How to Work Out What AI Is Worth to You

The number you will be asked for

At some point, someone with a budget will ask what this is worth. If your answer is that it feels faster, the conversation ends there. If your answer is a number you can explain in two sentences, you get to keep going.

The good news is that the arithmetic is genuinely simple. The bad news is that almost everyone does it wrong, in the same specific way, and produces a figure that falls apart the first time it meets a finance person.

The formula

Write it out once and you will never be caught short again.

Value equals hours saved, times loaded hourly cost, minus the review time the AI adds, minus what the tools and setup cost.

Two of those four terms are usually missing from the pitch you will hear, and they are the two that subtract.

The classic error

Here it is, because you will see it this month if you have not already.

Somebody says the team writes forty client summaries a week, each takes thirty minutes, AI does it in five, so that is twenty-five minutes each, sixteen hours a week, roughly two full days of capacity recovered.

Then you watch what actually happens. The draft arrives in five minutes. Someone reads it against the source material for eight minutes, fixes two figures, rewrites the opening because it is bland, and checks a client name. Real elapsed time is eighteen minutes, not five. The saving is twelve minutes, not twenty-five. Still good. Less than half of what was claimed.

The fix is not complicated. Measure the whole loop, from blank page to approved output, both before and after. That is your only honest comparison.

Help me estimate the value of one AI use case. Use only the numbers I give you.
Do not invent benchmarks, industry averages or percentages.
Task: [what it is]
How often: [12] times per [week]
Time today, start to finish: [40] minutes
Expected time with AI, drafting only: [6] minutes
Expected review and correction time: [15] minutes
Loaded hourly cost: [45]
Tool cost: [25] per person per month for [3] people
Show:

Hours per year today and after
Annual saving with review time subtracted
Net saving after tool cost
Every assumption you relied on, listed plainly
If any input looks optimistic to you, say so.

Get a baseline before you change anything

You cannot claim a saving against a number you never measured. Before the pilot starts, ask the people doing the work to note how long it took, five times, in a shared note. Five samples is not science, but it is infinitely better than the guess you would otherwise defend in a steering meeting.

Measure the same five things again after four weeks, using the same definition of finished. Finished means approved and sent, not first draft produced.

Here are timings my team recorded for the same task, start to finish:
[42, 35, 61, 38, 90] minutes
Give me:

The typical time (median, not average, and say why)
The worst case and how much of the weekly total it accounts for
Whether five samples is enough to be confident, in one honest sentence
Use only these numbers. Do not compare them to any benchmark.

What the saving actually buys

Be careful about the sentence that follows your number. Two full days of recovered capacity does not mean two days of payroll disappears, and if you imply that it does, your team will hear exactly one thing.

Say what it buys instead: the backlog that never gets touched, the response times that slip at month end, the analysis nobody has time to do, the overtime that quietly happens in December. Those are real, defensible and do not read as a threat.

Here is my draft business case for an AI pilot:
[paste your numbers and reasoning]
Act as a sceptical finance partner. Find:

Any saving counted twice
Any place where I ignored the time spent reviewing or correcting output
Any assumption stated as fact
Any claim that would not survive a question about where the number came from
List the three weakest points, most serious first. Do not soften them.

Run that before you present, not after. A colleague from finance will find the same holes, only in front of an audience.