Where to start with AI in your business.
Most leaders who say they want AI are stuck at the same place. They know they want it, they have no idea where to start. Here is the answer I keep landing on, across every engagement I run.
Founders, CEOs, and operations leaders keep landing in the same spot when they bring me in. They know they want AI in their business. They have no idea where to start. The conversation that gets the most attention right now is the cold-feet version. A CEO who was gung-ho on AI for several months hears a horror story (an expert testifying with hallucinated facts, a lawyer fined for citing made-up cases, a chatbot quoting the wrong policy on a recorded call) and freezes. The fear is rational, the freeze is rational, and the freeze rarely thaws on its own.
The underlying problem, though, is not unique to the leaders who got spooked. Almost everyone who is curious about AI is stuck on the same question. The cold-feet version is just the loudest. Every other variant lives in the same place: overwhelm at the number of tools, fatigue at the volume of vendor pitches, uncertainty about which use cases are real and which are demos, doubt about whether the team is ready to use any of it well.
Here is the answer I keep landing on, across every engagement I run.
The short version
Start with AI in the supporting layer of your operation. The deliverable itself, the parts of your work where a wrong fact is case-ending or trust-ending, stays with the humans who own it. The work that surrounds the deliverable, the intake, the scheduling, the summarization, the internal knowledge retrieval, the draft correspondence, the time tracking, all the work that humans do because someone has to do it and not because they were trained for it specifically, is where AI earns its place first.
The same answer fits the legal expert-witness firm whose CEO heard the hallucination horror story, the multi-location restaurant group thinking through their first AI pilot, the boutique investment fund, the umbrella manufacturer, the small architecture practice. Different industries, same starting place.
What I’m actually doing inside it
Across engagements, the supporting-layer work breaks into four shapes. Most operations have all four, in different proportions.
- Intake and triage. New client forms, prospect inquiries, support emails, vendor requests, the inbound that requires reading, sorting, routing, and acknowledging before the actual work begins. AI handles the reading and sorting reliably, drafts the acknowledgment template, and surfaces the items that actually need a human eye. The hour a day that used to go to this work goes back into the operation.
- Summarization of internal material. Meeting recordings, long email threads, prior client conversations, old proposals, anything where someone on the team needs the substance of something that already exists in writing somewhere. AI compresses cleanly and the human reads the summary, makes the call, and moves on. The forty-minute hunt for the right context becomes a four-minute read.
- Knowledge retrieval across your own archive. The question that comes up in every meeting: did we do something like this before, and if so, where is it. Before AI, that question cost half an hour of digging or got skipped entirely. AI handles the search in seconds and surfaces the artifact, leaving the human to decide what to do with what comes back.
- Drafts that humans finish. Internal updates, status notes, client follow-ups, the standard correspondence that takes time disproportionate to its impact. AI drafts, the human edits and adds the relationship texture only they have, and the draft goes out the door. The drafting cost goes down while the thinking and the relationship work stay exactly where they belong.
None of those four touch your deliverable. None of them touch the parts of your work where a wrong fact is case-ending or trust-ending. They sit on top of your operation. They free up time that goes back into the high-stakes work where the humans belong.
What this isn’t
A few framings this gets confused with.
- This isn’t “AI for the small stuff only.” The supporting-layer work tends to be where most of your team’s day goes, even though nobody designed the team around it. Naming it as the entry point is naming the largest reclaimable block of time in your operation.
- This isn’t a permanent ceiling. Once the supporting layer is working and your team has built real pattern-recognition for what AI does well and where it falls apart, the next round of work can push closer to the deliverable with appropriate guardrails. Starting in the supporting layer is the on-ramp. Where you end up depends on what your team learns.
- This isn’t avoidance.Some leaders read this as “we will only let AI touch the safe stuff, never the hard problems.” The hard problems are real and worth solving. Sequencing matters. The supporting-layer work is the right starting point because it carries the least risk and gives the team time to build pattern-recognition for what AI does well, which is the foundation for taking on harder problems later.
- This isn’t a productivity pitch. The lead value is keeping the stakes low while your team builds pattern-recognition for what AI does well. The productivity gains are real, and they show up as the follow-on benefit after the trust is built.
How to know if you need to start here
A few signals that the supporting-layer entry point is the right starting place.
- Your team is doing repetitive operational work that someone has to do but nobody on staff is specifically trained for. Intake, scheduling, knowledge management, internal correspondence.
- You have an instinct that AI should help you and you cannot picture exactly where. The instinct is correct, and the “where” is the layer on top of your operation. The work that goes out the door under your name comes later.
- Someone in your leadership has been spooked by an AI failure story in your industry, or they have not been spooked yet and you would like to keep it that way. Either way, starting in the layer where the deliverable is not at risk is the right move.
What to do next
If this sounds like your situation and you want to know which version of it fits your operation, take the AI readiness assessment. It is a 12-minute diagnostic that returns one of five archetypes plus a starting place specific to where your operation actually is. Most companies that recognize themselves in the description above come out as “The Entry Point.” Some come out as “The Stalled Stack” (already paying for tools that nobody is using) or “The Coordination Sprawl” (work fragmented across too many places). Each archetype routes to a different starting place.
Here’s what I do
I run fractional AI Operations engagements out of Portland for small businesses without a tech executive on staff. The work always starts in the supporting layer. Two weeks of scoping, then a build that ships in eight to twelve weeks, then an optional monthly partnership for clients who want one. The first build typically returns its cost in staff hours saved within the quarter. The sequencing of starting on top and earning the right to push closer to the deliverable is also described in What I’m actually building, when I build and AI consultant vs AI agency vs fractional AI Operations.