People ask me what I actually do. Which is fair, because "AI consulting" can mean anything from building custom machine learning models to selling chatbot subscriptions. What I do is neither of those things.
Here's what a typical engagement looks like, week by week. Not the polished version. The real one.
Before We Start: The Conversation
Everything begins with a conversation. Usually 30 to 45 minutes. No agenda, no slide deck. I ask what your business does, how things flow on a normal day, and where things get stuck. You talk, I listen.
What I'm listening for isn't what you think your AI needs are. Most people come in with something specific: "I want a chatbot" or "I need to automate my marketing." Sometimes that's right. Often the bigger opportunity is somewhere they haven't looked.
I'm listening for patterns. Repeated phrases like "that takes forever" or "it depends on who's doing it" or "we keep meaning to fix that." Those are the signals.
By the end of the conversation, I have a rough sense of whether AI would actually help, and where. Sometimes the answer is "you don't need AI, you need a better process" or "you need a part-time admin, not a robot." I'll say that. I'd rather be honest and lose the engagement than build something that doesn't help.
Week One: Discovery
If we move forward, the first week is all observation and interviews. I talk to the people who do the actual work, not just the owner. The person who processes invoices. The person who manages the inbox. The person who onboards new clients.
These conversations are revealing. Owners know the strategy. Employees know the friction. The gap between those two perspectives is usually where the best opportunities hide.
I also look at the tools you're already using. Your CRM, your project management system, your email setup, your spreadsheets. What talks to what. Where data moves manually that could move automatically. Where information gets re-entered because two systems don't connect.
By the end of week one, I have a map. Not a polished diagram, just a clear picture of how work actually flows through your business, and where it gets bottlenecked, duplicated, or dropped.
Week Two: Recommendations and Selection
This is the week where I come back with a prioritized list. Here are the five or six opportunities I found, ranked by impact and effort. Here's what each one would save in time. Here's what each one would cost to implement. Here's the risk profile.
We usually pick two or three to start with. Occasionally just one if it's complex enough to warrant full attention.
The selection conversation matters. I'll tell you what I think is the best starting point, but you know your business. Maybe the highest-impact automation is in a workflow that's about to change anyway. Maybe there's a lower-impact one that would get your team excited and build momentum. Context matters, and you have context I don't.
We also talk about tools. Which AI model, which automation platform, which integrations. I try to use tools you're already paying for when possible. Adding a new subscription for every automation gets expensive fast and creates its own maintenance burden.
Week Three: Build
This is where things get built. Depending on the complexity, "build" might mean configuring a Zapier workflow, writing and testing prompts, setting up integrations between your existing tools, or building a simple internal tool.
I work in short cycles. Build something small. Test it with real data. Show it to the person who'll be using it. Get feedback. Adjust. Test again.
The "show it to the person who'll be using it" step is non-negotiable. I've watched too many consultants build things in isolation and then present a finished product that doesn't fit how people actually work. By the time someone sees it, it should already be shaped by their input.
For automations, I build in monitoring from the start. Logs that track what the automation did, what it decided, and what it produced. This makes debugging easier and gives the team visibility into what's happening behind the scenes.
Week Four: Test and Refine
The automation runs in parallel with the manual process for a week. The person who used to do the task manually still does it, and we compare results. Does the automation produce the same quality? Does it handle edge cases? Does it break in any obvious ways?
This is where most of the refinement happens. Prompts get adjusted. Rules get tightened. Edge cases get handled. The gap between "works in testing" and "works in real life" is always bigger than expected, and this week is for closing that gap.
I also watch for the human side. Is the team member comfortable with the automation? Do they trust the output? Are they checking it appropriately, not too much (defeating the purpose) and not too little (missing errors)?
Week Five: Launch and Handoff
The manual process stops. The automation takes over. For the first week of full operation, I'm available for quick questions and adjustments. Things always come up that nobody anticipated. A new type of email the automation hasn't seen before. A tool update that changes an API response. A client with a name that breaks the data parsing.
I also do the handoff documentation. Not a 30-page manual. A short guide that covers what the automation does, how to tell if it's working, what to do if it stops working, and how to adjust the most common settings. Written for the person who'll be maintaining it, not for a technical audience.
After: The Check-In
I schedule a check-in for 30 days after launch. By then, the automation has been running long enough to reveal any persistent issues. We look at the data: how much time is it actually saving? Is the output quality holding up? Has anything drifted?
Usually there are a few adjustments. A prompt that worked fine for the first 100 uses starts producing slightly different output and needs tuning. An integration that was stable starts throwing occasional errors. Normal maintenance stuff.
We also talk about what's next. Now that one or two automations are running, the team has a better sense of what else could be automated. The opportunity scan we did in week one often gets revisited with fresh eyes.
What It Costs
I charge for the engagement, not the tools. The tools are yours. If we stop working together, your automations keep running. You're not paying a monthly retainer for something to keep functioning.
Typical engagements run four to six weeks for a small business. The cost varies with complexity, but for context, we're talking about a fraction of what you'd spend on a part-time employee for a year, for time savings that persist indefinitely.
I'm not going to put specific numbers here because every engagement is different, and I'd rather have an honest conversation about scope than anchor to a number that might not apply. If you're curious, let's talk. I'll give you a straight answer.
What I Don't Do
I don't build custom AI models. I don't develop software from scratch. I don't sell tools or take commissions from vendors. I don't do engagements that last six months, because a small business AI project shouldn't take six months.
I find the friction, build the automations, make sure they work, teach your team to maintain them, and move on. That's the job.
If you've read this far and it sounds like what you need, you can learn more about our Flow Rebuild service or just reach out directly. Either way, the first conversation is free and comes with zero obligation.