• Start with a real workflow, not a tool or a demo.

• A workflow is any recurring piece of work with a clear start and a clear end.

• Many workflows need simplification or better documentation before AI is even worth considering.

• Run a small, defined pilot with a named owner, a time limit, and a way to stop if it is not working.

• Measure net time saved, after checking, corrections, and maintenance, not the headline number.

If you run an established small business, you have probably been told a hundred times that you should be using AI. The advice usually arrives with a tool attached. Try this app. Book that demo. Add this subscription. So you end up with a folder full of half-tested products and very little to show for it.

There is a calmer way in. The sensible place to begin is a piece of work you already do. Where to start with AI in business is a question about your workflows, not about technology.

Let me walk you through how I think about this with clients, and how you can do the same on your own.

Why AI Projects So Often Begin in the Wrong Place

A tool-first approach quietly reverses the logic. You buy something clever, then you go looking for somewhere to use it. That search rarely ends well. You get scattered experiments, a growing pile of subscriptions, and very little day-to-day benefit. McKinsey describes the same trap as having “technology in search of a problem”, and recommends beginning with the business problem before developing the technology roadmap.

A workflow-first approach starts from the other end. You begin with a real problem: a delay, a cost, a bottleneck, a task that keeps landing back on your desk. Then you ask whether AI, automation, or something far simpler would help.

This matters because so many automation efforts come unstuck early. Research into the field found that a large share of automation projects fail because organisations automate a broken process instead of fixing it first. If the work itself is muddled, adding software just makes the muddle faster.

Key Point: Buying a tool before identifying the problem it solves is the most common and most expensive mistake in AI adoption.

What Counts as a Business Workflow

A workflow is a recognisable piece of work that moves from a clear starting point to an outcome. Nothing more complicated than that.

You already have dozens of them. Some common ones in a business your size:

• Preparing a quote and following it up

• Onboarding a new customer

• Handling an incoming enquiry

• Producing a weekly management brief

• Approving purchases

• Collecting information before an appointment

• Answering the same staff questions over and over

Each one has a beginning, a middle, and an end. That structure is what makes a workflow a good candidate for improvement. If you cannot describe the start and the finish, it is probably too vague to work on yet.

Key Point: If you can describe where a piece of work starts and where it ends, you have a workflow worth looking at.

Where to Look for Worthwhile Workflows

You do not need to review the whole business. Pick two or three workflows to start. The best first workflow is frequent enough to matter, reasonably stable, understood by the people doing it, and small enough to test safely.

When I sit down with an owner, these are the places I look first:

• Work that keeps waiting for you. Anything that stalls until you personally approve, sign, check, or reply.

• Information copied by hand between systems. The same details typed into the accounting package, then the job sheet, then an email.

• Repeated questions. Customers or staff asking the same thing week after week.

• Follow-ups that get forgotten. Quotes that go quiet, reviews never requested, that sort of thing.

• Reports pulled together by hand from three or four sources every Monday.

• Repeated drafting, checking, or sorting. Tasks where someone writes, reviews, or classifies the same kind of thing again and again.

• Processes where mistakes and rework are common, the ones the bookkeeper has to fix up again later.

None of these start with a tool. They start with friction you can already feel.

Key Point: The best workflows to improve are the ones already causing you grief, not the ones that sound impressive to automate.

How to Assess Two or Three Candidates

Once you have a shortlist, look at each one honestly. I use plain questions rather than a scoring system, because a workflow is a human thing and it resists being reduced to one perfect number.

Here is the table I work through with clients.

You are not trying to produce a mathematical result. You are trying to see which one or two workflows are worth the effort, safe to trial, and clear enough to succeed. A workflow that looks great on paper but is a tangle of exceptions is often a poor first choice. A smaller, tidier one usually teaches you more.

Key Point: A simple, stable workflow beats an impressive but messy one every time when it comes to picking your first pilot.

What Should Happen Before AI Comes Anywhere Near It

This is the part that gets skipped, and it is where money gets wasted. Some workflows are not ready for AI. They need something simpler and cheaper first.

The most common reasons automation efforts fall short are not technical. Studies of the field point to weak change management, poor training, and choosing the wrong process, well ahead of any software problem.

So before you reach for AI, ask whether the workflow first needs one of these:

• Simplification, because there are steps in there that made sense years ago and no longer do

• Clearer accountability, so everyone knows who owns each part

• Better documentation, so the process does not live only in one person’s head

• Better use of software you already pay for

It helps to see the full range of options and not jump straight to the fanciest one. There is a ladder here, from lightest touch to heaviest:

1. Remove the work entirely if it no longer earns its place

2. Simplify the workflow

3. Delegate it properly to the right person

4. Use existing software more effectively

5. Add conventional rule-based automation

6. Use AI to assist a person

7. Let AI take limited action inside clear controls

Most owners are surprised by how much can be solved on the first four rungs. If you would like a fuller picture of how these pieces fit together, I have written elsewhere about what an AI Operating System means for a small business.

Key Point: AI sits near the top of the options ladder. Work your way up, and you may find the right answer is two or three rungs lower.

Choosing a Safe First Pilot

Once you have picked a workflow and confirmed it is ready, keep the first trial small. A good pilot has a shape you can hold in your head.

• Narrow scope. One workflow, not five.

• A defined owner. One named person responsible for it.

• A clear start and outcome. You know when it begins and what finished looks like.

• Known information sources. You know exactly where the data comes from.

• Human review. A person checks the output before it goes anywhere sensitive.

• A way to handle exceptions. When an odd case shows up, there is a plan for it.

• Baseline measurements. You noted how long it took and how often it went wrong before you changed anything.

• A defined test period. Four to six weeks, not open-ended.

• A clear way to stop. You can switch it off and return to the old way without drama.

If you cannot explain the pilot on one page, it is too big to start well.

Key Point: A pilot you can explain in one page is a pilot you can actually run.

Measuring Whether the Change Actually Worked

Here is where honesty pays off. The number that matters is net time genuinely released, after you have accounted for checking, corrections, exception handling, and ongoing maintenance. A tool that saves an hour of typing but adds forty minutes of proofreading has not saved you what the brochure claimed.

Beyond time, useful measures include fewer interruptions landing on you, faster turnaround for the customer, fewer errors, and more consistent follow-up. Those last two often matter more to clients than raw speed. Digital automation has been reported to save some businesses meaningful hours each week, but the honest figure is always the net one, measured in your own business.

Key Point: Gross time saved is a sales number. Net time saved is the real one. Measure that.

A Couple of Worked Examples

Let me offer two hypothetical examples, purely to show the thinking. These are illustrations, not real clients.

Example one: quote follow-up. A builder sends plenty of quotes, but follow-up depends on whether anyone remembers. The workflow is frequent, the value is high, and the steps are clear. Before touching AI, the fix might be a simple reminder in the existing job software. If that is not enough, AI could draft a polite follow-up email for the owner to approve. Human review stays in place, and the pilot runs for a month.

Example two: the weekly management brief. An owner spends Sunday evening pulling figures from three systems into one summary. The information sources are known and the format rarely changes. This is a strong candidate. The first step is tidying where the data lives. Only then does automation, and perhaps AI to write the plain-English commentary, make sense.

What an AI and Automation Business Review Adds

You can do all of this yourself, and plenty of owners do. Where I come in is bringing an outside pair of eyes to the shortlist. An AI and Automation Business Review is a structured look across your business to find the two or three workflows worth improving, sort out what needs fixing first, and choose a safe pilot with proper measurement around it.

I have spent a long time on both sides of this. I still write code every day, and I have been advising small business owners for years. That combination means I can translate the tech into plain English and tell you honestly when a workflow is not ready, or when the right answer is no AI at all.

The One Question to Start With

If you take one thing from this, make it the opening question. Which piece of work is worth improving?

Answer that well, in two or three cases, and the choice of AI or automation becomes far easier. It comes later, once you know what you are actually trying to fix. That is what good small-business process improvement looks like, and it is a far calmer path than chasing the latest tool.

Frequently Asked Questions

Where should I start with AI in my business?

Start with a workflow that causes real friction, not with a tool. Pick two or three recurring pieces of work, assess them honestly, then decide whether AI, simpler automation, or better use of existing software is the right answer.

What is a workflow in this context?

A workflow is a recognisable piece of work with a clear start and a clear outcome. Preparing and following up a quote, onboarding a customer, or producing a weekly report are all workflows.

Do I need AI for every workflow?

No. Many workflows are best improved by removing unnecessary steps, clarifying who owns them, documenting them properly, or using software you already pay for. AI is one option among several, and often not the first one worth reaching for.

How do I know if a workflow is ready for automation?

It is ready when you can explain it on one page, the information it uses is clean and in one place, the exceptions are few, and you can test the change safely and undo it if needed.

How do I measure whether the change worked?

Measure the net time genuinely released after checking, corrections, and maintenance. Also watch for fewer interruptions, faster turnaround, fewer errors, and more consistent follow-up.

What is an AI and Automation Business Review?

It is a structured look across your business to identify the two or three workflows worth improving, work out what needs fixing before any technology is added, and set up a safe first pilot with proper measurement.

How many workflows should I look at to begin with?

Two or three is enough. Starting with more than that makes it hard to run a proper pilot or learn anything useful from the results.

What if my workflow is messy and hard to describe?

That is actually important information. A workflow you cannot describe clearly is not ready for AI. Fix the process first: simplify it, document it, and clarify who owns each step. Then reassess.

Key Takeaways

• Start with a workflow you already do, not a tool someone recommended.

• The best first workflow is frequent, stable, and small enough to test safely.

• Many workflows need simplification, clearer ownership, or better documentation before AI is worth considering.

• Run a small, defined pilot with a named owner, a time limit, and a clear way to stop.

• Measure net time saved, after all checking and corrections, not the headline number from a brochure.

• AI sits near the top of a ladder of options. Work your way up; you may find the right answer is simpler than you expected.

• The first question is always: which piece of work is worth improving? The tool comes after.

Talk It Over With Me

If you would like an outside view on where AI or automation could genuinely help, I would be glad to talk. Book an AI and Automation Business Review with me, Christine, at Gecko Gully, and we will find the workflows worth improving and a sensible place to begin.