The AI Investment Calculator: The Maths Every AI Budget Needs

Master of the E-Series Tarocchi, Arithmetic, 1460–70. The Metropolitan Museum of Art, public domain.

In most board meetings where an AI budget comes up, the CFO asks for the return and the vendor answers with a word like transformation. The budget often passes anyway, because AI is on the strategic priority list, and the firm ends up with a programme and no target. The AI Investment Calculator is built to put a payback number and its assumptions on the table before that approval.

Most AI pilots produce no measurable profit

MIT's State of AI in Business 2025 found that about 95 percent of enterprise AI pilots produced no measurable profit. Gartner expects at least 30 percent of generative AI projects to be abandoned after proof of concept. The studies define failure differently, and I would not rest a case on either figure alone.

They agree on direction, though. For a typical pilot, the likeliest outcome is nothing measurable. A budget that assumes success as the default is priced for the exception.

How the calculator works backwards from the value you want

A normal business case starts from the vendor's quote and guesses at the benefit afterwards. The calculator starts from the benefit, because that is the number you will have to defend a year from now.

You enter the annual value you are aiming for in three separate lines: cost savings, new revenue and retention value. Keeping them apart shows which one is carrying the case. Then you describe the context:

• Build approach: off-the-shelf, configure and integrate, or custom build.
• Sector, and the most sensitive data the system will touch.
• How many people will work differently because of it.
• How ready the organisation is for the change.
• The planning horizon.

Cash flow is modelled month by month, so value starts at go-live and not on day one of the project. That single change usually adds months to the payback figure people expected.

The three cost buckets, and why the middle one decides the outcome

Direct AI cost covers licences, models, data work and integration. It is what a vendor quote covers, and in most budgets I review it is the entire budget.

Change management covers training, process redesign, communication and the productivity dip while people adjust. BCG's 10-20-70 finding puts roughly 10 percent of the effort in a successful AI deployment on algorithms, 20 percent on technology and data, and 70 percent on people and process. If technology accounts for most of your line items, the part that decides whether the project works has not been costed.

AI risk provision covers governance, security, oversight and rework, scaled to your sector and data sensitivity. IBM's Cost of a Data Breach research found breaches involving shadow AI cost around US$670,000 more on average. A provision is cheaper than that.

How to read the payback figure

The output is a range with its assumptions visible. Nobody has an exact return figure for an AI project, including the vendor whose slide says four months.

Use this rule of thumb. If your figure is within a few months of the vendor's, go ahead and negotiate on price. If yours says 18 months and theirs says four, ask the vendor to show which of your assumptions is wrong; one of you has a concrete error, and finding it is the most useful meeting you will have that quarter. If yours says 40 months or more, shrink the scope or look at a different product before approving anything.

Where the model is weakest

The ratios come from research on large organisations. A twelve-person firm in Kingston is not the population BCG studied, and I do not know how well 10-20-70 holds at that size. My guess is that the change management share is too low for a small firm, because a small firm has no slack to absorb a productivity dip. I cannot prove that, and the calculator will give you a confident number either way.

It also takes your value target on trust. In my experience that target is the assumption most often wrong, and the calculator will size an investment against savings that were never achievable. Treat the list of assumptions it forces into the open as the main output; the total is secondary.

What to do before your next AI budget meeting

1. The initiative sponsor: run the calculator and send the output to the CFO at least a week before the meeting. It takes under an hour, most of it spent agreeing the value target.
2. The CFO or finance lead: set a threshold above which no AI purchase is approved without a payback figure. Firms with an AI Officer should make that person responsible for running it.
3. The sponsor: split the value target into savings, revenue and retention, and name the owner of each line. A line with no owner should be removed from the case.
4. Procurement: ask the vendor for its own payback assumptions in writing. Compare them line by line with yours and record where they differ.
5. HR and operations leads: cost the change management bucket separately. Include training hours per person, the length of the productivity dip and any backfill; if it comes to less than the direct AI cost, check it again.
6. The project lead: set a review date six months after go-live. Compare actual value with the target on each line and decide then whether to scale, fix or stop.

Frequently Asked Questions

What is the difference between payback period and ROI?

Payback period is how long it takes for cumulative value to cover the cumulative cost; ROI is the net gain as a percentage of the cost over a set period. A project can show a high three-year ROI and still have a long payback, which matters to a small firm with tight cash flow. Look at both, and at the month when cumulative cash flow turns positive.

What should count as retention value?

Retention value is revenue you keep that you would otherwise lose: customers who stay because service improved, or staff who stay and save you rehiring and training costs. Estimate it from your own churn or turnover figures, not a vendor benchmark. It is the easiest line to inflate, so give it a named owner and a way to measure it after go-live.

Does the calculator store what I type?

No. It runs in your browser and does not send your inputs to a server. If you want to keep a result, take a screenshot or copy the figures into your own business case, because they will not be saved for you.

How should a small firm adjust the output?

Add to the change management bucket, not the technology one. Small firms usually have one or two people doing the work of a whole function, so the productivity dip while they learn a new system hits output directly. A practical check is to cost at least two weeks of reduced output for each person whose job changes.

Where can I read the research behind the figures?

The MIT figure is from the report The GenAI Divide: State of AI in Business 2025 by MIT's NANDA initiative. Gartner's projection on abandoned generative AI projects came from its 2024 press release on generative AI projects after proof of concept. BCG publishes its 10-20-70 work on its website, and IBM publishes the Cost of a Data Breach report each year.