What should your business automate first? A ranked method

Automate the highest-volume, lowest-judgment, back-office work first. The boring repetitive tasks. Score every candidate process on two axes, effort to automate and monthly payback, then do the low-effort, high-payback quadrant before anything else. Do not start with the flashy customer-facing thing. That is the mistake that kills most projects in year one.

I have run this exercise dozens of times, on our own group and on the businesses we audit. The answer almost never changes. The process you are excited about is rarely the one you should build first.

You are not behind, so stop letting that drive the decision

Most small businesses are not behind on AI. As of May 2026, U.S. Census Bureau BTOS data showed firms with 250 or more employees using AI at 37%, mid-size firms of 100 to 249 people at 32%, and firms under 20 employees at under 20%. Adoption climbs with headcount, not with ambition.

The pressure to catch up is real, but it pushes owners toward the wrong first move. A Thryv survey of 540 SMB decision-makers (published 2025) found small-business AI adoption rose from 39% in 2024 to 55% in 2025. Fast growth. It does not tell you those businesses picked well.

Here is what the fear does: it makes you reach for the visible win, the customer-facing chatbot or the marketing generator, because that is what you can show people. Visibility is not the same as payback. Pick the process that quietly saves hours every week, and let the results argue for you later.

Why does the flashy first project usually fail?

The flashy first project fails because it is hard to integrate and hard to measure. MIT’s Project NANDA study “The GenAI Divide” (August 2025) found 95% of enterprise generative-AI pilots delivered no measurable P&L return. The cause was not weak models. It was poor workflow integration and choosing the wrong process to automate.

That last finding is the backbone of everything I am about to say. The report was clear that back-office automation produced the highest returns, while the front-office pilots that get all the attention mostly stalled. The model was rarely the bottleneck. The choice of where to point it was.

The abandonment numbers back this up. Gartner projected (July 2024) that at least 30% of generative-AI projects would be abandoned after proof-of-concept by the end of 2025. Then in June 2025 it went further: over 40% of agentic-AI projects will be canceled by the end of 2027, driven by unclear business value and escalating cost.

Read those two together. Unclear business value is the killer. A boring back-office process has a clear, countable value: the hours it eats today. A customer-facing experiment has a value you have to argue about in a meeting. One of those survives the budget review. The other does not.

What should you automate first, concretely?

Automate the work that is high-volume, rules-based, and low-judgment. Invoice processing, data entry, order confirmations, HR onboarding paperwork, report generation, inbox triage, reconciliation. These are the tasks nobody enjoys and everybody does the same way every time. That sameness is exactly what makes them safe to hand to a machine.

There is a reason these show up first on every serious list. McKinsey’s “The State of AI” (2025) reported that 88% of organizations now use AI regularly in at least one function. By function the leaders were marketing and sales at 42%, product development at 38%, and service operations at 35%. Notice the gap: most usage is still clustered where the demos are, not where the quiet savings live.

That gap is the opportunity. If everyone is crowding into the marketing use case, the underpriced win is the invoice queue that three people touch every morning. Boring is a feature. Boring means the task is stable, the output is checkable, and the payback lands on a line you already track.

The test I use is simple. Would the task change if a customer had a bad day? If yes, it needs judgment, so it waits. If no, it is a candidate for first.

The ranked-list method: score effort against payback

Score every candidate process on two axes only. Effort to automate on one, from low to high. Monthly payback in hours or dollars on the other, from low to high. Plot each process. You will get four quadrants, and you do the low-effort, high-payback quadrant first. All of it, before you touch anything else.

This is the whole method. It stays on one page. I have never needed a fancier framework, and the businesses that reach for one usually do it to justify a project they already decided to build.

Effort to automate covers three things: how structured the inputs are, how many systems the task touches, and how tolerant it is of a wrong answer. Structured inputs, one or two systems, and a checkable output mean low effort. Free-form inputs, six integrations, and a mistake that reaches a customer mean high effort. Be honest here. Optimism about effort is how six-week builds become six-month ones.

Payback is easier. Count the hours the task consumes across everyone who touches it, every month. Multiply by loaded cost. That number is your payback, and it is the number that survives the budget conversation later.

Here is how the common first candidates tend to rank:

Process	Effort to automate	Monthly payback	Do it first?
Invoice and AP data entry	Low	High	Yes, start here
HR onboarding paperwork	Low	Medium-High	Yes
Report and dashboard generation	Low	Medium	Yes
Inbox triage and routing	Medium	Medium	Soon
Order and booking confirmations	Low	Medium	Yes
Customer-facing chat/support bot	High	Uncertain	Wait
Marketing content generation	Medium	Uncertain	Wait

The pattern is not subtle. The top of the list is unglamorous and countable. The bottom is exciting and hard to measure. Work the top down.

One caution from doing this repeatedly: the “uncertain” payback rows are not bad ideas. They are second-year ideas. You build credibility and internal data on the low-effort wins, and then the hard projects get funded because you have a track record instead of a hope.

What does “start with back-office” look like in practice?

At scale it looks like a stack of unglamorous processes, done one after another. A Deloitte case study on a UiPath deployment (2025) described a large firm that automated more than 600 processes for over 4 million cumulative labor hours saved. The headline examples were invoicing, HR onboarding, and benefits administration. Not a chatbot in sight.

You will not automate 600 processes as a small business, and you should not try. The lesson is the sequence, not the scale. That firm did not begin with a customer-facing showpiece. It began with invoicing, the most countable process it had, and compounded from there.

We run the same pattern on our own group. The agent fleet that runs our own businesses started with back-office queues, the reconciliations and confirmations and reports that ran the same way every day, and grew outward from proven wins. If you want the specifics, our record of agent runs behind the group carries the numbers so this post does not have to.

The through-line across the MIT, Gartner, and Deloitte evidence is one idea. The teams that won did not have better models. They pointed ordinary automation at the right first process. If you want a deeper walkthrough of how we pick that process on a client engagement, the breakdown of what an operations audit produces covers the scoring in detail, and the case for an outside team versus building it in-house covers who should do the work.

Where this leaves you

The first project decides the second one. Pick a boring back-office process with high volume and low judgment, score it against the effort-versus-payback grid, and ship the low-effort, high-payback quadrant before you let anyone talk you into the chatbot. That is how you land on the right side of the 95% that MIT measured in 2025.

If you want a second pair of eyes on your list, that is exactly what our operations audit does: we score your processes on the same two axes and hand you a ranked build order you own.

Start async. Reply through the start page with the two or three processes eating the most hours in your week, and I will come back within one business day with a first read on which one to build first. No meetings, no call to book. Just send the list, and we go from there.