Will AI pay backfor your gyms?

A five-minute test for gym operators. Pick where it hurts. Step 1 checks it's worth a proof of concept. Step 2 checks the numbers still work in a bad year, when AI costs more, takes longer to check and fewer members use it. The questions were written by Mikey, who ran London Weightlifting for five years.

Where it hurts
Filled in with an example: winning back members who've stopped coming, across a multi-site gym business.

Is it worth a proof of concept?

Score each question 0 to 3. The two marked "must have" stop a project on their own if they score 0.

How many members do you have across your sites?
0 = Under 1,0003 = Over 20,000
Would you let AI send follow-up messages to members without someone approving each one?Must have
0 = Every message must be approved first3 = Yes, within rules we've agreed
Is your member data in one system?
0 = Spread across door entry, booking and payment systems3 = Visits, bookings and payments all in one system
Do you have clear rules for what members can be offered to stay, like a freeze, a cheaper plan or a free PT session?
0 = Staff decide case by case3 = Written rules for who gets what
Is there someone who would own this day to day?
0 = No one has the time3 = A named person who'd run it and report on it
Could you tell whether a follow-up kept a member, compared with members who weren't contacted?Must have
0 = No way to link the two3 = Yes, we can compare who was contacted with who stayed
25 / 30

24 to 30: a strong fit, ready to go. 15 to 23: worth doing, but start with a small pilot to prove it. Under 15: probably not worth the spend right now.

Return on investment

Rough numbers are fine. Only money you actually save or make counts as profit: freed-up staff time counts only if you stop paying for it. You'll see a normal year and a bad year. The bad year assumes half as many cases use AI, each one costs twice as much to run, and takes 50% longer to check.

The work
Members whose visits are dropping off
%
Of your at-risk members, the share AI would follow up and win back
What it saves
min
Time staff spend spotting them and writing to them
%
83 of 167 hours a year, through fewer hires, less overtime or agency staff. The rest frees staff up for other work and isn't counted
£
Including on-costs
£
Over the rest of their membership, e.g. 20 more months at £50 a month = £1,000
%
After the direct cost of serving them
What it costs
£
min
Checking a sample, not approving each message
£
One-off, year one only
months
Nothing is saved while it's being built
Normal yearBad year
Members won back1,000500
Staff cost you stop paying£2,500£1,250
Profit from members kept£700,000£350,000
AI running cost−£1,000−£1,000
Time spent checking−£100−£75
Build cost−£200,000−£200,000
First-year profit (after a 3-month build)£326,050£62,631
Pays for itselfMonth 7Month 10
Profit from year two£701,400£350,175
Spare staff time, not counted as profit83 hours42 hours

In the bad year, the AI's running cost is under 1% of what it saves, and checking its work is 0%. The AI bill matters more than checking time, so test a smaller, cheaper model.

Good first project

It's worth a proof of concept, and still makes £62,631 in its first year if things go badly.

The bad-year test comes from Duncan Anderson's essay “Capable enough to attack a bank, not reliable enough to run a shop”: an AI project is only worth backing if it still pays back when things go worse than planned.

A rough guide, not financial advice.

Got a result worth pursuing?Let's talk it through.

Bring your numbers and we'll show you what it would take to build, and what it would cost to run.

Book a call

Mikey Glenn

Ran London Weightlifting for five years, then sold it