longest wait. “Why does every number take so long” deserves a number of days, not an adjective.

Disputed numbers cost meetings, not analyst hours, and they usually make the stronger case. The money from repeat requests can look small next to a salary; a figure challenged in front of the CEO does not. Count disputes rather than pricing them: “revenue was challenged in [D] meetings in two weeks” needs no multiplier. If you can, attach one exhibit: last month’s revenue from sales next to finance’s, with each difference labelled. Our guide to one source of truth shows how to build that bridge.

To turn hours into money, ask finance or HR for the fully loaded hourly cost from your own payroll: salary plus benefits and employer taxes, per hour worked. If you would rather not put your own pay in a document, ask for a team-average rate instead. Only if neither is available, use a US-only fallback: the Bureau of Labor Statistics reported that in June 2026 wages and salaries were 70.0% of what private employers paid per hour worked, and benefits the other 30.0% (BLS, Employer Costs for Employee Compensation). So a US hourly wage divided by 0.7 is a fair rough estimate. Say in the note that it is an estimate.

The arithmetic is one line: hours on repeat questions × loaded hourly cost. Call the result what it is: time already paid for, not a cash saving. What makes it count is what those hours would go to instead, and only you can say that. For the queue meanwhile, see the analyst’s guide to one-off requests.

What governed self-service means, in plain words

The executive will not read a definition, so give them one sentence: every important number has one written definition and one owner, and people get it themselves instead of asking for it. Underneath, that is four things:

1. One definition per number. “Revenue” means one thing, written down, with the rules in the open: which orders count, which date, whether refunds come out.
2. One owner per number, from the business. Finance owns revenue. Marketing owns its own channel figures. The analyst writes the definition; the owner decides what it says.
3. Written once, in the central reporting database. The analyst turns each definition into SQL in the company’s data warehouse, with a plain description, so every report reads the same rows.
4. Reached from the tools people already use. A spreadsheet, a dashboard or an AI assistant, each reading the same definition. A BI tool, Power BI included, stays in step only if it reads the definition from the warehouse rather than redefining it in its own model.

It is not warehouse access for everyone. It is not a new reporting tool rolled out to the whole company. The analyst still decides what is published, and a question outside the defined numbers still comes to the analyst. The longer argument is in governed self-service analytics: fix the real bottleneck.

What it costs, stated honestly

A business case that hides the cost gets sent back. This one has three kinds, and none of them is a new hire.

Time

• The analyst’s time to write the first definitions. Start with the handful of numbers your log shows are asked for most. Your log already gives you a fair estimate: you have written each of those queries at least once. Writing it properly, with a description and checks, takes longer than the quick version did.
• Each owner’s time to agree their definition. Most definitions take one meeting. A contested one can run for weeks if nobody has the authority to end the discussion. That is why the note asks the executive to settle disagreements, and caps what that costs them: one short decision meeting per dispute, with both versions side by side. Whoever settles a revenue dispute, management revenue must still reconcile to the ledger and to finance’s revenue recognition policy.
• Changes to a definition. Definitions change when the business does: a new product line, a new pricing model. The number’s owner approves a redefinition (finance, for the numbers finance owns). Keep each definition’s history, a version and a date, and tell the people who use it, in the channel they already read. Figures already published under the old definition are either restated or labelled with the definition they used; the owner chooses which.

People

One analyst who can write SQL against the warehouse. One business owner per number, with the authority to make a definition final. And one executive who settles it when two owners disagree. The second role is the one companies forget to assign, and it decides whether the project lands.

Money

The first step needs no new tool: agreeing definitions and writing them as SQL in the warehouse you already have. If you later propose a tool to deliver the numbers, bring it back as its own decision, with its price taken from the vendor’s own page. And ask whoever pays the warehouse bill how queries are charged: when more people ask their own questions, the warehouse runs more queries. Propose a monthly ceiling and name who watches it.

Risks of doing it

Name them before the executive does. Other requests slip while you set up: say which ones, and agree what waits. Owners may not agree: that is what the decider is for, and a definition nobody agrees on is not published. Self-service sprawl: old dashboards and spreadsheets keep computing the same numbers their own way, and now there is one more version, not one fewer. That is why the note asks for a date to retire or repoint them. If an AI assistant reads the numbers, the query results it receives go to the AI provider, so security or IT signs off before anyone switches it on.

The risk of doing nothing

The alternative is not perfection. It is the queue you already have, and it does not stay the same size.

• Requests grow with the number of people asking, not with the analyst’s capacity. A second analyst adds capacity, and also writes a second version of the same queries, so the company gets two answers to one question faster.
• Two teams keep bringing two numbers to the same meeting. The time goes on which figure is right instead of on the decision. Often the gap is definition, not error; the reconciliation in our guide to one source of truth shows how to tell.
• The definitions live in one person’s head. If the analyst leaves, the rules for “active customer” or “net revenue” leave in their queries.
• If people are already pasting numbers into AI assistants, the answers use whatever meaning of “revenue” the assistant assumes. Pointing it at defined numbers, with their descriptions, makes the answers checkable.

What changes, and what does not

• The CEO or CFO asks for a defined number in a spreadsheet, a dashboard or an AI assistant and gets a figure built on the definition finance signed off. Each number on the board slide has an owner and a written definition behind it.
• Department heads answer their own repeat questions without waiting.
• The analyst stops pulling the same numbers every week, and spends the time on the questions that are actually new: why a number moved, not what it is.

And what it does not change, which belongs in the note too:

• Wrong source data stays wrong. If orders are missing from the system they come from, a definition will not bring them back. It does make the gap easier to find; data quality checks on the data itself are a separate job.
• New questions still come to the analyst. That is the point. Each one that comes back twice becomes the next definition.

How you will know it worked

Put the test in the proposal, so the review is not a matter of opinion. After the first definitions are live, run the same two-week log again, ask someone outside the data team to check it, and keep both logs away from month-end close. Compare:

• Repeat requests for the defined numbers: these should fall, without anyone being told to stop asking.
• Disputed numbers in meetings: these should fall for the defined numbers.
• Use of the defined numbers: who opened or pulled each one. A definition nobody uses has not replaced anything. A short walkthrough for the people who will use them, at launch, is part of the setup.
• The same question in two places: ask for one defined number in a spreadsheet and in a dashboard. If both read the definition, they match, allowing for display rounding; if they do not, one of them is still computing its own version.

Write the target and the stop rule into the note yourself, so nobody argues about them later: for example, “[repeats for the defined numbers fall by half by the review date]; if not, we stop and I report why”. Pick a target your log makes plausible; there is no standard one.

If repeats do not fall, the definitions cover the wrong questions: build the next one from the log, not from a whiteboard. If a team rebuilt a number in its own spreadsheet, find out what the definition missed.

Where OWOX Data Marts fits

The note names no tool. If you use OWOX Data Marts, the pieces map like this (documentation):

• A Data Mart is the written definition: SQL, a table or a view in your own warehouse, with a description and owners. Related Data Marts form a data model. The Business User role reports on shared Data Marts and cannot edit them (roles).
• Reports deliver the same rows to each tool. Google Sheets reports can run on a schedule; the Microsoft Excel add-in has no scheduled refresh, so the user refreshes it in the workbook (Excel); Data Studio pulls from the Data Mart.
• AI assistants read the defined numbers. Through the OWOX Data Marts Cloud MCP server, an AI assistant such as Claude or ChatGPT runs bounded queries on published Data Marts; the result rows go to its provider.

The note to send upstairs

Copy this, fill every bracket from your log, and cut any line you cannot stand behind. Never send it with a bracket left in: an empty bracket tells the reader you have not counted yet. It is written for someone who has four minutes and no interest in how the data works. If your manager should see it first, send it to them; it reads the same. If a data team is sending it, write it as “we”.

To show the shape, here is the first paragraph filled in with made-up figures (an example only, not a benchmark): “Over two weeks I logged 48 data requests, taking about 31 hours. 19 were questions I had answered before. The average wait was 2 days and the longest 9. Revenue was challenged in 3 meetings.”

Subject: One definition per number, and fewer data requests

[Name],

You asked why numbers take so long and why teams bring different ones. Here is what I found, and what I would like to do about it.

What I counted. Over two weeks I logged every data request: [N] in total, taking about [H] hours. [R] of them were questions I had already answered before. The average wait for a number was [days], the longest [days]. In the same two weeks, [D] numbers were challenged in meetings, mostly [revenue / customers / pipeline]. At [our loaded cost / a team-average rate] of about [$ per hour, from finance], the repeat questions alone took about [$X] of time we already pay for. That is not a cash saving; I would spend those hours on [what you would do instead].

What I am proposing. For the [five] numbers we are asked for most, [list them], we agree one definition each and one owner for each: [finance for revenue, marketing for its channel figures]. I write each definition down once, in our central reporting database (the data warehouse). People then get those numbers themselves, in [the spreadsheet / the dashboard / our AI assistant], instead of asking me, and everyone works from the same definition. I still decide what is published, which is what stops a new sprawl of versions; numbers finance owns are published only with finance’s sign-off.

What it costs. About [hours] of my time over [weeks] to set up, about [hours] from each owner, and a little of my time each month to keep the definitions current. Most definitions take one meeting with their owner; a contested one can take weeks, which is why I need you as the decider. The first step needs no new tool; if I propose one later, it comes back to you as its own decision, with its price. Our database bill may rise as more people run their own queries; I propose a ceiling of [$ per month], watched by [name]. No new hire.

What could go wrong. While I set this up, [which requests] will wait. Owners may disagree; a definition nobody agrees on is not published. Old reports may keep their own versions, which is why I am asking for decision 4. [If we use an AI assistant: the query results go to the AI provider, so IT signs off first.]

What happens if we do nothing. Requests keep growing with the team, we keep spending meetings on whose number is right, and the rules behind our key numbers stay in my head and my queries.

What will not change. If the data in our source systems is wrong, it will still be wrong. Questions outside the [five] numbers still come to me, and the ones that come back twice become the next definitions.

What I need from you.

1. Confirm an owner for each of the [five] numbers, and settle it when two owners disagree: one short decision meeting per dispute, and I bring both versions side by side.
2. Agree that once a number is defined, teams pull it themselves, so I spend my time on why numbers move, not on pulling them again.
3. Pick a review date: [date]. I will re-run the same log. The target is [repeats for the defined numbers fall by half]; if we miss it, we stop and I report why.
4. Set a date, [date], by which the existing reports that compute these [five] numbers are retired or repointed to the new definitions.

[Your name]

Data AnalyticsAI Analytics Team: Replace Your Backlog, Not Your PeopleJune 3, 2026→Data AnalyticsAn Analyst’s Guide to Handling Ad-hoc RequestsMarch 23, 2026→Data AnalyticsAsk Your Business Data in Your AI ChatJuly 20, 2026→

See all articles →