Before the AI

Assisted by AI

The messy middle was about what to fix before you try AI. This one is about how. Not a framework. Not a methodology. Just the practical steps that actually move things forward.

Start with where the time goes

Before you map a process or clean a dataset, figure out what’s actually eating hours.

Pick one team. One department. One week. Track where the time goes. Not in a time-tracking tool - just a shared spreadsheet or a notepad. What did people actually spend their day on? Not what they were supposed to spend it on. What they actually did.

Most businesses have never done this. They guess. And they guess wrong.

The insight is rarely the big obvious thing. It’s not the major project that took too long. It’s the twenty small things nobody counts: the re-keying of data from one system to another, the email chain that should have been a form, the report assembled manually every Monday morning from three different sources, the meeting that exists because information doesn’t flow any other way.

Those twenty small things are where you start.

Map the process before you fix it

Take the one process that showed up worst in the time audit. The one where everyone said “yeah, that’s painful.”

Now map it. Not in a flowcharting tool. Not yet. Just describe what actually happens, step by step, from trigger to completion. Who does what, in what order, using what systems. Use AI to help with this part. Record yourself doing the process - screen recording, voice memo, video on your phone. Walk through it out loud as you go. Then hand the recording to AI and ask it to generate the first-pass documentation.

It won’t be perfect. It doesn’t need to be. You need a starting point - something to react to and correct. That’s faster than writing from scratch, and it’s faster than staring at a blank document trying to remember every step.

Same principle applies when you’re shadowing a team member to understand their workflow. Record the conversation. Let AI structure the notes. You focus on asking good questions instead of trying to capture everything.

The tool you’re getting ready for can help you get ready.

Audit your data

Where does your customer data actually live? Your job data? Your financial data? List every system, every spreadsheet, every shared drive, every inbox where business information accumulates.

Now ask three questions about each source: What’s duplicated across systems? What’s contradictory between them? What’s missing entirely?

You don’t need a data warehouse. You don’t need a data strategy. You need to know what you have and where it is.

The ugly spreadsheet that maps “system → what data → how current → who owns it” is more valuable than any data strategy document. It’s not impressive. Nobody’s going to frame it. But it tells you exactly where the problems are and who can fix them.

Fix one thing

Don’t try to fix everything. That’s how these projects stall - the audit reveals forty problems, someone tries to solve all of them, and six months later nothing’s changed.

Pick the highest-impact, lowest-effort problem from what you’ve found. One thing.

Common first wins: connect two systems that should already talk to each other. Template a document that gets rewritten from scratch every time. Automate a reminder that currently relies on someone remembering.

These are often not AI projects. They’re spreadsheet projects. Automation projects. “Just use the feature your existing software already has” projects. The solution to a painful manual process is frequently a built-in feature that nobody set up.

The win isn’t the fix itself. It’s proving the pattern works. You audited, you identified, you fixed, and things got better. That builds momentum for the next one.

Now try AI on something small

With one process mapped and one data problem fixed, you have enough foundation to test AI meaningfully.

Pick something contained: draft a document using the template you built, summarise the data you cleaned, generate the report you’ve been assembling manually every week.

The difference between this and “just try ChatGPT” is that you’re giving AI structured inputs instead of asking it to work from nothing. You’ve got a documented process, so AI knows the steps. You’ve got clean data, so the output makes sense. You’ve got a template, so there’s a format to follow.

That’s why the boring work matters. Not because AI demands perfection, but because structured inputs produce useful outputs.

The easy ways AI helps right now

Here’s the part that surprises people… You don’t have to wait until everything’s clean and documented to start using AI. Some of the best early uses are the readiness work itself.

• Record a meeting and have AI pull out the action items and decisions. Not a transcript - the structured output that actually matters.

• Photograph a whiteboard process map and have AI turn it into a clean flowchart.

• Dump a messy spreadsheet in and ask it to spot duplicates, inconsistencies, and gaps.

• Dictate how a process works while you’re doing it and have AI draft the standard operating procedure.

None of these require clean data or mapped processes as prerequisites. They’re bootstrapping tools. They help you build the foundations faster. The irony is worth sitting with: the best first use of AI in most businesses isn’t a business application. It’s using AI to get ready for AI.

Make it a habit

Getting ready isn’t a project with a start date and an end date. It’s a habit. Audit, map, fix, repeat. Each cycle gets faster because you’ve built the muscle and you understand your own operations better.

The businesses that’ll get the most from AI in two years are the ones building these foundations now. Not because AI requires perfection, but because AI amplifies whatever you give it - and you want to be amplifying the right things.