Automate first the weekly work that starts with something your systems can see and ends in a clear done state, such as turning a new enquiry into a CRM record and a confirmation email. Keep judgement with a person, and add the next automation only once the first runs cleanly.
In over five years of business analysis, I've seen two mistakes again and again: automating everything at once, which stalls, or the wrong thing first, which burns the team's trust. Automate a mess and you get a faster mess. Here's how I'd decide what to automate first in a small service business, in the order we work at Oduro: process first, tools second, a person on every judgement call.
Start where work is repeated and owned
Your first automation should remove retyping and forgotten handoffs, not invent a new way of working. Look for weekly work with a clear trigger and a proper home in a tool you own.
That last part matters more than people think. If nobody owns the client list, or there are three versions of it, automation copies the confusion faster. So ask: where does this live, and who keeps it right? If the honest answer is "depends who you ask", that's your first job, and it isn't a tech one. Some spreadsheets should be software, and it's worth knowing the real cost of disconnected tools.
If this feels basic, you're not alone. When the British Chambers of Commerce surveyed over 1,500 UK firms in 2025, just 11% said they used technology to a great extent to automate or streamline operations. And the ONS found that better-managed firms were far more likely to have adopted at least one advanced technology, such as cloud systems or specialised software: 88% of the best-run tenth, against 51% of the bottom tenth. Tidy processes first, then tools.
Say you run a sales and lettings agency and want every new Rightmove enquiry pushed into your CRM with an automatic viewing invite. But your applicants live in the CRM, a negotiator's spreadsheet and two shared inboxes, and nobody owns keeping them straight. Switch it on and the same buyer gets three invites while those on the spreadsheet get none. First, agree the CRM is the only applicant list and name who keeps it clean.
A simple priority test
Run each candidate job through four questions:
- Does the same sequence run at least weekly?
- Is the trigger something the system can see (form submit, status change, calendar event)?
- Would a wrong run be easy to spot and reverse?
- Does a person still make the judgement calls that matter?
If you answer no to the last two, shrink the scope: automate the boring middle and leave judgement at the edges.
The first two questions say whether it's worth it; the last two, whether it's safe. Be strict: "usually" counts as a no, and a job that starts "when someone remembers" isn't a candidate yet.
Judgement has a legal side too. Since February 2026, the Data (Use and Access) Act 2025 has let businesses make significant decisions about people by purely automated means, such as turning down a credit application, in more situations than before, with tighter limits for health and other sensitive data. In return, you must tell people about the decision, let them put their side, offer a human review and let them challenge it. Gov.uk summarises the change, and the ICO is updating its guidance to match. Most first automations sit nowhere near that line, but it's another reason to keep people on decisions. See the decisions I'd never hand over to AI.
Common first wins in service teams
Where I'd usually look first:
- An enquiry that becomes a CRM row and a confirmation email. No retyping, and the customer knows you've got it. More on what your website should hand off after the enquiry.
- A booked call that creates a prep checklist for the owner. Whoever takes it turns up ready.
- A closed job that sends an invoice reminder and a feedback request. The bit everyone forgets on a busy Friday.
Next come following up leads who don't book and reminders for when show rates drop. Both have clear triggers and are easy to check.
Weak first candidates need a creative decision, legal wording or a one-off exception every time. Pricing an awkward job, handling a complaint, deciding a refund: keep those with people. Automate the admin around them, like logging the complaint and alerting the right person, not the decision.
Picture a small accountancy practice in January, with records arriving by email, upload and post. A good first automation: when a client is marked "records received" in the practice management software, the client gets a short acknowledgement and the right accountant gets a review task. "Done" means the email has gone and the task is assigned; the practice manager owns it and is alerted if a run fails. The accountant still reviews the records; the automation just stops them sitting unnoticed in an inbox.
Build the first connection
Connect the tools and write down what starts it, what each step does, what "done" looks like and who hears if it fails. That page is your SOP, so someone other than the builder can fix it later.
Then run it alongside the manual way, checking every run at first, so you hear about breaks before your customer does. If there's an AI step anywhere in the chain, test it properly before customers see it.
Take that estate agency again, once the CRM is the only applicant list. For the first few weeks, a negotiator still logs enquiries by hand and compares the two lists each evening. Any enquiry missed or doubled gets traced to its cause before the manual step is dropped.
Add the next automation only once that one runs cleanly. We design this layer as Workflow automation so the first win is measurable and the next ones can plug in. Decide up front what you'll watch, such as missed enquiries or unpaid invoices; the nine numbers every service dashboard needs are a good start.
Where to go next with automation
Our automation guides start here. Pick the one matching where your week leaks:
- Leads going cold: how to follow up with leads who don't book
- No-shows: why show rates drop and the automations that fix it
- Enquiries after hours: how to serve your customers out of hours
- Tools that don't talk to each other: the real cost of disconnected tools
- Adding AI on top: is your business ready for AI?
Or browse all our guides.
Questions people ask
How do I decide what to automate first in a small service business?
Pick weekly work with a trigger your systems can see and a clear done state, like an enquiry becoming a CRM record and confirmation email, then apply the four questions above.
Should I start with automation or AI?
Usually plain automation, because rule-based steps are easier to check. Once the process is written down and the data lives in one place, AI has a solid base.
What shouldn't I automate?
Anything needing judgement, legal wording or a one-off exception every time. Automate the admin around those jobs; leave the decision with a person.
Do I need new software to get started?
Not always. Many first automations just connect tools you already use. If key information lives in a spreadsheet only one person understands, fix that first.
Sources
- Turning Point As More SMEs Unlock AI (British Chambers of Commerce (with Intuit))
- Management practices and the adoption of technology and artificial intelligence in UK firms: 2023 (Office for National Statistics)
- Data (Use and Access) Act 2025: data protection and privacy changes (GOV.UK (DSIT))