Your 2027 budget is being set now. Don't ask for AI money, name what it replaces

Somewhere in your inbox there’s a 2027 planning spreadsheet with a deadline on it. If you want AI in next year’s plan, don’t add a new line called AI. Find something you already pay for that isn’t working, and make the case for replacing it with something that does. A finance director who’d never sign off an AI budget from nothing will often happily move money from a tool, a service or a manual process that everyone already moans about.

If your company runs a calendar year, that conversation is happening now. The plans for 2027 get argued over in October and November, and by January the numbers are mostly set. So the time to write the case is this month.

I don’t have to ask anyone for budget, because I own the company. But I do have to convince my own team, and I know what some of them think when I walk in on a Monday with a new idea. Another one from Chris. The ideas that land are never “let’s try some AI”. They’re “this thing we all hate doing, we could stop doing it”.

Why does asking for an AI budget so often fail?

Because a new line called AI is the easiest thing in any plan to cut. It has no history and no owner who’ll fight for it, and nobody can hold it to a number. The person asking usually can’t say what it will stop costing, either, which is the first thing finance will ask. When the plan has to come down by 10%, the experimental line goes first.

It also asks the wrong question. “Can we have money for AI?” makes the finance director judge a technology they’ve mostly read about in the papers. “Can we stop paying for this and get a better result for less?” asks them to judge a trade, and trades are what finance directors are good at.

So the job, before you write a word of the business case, is to find the trade.

What should you look for in the budget you already have?

Look for the lines that are expensive, inconsistent or quietly growing, and that nobody would defend in a meeting. In marketing and sales budgets they usually fall into four groups:

• Per-seat software that a handful of people use. Approval tools, survey tools, dashboards built for one purpose. The renewal creeps up every year and nobody checks who actually logged in.
• An agency or freelancer doing repeatable work by hand. The same report, the same posts, the same list build, month after month.
• Data your team no longer trusts. If people double-check a provider’s records before they use them, you’re paying twice.
• A manual job someone does every Friday. Nobody has ever costed it, because it’s “just part of the role”.

Each of those has two things a new AI line doesn’t. A number already in the budget, and people who’d be glad to see it go. The case moves money to something that works better rather than adding any, and the person who used to do the Friday job gets their Friday back.

When you’ve found the line, write down what it costs now, what’s wrong with it, and what good would look like instead. That’s the start of your marketing budget planning, and it beats any slide about what AI can do.

Which SaaS subscriptions could AI replace?

The thin ones. Software a few people use now and then, priced per seat, that’s mostly a database with some screens on it. Your systems of record, the CRM, marketing automation and the data warehouse, are the core of the martech stack and they aren’t going anywhere.

I found one of mine earlier this month. A SaaS tool I’ve paid for every month for 12 years emailed me about a price increase. I asked Claude to read its API docs, screenshot every part of the interface and clone it. Now I’ve got a 100% copy I can shape into the version I actually want (I only use about 40% of its features). I don’t mind paying. But I can’t unsee it either.

I wrote about the bigger version of this in Is SaaS dead? The rise of throwaway software . In the last week of September my AI tools built me 20 tiny web tools, each for one job, and most were binned within a day or two. Nobody logged in and nobody paid for a seat. The test from that piece is the one I’d take into budget season. At the next renewal, ask how many people used the tool last month and how much it really helped them. If the honest answer is a handful of people, you’ve found a line.

One caution. My copy holds real data, so it isn’t throwaway. It needs an owner, a security check and someone to keep it running, and those go in the costs.

What does a replacement look like in practice?

It looks like a swap. Something you already pay for comes out, and something that works better goes into the same slot.

Northern Data Group is a good example. Their existing data enrichment tools were, in the words of our case study, “expensive, inconsistent and full of hidden costs that made it difficult to plan and budget effectively”. Their ABM campaigns were underperforming, because the data underneath them wasn’t accurate or targeted enough.

What they needed was to stop paying for tools that weren’t working. We put Compass Data, our fully managed data enrichment service, in their place.

That’s the shape of every easy yes I’ve seen. A cost that was already there, a result that was already disappointing, and something better in the same slot. The sale was better data for the money, and AI barely came into it.

Sometimes the cost already in the budget is your own team’s time. When TCS’s Head of Social asked us to help his team with content creation, we set them up on Compass Agents, trained on the TCS brand and tone of voice. The case study counts more than 80 posts a month, more than 60 hours of production time saved a month, and 3 times faster from idea to post (case study ). Hours like those rarely show up as a line in anyone’s budget, which is exactly why they belong in the case.

What should go in the AI budget line?

Keep it short, and make it about the trade. I’d use this shape:

1. Name the line you want to replace. What it costs, how long you’ve had it, and what’s wrong with it, in plain words.
2. Say what replaces it. One sentence on what the new thing does for the team, not how it works.
3. Cost the whole thing. The tool, the setup, and the person-time for checking results, so there are no surprises later.
4. Say how you’ll know it worked. One or two numbers you already track, like open rates or hours spent on the job, with where they are today.
5. Start small. A first step you can finish in a few weeks and measure, before anyone commits the full year.

That last point matters more than it looks. A small, measurable first step is much easier to approve than a year-long programme, and if it works, next year’s case writes itself.

How would I write the line for my own SaaS tool?

This is the tool I cloned, written up the way I’d put it in a plan.

I haven’t cancelled it, by the way. But if it were going into a 2027 plan, that’s the line I’d write, and the first two rows don’t mention AI at all. They’re about a cost and a problem, which is what finance reads first.

The same table works for a manual job. Put the hours in the first row instead of the subscription, like the 60 hours a month in the TCS case study, and say who gets that time back.

Where does AI spending actually go?

Mostly not on the AI. The model is rarely the expensive bit, and it keeps getting cheaper. Anthropic says Opus 5.5 costs about 40% less to run than Opus 5, which came out two months earlier, as I wrote in a piece on how fast AI is moving .

The real cost is setting it up properly and checking what comes out. That means connecting it to the systems your team already uses, deciding what it should and shouldn’t do, and having a person check the results until you trust them. That work is where AI spending actually lands, and it’s the part people leave out of the business case, so the numbers look wrong six months in.

So put it in. A case that says “the model costs almost nothing, the setup and the checking cost this much, and here’s who does the checking” is one a finance director can believe. It also stops the awkward conversation in April when someone asks why the AI costs more than the slide said.

What will finance ask about your AI business case?

Much the same six questions every time. Have a one-sentence answer to each before the meeting, and you’ll spend the meeting on the trade instead of the technology.

• What does it cost today, all in? The invoice, plus the hours people spend working round it, plus any price rise you already know about.
• What stops if we say yes? Name the contract, the renewal date and the notice period. Miss the notice date on an annual contract and the saving can start a year late.
• What does it cost in year two? The model keeps getting cheaper, so year two is mostly people: who owns it, and who checks it.
• Who checks what it produces? Name a person. “The AI does it” isn’t an answer finance should accept.
• What’s the AI ROI, and when will we know? One or two numbers you already track, where they are today, and the date you’ll look again.
• What if it doesn’t work? Say what you’d go back to. Running the old and the new side by side for a few weeks is cheap insurance.

Finance would ask every one of those about any spend, and the trade you’ve already written answers most of them.

So how do you get budget for AI in 2027?

Stop asking for an AI budget, and ask to replace something instead. Find the line in your current plan that costs too much and delivers too little, whether that’s a per-seat tool, a service or a job someone does by hand. Cost the replacement honestly, including the setup and the checking, and make it small enough to try first. Your finance director only has to believe the trade, and that’s a much easier conversation to win this autumn.

Frequently asked questions

How do I get budget for AI if our marketing budget is flat?

Replace something rather than adding to the plan. Find a tool, service or manual process you already pay for that isn’t working, and make the case for moving that money to something that does.

Which SaaS subscriptions can AI replace?

The thin, per-seat tools a few people use occasionally, such as approval tools, simple survey and comparison tools, and one-purpose dashboards. Your CRM, marketing automation and data warehouse stay.

What should an AI business case include?

The line you’re replacing and what it costs, what the new tool does for the team, the full cost including setup and checking, how you’ll measure it, and a small first step.

How do I show finance the ROI of AI?

Use one or two numbers you already track, like open rates or hours spent on a job. Say where they are today and when you’ll check them again.

Is the AI model the expensive part?

Rarely. The model keeps getting cheaper. Most of the cost is setting it up properly, connecting it to your systems, and checking what comes out.

When should we make the case for next year’s AI budget?

If your company runs a calendar year, now. Plans for the next year are usually argued over in October and November.

Frequently asked questions