If you've been paying attention to AI this year, you probably have a list by now. It came from a conference session, a peer group, a vendor, or an industry report: firms like yours are using AI to answer inquiries faster, chase documents, draft proposals, prep jobs, follow up on unpaid invoices.
The list is useful. It tells you where other businesses in your industry are finding a return, which saves you from guessing in the dark.
What it can't tell you is whether any of those ideas is worth money in your business. That depends on numbers only you have: how often the work happens, who does it, what it costs when it goes wrong, and what your team would do with the time back.
That's the gap between an industry idea and a business case. Closing it takes about a week of light counting, not a consultant's model.
Why the industry number isn't your number
Industry figures are averages across businesses that aren't yours. A report might say firms in your sector typically see the biggest gains in client intake. Maybe. But if your intake is already fast and your real drag is the back-and-forth before work can start, the average points you at the wrong job.
There's a second problem. A lot of published AI results come from larger companies, early adopters, or the vendors selling the tools. That doesn't make them wrong. It makes them directional.
So treat the industry list as a set of places to look. Then go measure.
Pick one idea and make it a job
"Use AI for client communication" is a category. You can't measure a category.
Turn the idea into one specific job with a start and an end. A few examples of the translation:
- "AI for intake" becomes "replying to new website inquiries and booking the first call"
- "AI for collections" becomes "sending reminders on invoices more than 15 days past due"
- "AI for reporting" becomes "building the Friday client status update"
If you're not sure which idea to start with, our guide on where to put your first lever walks through ranking the candidates. For this exercise, one job is enough.
The five measurements
Count these from your own work over two to four recent weeks. Rough is fine. Invented is not.
1. Volume. How many times does the job happen in a typical week or month? Pull it from your CRM, inbox, invoicing system, or job board rather than from memory. Memory tends to round up for annoying work.
2. Hands-on time. How long does one instance take the person doing it, including the checking and the fixing? Time three or four real ones. Include the one that went sideways.
3. Wait time. How long does the job sit between steps: waiting for information, an approval, or someone to get to it? This is often where the money is. A proposal that takes 40 minutes to write but sits four days before it goes out is a speed-to-revenue problem, and faster drafting won't fix it alone.
4. Misses. How often does the job go wrong or not happen at all? Inquiries that never got a reply. Invoices that went out late. Follow-ups that slipped. Rework because something was missed. Count them.
5. Who does it. Name the person, and note what their time is worth and what they'd do with more of it. An hour of your senior project lead is not the same as an hour of an admin, and neither is the same as an hour of yours.
Write those five down for the one job. That's your baseline. Without it, nobody can tell you later whether the change worked, including you.
Name the return before anyone quotes a price
Every AI idea promises to save time. Time is the input. The return is what happens to the business because of it, and it usually shows up in one of three places.
Revenue the work protects or adds. Faster replies that book more first calls. Fewer quotes that go cold. Count the added contribution after the cost of delivering the work, not the full sale.
Cost that actually leaves. Less overtime, a cancelled subscription, outside help you no longer need. If it doesn't show up in your financials, it isn't a cash saving.
Capacity your team gets back. Hours returned to people who can spend them serving customers, solving problems, or doing work that needs judgment. That's a real return. It's just a different one, and it only turns into money if you decide what those hours go toward.
Pick the one that matters most for this job and say it plainly. Don't count the same hours as payroll savings and as capacity for new work. We went deeper on that trap in AI ROI for small businesses.
A hypothetical, start to finish
Here's an illustration with made-up numbers, not a client result.
Say you run an IT services firm, and the industry idea that caught your eye was "AI-assisted inquiry response." You turn it into a job: replying to new inquiries from your website and booking a first call.
You count the last four weeks:
- Volume: about 30 inquiries a month
- Hands-on time: around 15 minutes each to read, look up the business, and write a reply
- Wait time: the median reply goes out the next business day, because the person who handles it is also on service tickets
- Misses: four inquiries a month get no reply at all
- Who does it: your service coordinator, who'd rather spend the time on existing clients
Now the business case gets interesting. The hands-on time is only about seven and a half hours a month. If that were the whole story, this might not be worth building.
But the misses and the wait time point to a different return. If four inquiries a month go unanswered, and next-day replies lose some of the rest, the real question is how many first calls you're not booking. That's the number to estimate, cautiously, using your own close rate and the contribution from a typical new client.
So the case you write isn't "save the coordinator seven hours." It's "reply to every inquiry within a business hour and stop losing the four a month that fall through." Capacity for the coordinator is a bonus, tracked separately.
That shift, from the time the vendor talks about to the return your numbers point to, is the whole point of measuring first.
Set the target and the stop sign
Before anything gets built, write down two numbers.
The target. What the measure should reach and by when. "Every inquiry gets a reply within one business hour, with zero unanswered, by the end of the first month."
The stop sign. The result that tells you to rethink. "If reply time hasn't improved and the coordinator is spending more time checking drafts than writing them, we pause and review."
The stop sign matters more than people expect. It keeps a project from drifting for six months on hope.
Then do the same honest math on cost: setup, integration, data cleanup, training, monthly usage, and the review time someone will spend checking the output. Put the first year's costs next to the first year's expected return. If the case only works when you leave costs out, it doesn't work.
The one-page business case
When you're done, you should be able to fill this out:
Industry idea: [what you read or heard]
Our job: [one specific job with a start and an end]
Baseline: Volume: [#/month]. Hands-on time: [minutes each]. Wait time: [typical delay]. Misses: [#/month]. Done by: [name]
Return we're after: Revenue protected or added / Cost removed / Capacity returned (pick one, say how you'll see it)
Target: [measure] reaches [level] by [date]
Stop sign: [result that means we pause]
First-year cost estimate: [setup + usage + review time + cleanup]
Owner: [who checks the result]
That page is worth more than the industry report that started it, because it's about your business. It also makes vendor conversations shorter, because the question stops being what AI can do and becomes whether this tool can move this number.
Before you build, run the job through the three people, process, and data checks. The baseline you just counted will make those checks faster.
If you don't have an industry list yet, start with the free AI Leverage Report for your sector. It shows where the return usually shows up in businesses like yours. If you have the list and want help turning one idea into your numbers, that's exactly what the free 30-minute AI Leverage Call is for. Bring the one job and whatever baseline you've counted, even if it's rough.
