Automate first the task that happens often, follows clear rules, costs you money when it's late and right now depends on someone remembering to do it. In most service businesses that's what happens to a new enquiry, so the first reply, the follow-up and the booking. Score your options with the test below, build the winner properly, and only then move on to the next one.
Why "automate everything" projects stall
When a big automation project stalls, it's hardly ever the technology. A lot of times the process was never written down, so nobody agrees what the automation should do. Or five tools need connecting before anything works, so nothing goes live for months. Or the team doesn't trust it and keeps doing it by hand anyway.
Doing one flow at a time gets round all three. It's small enough to finish, the team actually notices the difference, and it gives you one clean record to build the next automation on.
Score your candidates: a five-question test
Ask your team "what do you do by hand that annoys you?" and list every repeated task they mention. Then score each one from 0 to 2 on these five questions.
| Question | 0 | 1 | 2 |
|---|---|---|---|
| How often does it happen? | Monthly or less | Weekly | Daily or more |
| How long does it take each time? | Under 2 minutes | 2 to 15 minutes | Over 15 minutes |
| What does a delay or mistake cost? | Nothing much | Annoyance or redoing work | Lost customers or money |
| Are the rules clear? | Judgement every time | Mostly clear, some exceptions | Same steps every time |
| Does it rely on someone remembering? | No, it already runs by itself | Sometimes | Yes, completely |
The top scorer is your first project. A score of 8 or more is usually worth automating now. Below 5, leave it alone for now.
Pay close attention to the fourth question. If a task needs judgement every time, it's a bad first automation, however much time it takes. It might suit AI help later, with a person approving the result, but it shouldn't be step one.
A worked example
These are made-up numbers, not client data. Say you've got a physiotherapy clinic and it lists four tasks.
| Task | Often | Time | Cost of delay | Clear rules | Relies on memory | Score |
|---|---|---|---|---|---|---|
| Replying to new enquiries | 2 | 1 | 2 | 2 | 2 | 9 |
| Appointment reminders | 2 | 0 | 2 | 2 | 2 | 8 |
| Weekly revenue report | 1 | 2 | 1 | 2 | 1 | 7 |
| Handling complaints | 1 | 2 | 2 | 0 | 1 | 6 |
Enquiries win, reminders come next, and then the weekly report. Complaints score high on cost but low on rules, and that's exactly the right answer, because complaints need a person. All the automation should do there is get them to the right person fast.
The usual order for service businesses
Your scores decide, but most owner-led service businesses end up in a similar order. Each step makes the next one easier, because the data's already in one place.
- New enquiries. Reply within minutes, answer the common questions and book. See speed-to-lead for why this goes first.
- Follow-up. Enquiries that don't book get a short, useful sequence that stops when they book, reply or opt out.
- Bookings. Confirmations, reminders and prompts to rebook, which also cut no-shows. We went through the detail in missed visits and patient no-shows.
- Handoffs and admin. Passing details from sales to delivery, updating the CRM and filing documents without typing things twice.
- Reporting. The weekly numbers made automatically, from the records the first four steps now keep clean.
You'll notice reporting comes last. If you automate a report built on messy data, you just get the wrong answer faster.
What to keep a person on
Automation should handle the predictable work and get the rest ready. Keep a person on these.
- decisions involving money or legal commitments
- complaints and upset customers
- messages sent in your name that aren't routine
- regulated advice, like medical, legal or financial
A good system gets these to the right person quickly, with the context attached, instead of trying to answer them itself. We list the decisions we never automate in where people must stay in an AI workflow.
Off-the-shelf tools or a custom build?
For a first automation, start with the tools you already pay for. Most CRMs and booking systems come with reminders and basic sequences. Connector tools like Zapier can fill the gaps.
Know how connector costs grow before you rely on them. On Zapier, each successful action step counts as a task, triggers and filters don't, and when you hit your plan limit new runs are held until you pay for more. So a five-step automation running on 40 enquiries a day uses 200 tasks a day. That's one of the reasons we run everything on one platform instead.
A custom build makes sense when the flow goes across several systems, needs rules the tools can't handle, or when connector costs and breakages start costing more than the build. When we do it, it runs on Agentryx One and we run it for you. The price is set per business once you've seen a demo of your own system, and your data stays yours. If your tools already cover it, you don't need us for this step.
When not to automate yet
Hold off if the process still changes week to week, if nobody can describe the steps the same way twice, or if a tool you've got already does it well. In those cases, write the process down and run it by hand for a month first. That way you automate something that works, and you don't lock in something that doesn't.
Do this week
Ask each person on your team for the three tasks they'd most like to stop doing by hand. Score every task with the table above in one sitting, and pick the top one. Then write its steps from start to finish and mark every point where someone waits, chases or types something in again. That list is the brief for your first automation.
If you'd like a second opinion on the scores, book a call and we'll go through your list with you.
