It's September, which means that somewhere in your building a spreadsheet called something like FY27 Budget — v3 is already doing the rounds, and the line for your data team is being decided. Often without anyone from the data team in the room. If you want a say in your 2027 number, the window is open right now, and it closes a lot faster than most people expect: the typical calendar-year budget kicks off this month, gets its serious review through October, goes to the executives as a draft by mid-November, and is locked by around the 15th of December. Which means the time to make your case is the next few weeks — not January, when you find out what you were given.

I've been on both sides of this: the engineer whose sensible ask got cut because nobody translated it, and the architect who learnt, slowly, how to argue for a number in a language the people holding the money actually buy. Here's what I wish I'd known earlier.

Sell the outcome, not the platform

Finance does not want to buy a Fabric capacity. They have no idea what an F64 is, and — this is the part engineers resent — they are right not to care. What they buy is a month-end close that lands two days sooner, a pricing decision made on numbers people trust, an AI pilot that can finally go to production instead of dying in a risk review. Every line in your ask should be a business outcome or a risk avoided, with the technology as the footnote. "We need capacity for the new lakehouse" loses. "We need to keep the sales forecast trustworthy as we double the data behind it, and here's what that costs" wins.

The clearest version I ever watched: a team asked, two years running, for money to "modernise the platform." The first year it was cut without discussion — it sounded like the data team wanting nicer tools. The second year they asked to "keep the month-end close from slipping past day five, which slipped twice last year and held up the board pack both times," attached the identical engineering to it, and got the lot, with a note of thanks. Not a line of the technical plan had changed between the two asks — only the sentence in front of it.

Reframe governance from overhead to insurance

The line items that get cut first are always the unglamorous ones — the catalogue, the stewardship time, the data-quality work — because they read like housekeeping. Reframe them. Since the EU AI Act's high-risk obligations came into force, "we can't document how our AI reaches a decision" stopped being a tidiness problem and became a legal and board-level one. Governance spend is not overhead; it is the insurance premium on every AI and reporting bet the business is about to make. Put a name to the risk you are buying down, and the boring line item suddenly has a sponsor. Make it specific: "we can't currently show which data an AI-assisted decision touched" stops being housekeeping the moment a regulator or an internal auditor can ask the question — it becomes an exposure with a plausible fine attached, and finance funds exposures with fines attached far more readily than it funds tidiness. The discipline is not to inflate the risk, but to name the real one plainly and let its size do the arguing.

Show the cost of not spending

A budget ask framed only as "give us money to do more" competes with every other team's ask to do more. A stronger frame is the cost of standing still. I wrote a while ago about the governance debt behind every stalled AI project — the decade of small shortcuts that all come due the moment you point an AI at production. Budget season is when that debt gets its interest bill. Make it concrete: the capacity you will over-buy in a panic next summer because nobody sized it deliberately; the shadow spreadsheets multiplying because the platform is painful; the pilot that will stall, visibly, in front of the executives who funded it. Nobody funds a vitamin. Plenty of people fund a painkiller.

Lead with one number

The instinct is to prove rigour by showing all your working — a forty-row spreadsheet with every SKU and licence. It does the opposite. A wall of detail invites line-by-line haggling and signals that you are not sure which parts matter. Bring one defensible headline number with a clear story behind it, and keep the detail in your back pocket for when they ask. Building that number is less work than the spreadsheet instinct fears: take the one or two outcomes you're protecting, put a rough figure on what they're worth or what their failure would cost, and make sure the ask is visibly smaller than that. You're not proving you can do arithmetic — you're showing that the money buys back something worth more than itself. "We're asking for X, it protects these two outcomes and closes this one risk, and here's roughly how it breaks down if you want it" is a position. A spreadsheet is a negotiation you have already half-lost.

Ask in their calendar, in their words

This is the part that is pure communication, and it is the differentiator. The finance team has a process, a template, and a deadline, and your ask has to arrive inside all three. A brilliant business case delivered in the wrong week, in engineering language, on a slide that doesn't match their format, is a brilliant business case nobody reads. I've made this point before in a different frame — you can be completely right and completely unfunded — and budget season is where it bites hardest. Find out who owns the FY27 template, what date they need your input, and what shape it has to be in. Then meet them there.

Know whose number it actually is

The other thing I got wrong for years was assuming the decision sat with my own manager. It rarely does. Somewhere there is a finance business partner who owns the template your ask lands in, and a budget holder — often not in the data team at all — whose total your line comes out of. Those two people are worth ten minutes each, early, and not to lobby them: to understand what they are under pressure to deliver next year, so you can hang your ask on their problem instead of yours. A data platform that makes the CFO's cost target easier to hit is a very different proposition from a data platform that happens to be good engineering. Both can be true at once. Only one of them gets funded without a fight.

The honest caveat

Sometimes the reason your ask gets cut is that it deserves to. If you can't yet show what last year's spend actually bought, or you're asking for a platform-sized number to solve a problem you haven't scoped, no amount of framing will save it — and it probably shouldn't. When that's the case, the credible move is to ask for a smaller, provable slice: fund the one thing you can measure, deliver it, and come back next September with a result instead of a promise. A modest ask you can defend beats a moonshot you can't, and it earns you the room next time.

Because that's really what budget season decides — not a spreadsheet cell, but whether the organisation thinks of its data team as a cost to be trimmed or a capability to be resourced. That decision is being made now, in conversations you can still be part of if you move this month. The number gets signed in December. It gets won in September.