WhatsApp AI agent case study: what actually happened after 90 days in production

Most content about AI automation stops at the explainer stage. Here is how it works, here is what it could save you, here is the theoretical upside. What's harder to find is what actually happens once a system like this is running in a real business, handling real client work, for months rather than days.

Fan Rescue is a UK field services business — commercial kitchen extraction and fire safety compliance. Their WhatsApp AI agent has been live for between one and three months. This is what it has actually done in that time, not what it was projected to do.

For a full walkthrough of what the agent actually does — invoicing, proposals, approval routing and CRM management, with a recorded demo of it running live — see what a live WhatsApp AI agent actually does. This post picks up from there with the numbers.

The numbers

These figures are pulled directly from the connected systems the agent writes to — Xero, the proposal system, and the Make.com scenario history. They are not estimates of what the system is capable of. They are what it has actually done.

• 22 invoices raised directly in Xero, triggered from a WhatsApp message
• 14 proposals generated and sent to clients from the field
• 10 hours a week saved on CRM admin, by Fan Rescue's own estimate — the agent updating records instead of a team member doing it by hand
• 8 automation scenarios built, shipped, and running continuously in Make.com
• 2,800 Make.com operations consumed processing the above — a useful reference point for anyone scoping the running cost of a similar build

The CRM time-saving figure is Fan Rescue's own estimate based on what the equivalent manual admin used to take, not an independently audited number. We're including it as reported because it's consistent with what we'd expect from removing manual double-entry across Monday.com and Xero — but it's worth being precise about what kind of figure it is.

What the 22 invoices actually represent

Every one of those 22 invoices started as a WhatsApp message from a field engineer, not a return trip to the office and not an end-of-week batch of admin. A job finishes on site, a message goes out, a draft invoice exists in Xero within seconds. What used to be same-week invoicing became same-day invoicing, without adding a step to anyone's job.

That shift matters more than the raw count suggests. Invoicing delay is one of the quiet drags on cash flow in field services businesses — work is done, but the invoice sits in someone's head or notebook until there's time to sit down and process it. Removing that delay doesn't require working faster. It requires removing the step where the invoice waits for someone to have a spare ten minutes.

What the 14 proposals represent

Proposal generation was the workflow we highlighted in the original piece on WhatsApp AI agents for business operations — a site visit that used to end with "I'll write that up when I'm back" now ends with a client-ready proposal published to a live link before the engineer has left the car park.

14 proposals in this window means 14 opportunities where the gap between a completed site visit and a document landing in the client's inbox shrank from days to minutes. That's not a productivity statistic — it's a competitive one. The business that quotes first, while the job is still fresh in the client's mind, wins more of those jobs.

8 scenarios, 2,800 operations — what that means practically

For anyone weighing up whether a build like this is worth doing, the Make.com operation count is a genuinely useful data point. 2,800 operations to run 8 live scenarios, covering invoicing, proposals, and CRM updates, sits comfortably inside Make.com's Pro tier allowance. This is not a system that requires enterprise-level automation spend to run — it runs on the same plan tier we'd recommend to any SME client.

The 8 scenarios themselves break down roughly as: message intent routing, Xero invoice creation, Monday.com record updates, proposal generation and publishing, client record creation, and a handful of smaller supporting flows that keep the above running reliably rather than doing anything visible on their own. That's a realistic scope for what a properly built operational WhatsApp AI agent actually requires — not one giant scenario, but several focused ones working together.

What this doesn't cover — and why that matters

In the interest of being straight about what these numbers do and don't prove: this is 90 days of data from one business, in one sector. It's real, but it's not a controlled study, and results in a different business — different call volume, different systems, different starting point — will differ. Some of that difference is favourable (a business with messier existing data will see a bigger before/after gap once it's automated); some of it isn't (a business with lower message volume simply won't generate 22 invoices through the agent in the same window).

What we think is genuinely transferable is the pattern: messages that used to sit and wait for someone to act on them manually now trigger the action directly. That pattern holds regardless of sector. The exact numbers will vary with how much of your operation already runs through WhatsApp.

Where to start if you want to see this for your business

The fastest way to know whether this is worth building for your operation is the same starting point we recommended in the original piece: map the messages your team sends every day that currently result in someone manually opening Xero, your CRM, or a document template. If that list is more than a handful of actions a day, the case is there.

Our operations audit checklist is a structured way to do that mapping yourself. If you'd rather talk it through, the WhatsApp AI chatbot service page covers exactly how we scope and build these.