Most businesses probably want to bring AI into their operations but aren't sure where to begin. The instinct is to start with the tools and work backwards, but that's the wrong order. The tool is the last thing you should be worrying about. Before any of that, you need to know which part of your business is worth automating, and whether it should be automated at all. Here's the process I work through.

1. Identify a workflow

Start by listing the workflows in your business that are repetitive and resource consuming. Once you have that list, narrow it to the top three to five based on how much time and resource they eat up.

From there, look at risk. You want workflows that sit in the low to medium range, where a mistake is quick to fix and won't do any real damage. If an error could cost you a client or create a mess that takes days to untangle, it's not where you want to begin.

The next part is more subjective and comes down to your own judgement. Ask whether the workflow is still best handled by a human. Cold outreach is a good example of one that isn't. Sending high volume outreach to a large list is a sensible use of automation and AI, and I'd recommend it. Hot leads are a different story. People are far better at reading emotional nuance and building a real connection, so handing hot leads to AI doesn't make much sense. There's an argument for letting AI cover them outside working hours or when no one is available, but the rest is better left to a person.

2. Document the process

Once you've settled on a workflow, document it. This sounds simple, but it's usually the hardest part of the whole exercise, and that's exactly why it's worth doing.

When a business sits down to write out how it actually does something, a few realisations tend to surface. This is far more complicated than it needs to be. I don't know why we do it this way. There are mistakes in how we currently do this. We could be doing this better. I've worked with businesses that needed several attempts to get a single process down on paper, and it's an eye opening exercise every time. You get value from the workflow you'll eventually automate, and you tighten up the internal process around it before any development starts.

Documenting properly means more than writing down the steps. Note who is responsible for each part of the flow, and capture any rules or guardrails that apply. Consider whether there's business specific context or training tied to the workflow. All of this becomes important once the AI enhancement layer goes in. The more context and business specific rules you give the model, the less room there is for hallucination. An LLM will always work from generalities unless you tell it otherwise.

Make sure someone involved in the day to day running of the workflow is part of the documentation. They understand why things are done a certain way better than anyone else. Whether you build the solution internally or through a third party, that person should be one of the project sponsors.

3. Development

I'll keep this one high level for now. If you decide to build through a third party rather than in house, make sure it isn't a plug and pay service. Those have their place for small businesses with limited context and a limited budget, but they only take you so far. A solution that genuinely fits your business comes from partnering with a provider who takes the time to understand how it runs.

That's the groundwork for getting AI into a business. Part 2 will cover the post development and things to look out for.