Key takeaways
- Score every candidate process from 1 to 5 on volume, rules, data, risk and measurability, then add the scores up.
- Good first projects are dull ones: invoice data entry, lead routing, report assembly. Pricing decisions and difficult customer conversations come later.
- Write down today’s numbers (time per task, error rate, response time) before you build anything. Without a baseline you can’t show a return.
- McKinsey’s research found workflow redesign had the biggest effect on whether companies saw bottom-line impact from gen AI, so fix the process before you automate it.
Why the first choice matters so much
Your first automation sets the tone for every one after it. If it saves visible time, people ask for the next one. If it stalls or misfires, you’ll spend a year arguing for the second.
Plenty of businesses are still at that starting line. The US Census Bureau’s Business Trends and Outlook Survey put overall business AI use between 17% and 20% from December 2025 to early May 2026, according to its May 2026 report. Firms with at least 250 employees were at 37%, and use among firms with fewer than 20 employees didn’t change significantly over that period. So if you haven’t started yet, you’re in the majority. Picking well is what matters.
Score each process on five questions
List ten or so tasks your team repeats every week. Then give each one a score from 1 (poor) to 5 (strong) on these questions:
- Volume. How often does it happen? Fifty times a day beats twice a month, because the savings repeat.
- Rules vs judgement. Could you write the steps on one page? The more the work follows clear rules, the higher the score. Heavy judgement or negotiation scores low.
- Data availability. Is the input already digital and in a system you can connect to, like a CRM, inbox or accounting tool? Paper forms and knowledge held in one person’s head score low.
- Risk. What happens if the automation gets one wrong? A misfiled lead is cheap to fix. A wrong payment or a wrong medical note isn’t. Lower risk earns a higher score.
- Measurable outcome. Can you count the result: minutes saved, errors caught, replies sent within the hour? If you can’t measure it, you can’t show it worked.
Add the five scores. Anything at 20 or above out of 25 is a strong first candidate in our view. Treat a 1 on risk as a veto whatever the total. And if two processes tie, pick the one a single team owns end to end, because fewer handoffs means fewer surprises.
This lines up with older research on where automation works. McKinsey’s 2016 analysis of work activities estimated the technical potential for automating data processing and data collection at over 60%, compared with 18% for applying expertise to decisions, planning or creative work, and 9% for managing and developing people.
Good first candidates and poor ones
Here’s how the scoring tends to play out for common small and mid-sized business tasks. Your numbers will differ, so treat this as a starting point, not a verdict.
| Process | Why it scores well or poorly | Verdict |
|---|---|---|
| Copying invoice details into accounting software | High volume, clear fields, digital input, easy to check | Good first candidate |
| Routing and tagging new website enquiries | Frequent, rule-based, low cost if one is misrouted | Good first candidate |
| Assembling a weekly sales or operations report | Same sources every week, time saved is easy to count | Good first candidate |
| Answering repeat customer FAQs with human handoff | Plenty of volume, answers exist in a knowledge base | Good, with review of edge cases |
| Setting prices or approving discounts | Judgement-heavy, mistakes cost real money | Poor first candidate |
| Handling complaints or sensitive customer cases | Needs empathy and discretion, high reputational risk | Poor first candidate |
| A process that changes every month | Rules aren’t stable enough to build on | Fix the process first |
Redesign the process before you automate it
Automating a messy process gets you a faster mess. McKinsey’s March 2025 State of AI survey tested 25 attributes and found that redesigning workflows had the biggest effect on whether organisations saw EBIT impact from gen AI. Only about 21% of respondents whose organisations used gen AI said they’d redesigned at least some workflows. The survey covered 1,491 participants and was fielded in July 2024.
In practice, walk the process with the people who do it. Cut steps that exist only because of an old system. Agree what “done” looks like. Then decide whether a simple rules-based workflow will do or whether you need an AI agent; our guide to AI agents vs workflow automation covers that choice.
How to measure the return
The same McKinsey survey found fewer than one in five respondents said their organisations tracked KPIs for gen AI solutions. That’s an easy gap to close if you plan for it on day one.
- Record a baseline for two to four weeks. Count how many times the task happens, how long each one takes, and how often it goes wrong.
- Pick one or two metrics only. Hours saved per week and error rate are usually enough. Response time works for customer-facing tasks.
- Count the full cost. Include build fees, software subscriptions, usage charges and the staff time spent checking outputs. Our AI agent cost guide breaks these down.
- Compare after 60 to 90 days. Hours saved multiplied by a loaded hourly cost, minus running costs, gives you a rough payback figure.
- Note what people did with the time. If the hours went into faster follow-ups or more sales calls, that’s the real story to tell.
Keep a person reviewing outputs for the first few weeks. It protects you while the system beds in, and the corrections tell you exactly where the rules need tightening.
Where to go from here
Our advice: pick one process, prove it, then expand. A single well-measured win does more for buy-in than five half-finished pilots. If you’d like help running the scoring with your team, our AI strategy consulting work starts with exactly this kind of opportunity mapping, and our AI automation services team can build what comes out on top. You can also book a strategy call to talk through your shortlist.
Frequently asked questions
What is the easiest business process to automate first?
Should small businesses automate at all?
How long before an automation shows a return?
Do I need AI, or is simple automation enough?
Sources
- U.S. Census Bureau: Large Firms With at Least 20 Employees Biggest AI Users (May 2026)
- McKinsey: The state of AI: How organizations are rewiring to capture value (March 2025)
- McKinsey Quarterly: Where machines could replace humans, and where they can’t (yet) (July 2016)