I was a guest on Granger Fawson’s Stronger Business Together podcast, and the episode went out this week as Stop Bolting On AI. Granger has worked in lean and continuous improvement since 1994, and I came to AI through process improvement and operations, so it’s no surprise that two lean people talking about AI ended up talking mostly about process.
You can watch the full episode on YouTube. These are the points I’d want a business owner to take from it.
Business hasn’t changed
The technology has changed, and it gives us a huge amount of power, probably more than the internet did when it arrived. But business principles didn’t change when the internet came in. What changed was how we delivered the business and how we dealt with people.
AI is the same. The businesses I work with still need clear processes, defined roles and a way of measuring whether things are working. AI doesn’t replace any of that. It’s a means to an end, and it works best on top of a business that already knows how it runs.
Most of a process doesn’t need AI
This was the point I most wanted to make. Only a small part of most business processes needs judgement, perhaps 10 or 20%, and that’s a rough figure rather than a measured one. The rest are straightforward steps that can run on ordinary automation.
Whatever you can mechanise without AI, you take a variable out of the equation, because you’re not introducing judgement where none is needed. The output becomes more consistent, and consistency is one of the main things automation should give a small business.
I was in a meeting recently with someone describing all the AI agents they had running, one doing this and another doing that. My thought was that half of those jobs didn’t need an agent. And where one does, the question is whether the right guardrails are in place: is it leaking information, is it looking at the wrong thing, is it doing the wrong thing? The moment you give a system autonomy, you have to put guardrails around it and monitor it.
Granger put this in lean terms, quoting Toyota’s principle that you have to earn the right to automate. Simplify the process, halve it, halve it again, and only then look at automating what’s left. Often you find you no longer need to.
AI adds a particular risk here. It’s so easy to tell an AI to do something that you can very quickly build in complexity that will cost you later.
Where I draw the line myself
Two rules I keep in my own business:
- AI drafts my emails, but I send them. My system puts a draft in my drafts folder and I read it, change what needs changing and send it. The only emails that go out automatically are fixed templates where just the name changes, and those don’t need AI at all.
- AI doesn’t write to my finance system. It can do the work of my year-end bookkeeping, but I enter the results through the accounting software myself. I could let it write directly, but there’s no good reason to add that risk.
Both are ordinary prudence about where a mistake would cost the most.
Why piecemeal AI produces inconsistent work
The hurdle Granger and I worked through was a hypothetical one, though I’ve seen versions of it many times: a 12-person marketing agency that has added AI in various places and is now getting noise and inconsistency in its output. They expected it to save time and make money, and it hasn’t.
The likely causes are mostly not about AI:
- No joined-up process. They have plenty of AI-powered tools, but they aren’t used in a connected way, so the output varies. Granger made the point that the same usually happened earlier with their software, picked one problem at a time with no overall plan, and AI has been bolted on top.
- Processes that aren’t written down. I’d guess very few small businesses have theirs documented.
- Not measuring. Every business should have a scorecard of the few measures that matter, covering sales and business development but also operations and delivery. Without it you can’t tell whether a change helped.
- Roles that aren’t clearly defined.
None of these are AI problems, and AI won’t fix them. If the underlying process is chaotic, AI just makes the chaos happen faster.
It’s a continuum
There isn’t one right destination for every business. It depends on the business’s history and especially its culture, and many staff are wary of AI because of everything they’ve read about it taking their jobs. In a business that size the opportunity is in elevating people rather than replacing them.
The stages I see run roughly like this:
- Experimentation, which most businesses have been through or are still in.
- Automated workflows, where some automation and some AI sit inside a known, defined workflow.
- Whole sections of the business running automatically, using a mix of AI and conventional, rules-based automation.
- AI agents with the autonomy to make decisions and act within the business.
- AI-first, where the business is designed from the ground up with AI in every area it can be.
You can get to a full system running the whole business, but where you start depends on where you are now.
Where to start
I’d always start with an opportunity scoping exercise. For a simple business that might be an hour’s conversation about how the work flows. For a larger one it could be a couple of days interviewing people to establish what the processes are.
Then look for the bottleneck. If you improve a step that isn’t the constraint, the process as a whole gets no faster. If you want to experiment, AI lets you iterate quickly, but keep the experiments away from anything your customers see.
Granger had a good suggestion for the mapping itself. Rather than starting with a prompt, start with a question: tell an AI what isn’t working and let it ask you questions back. I’d add that you can now talk a process through out loud and ask it to draw the diagram, then check whether that’s a fair reflection of what you said. Granger’s view was that six months ago it wasn’t much good at this, and now it makes the job a lot easier.
If this sounds familiar
If your business has picked up AI tools one at a time and the results are patchy, the fix usually starts with the process rather than another tool. I’m always happy to have that conversation, and I’ll start by trying to be useful.
My thanks to Granger for having me on. You can find more of his conversations with business owners at Stronger Business Together.