A compliance officer at a mid-size financial advice firm called me last month. Her firm had been quoted a small fortune for a bespoke AI build to summarise client meeting notes. She was not sure it was worth it.
I asked what tools they already had. She mentioned they bought enterprise licences for ChatGPT a few months ago, but the team rarely used them. Twenty minutes later, we had a working prompt that did exactly what the expensive build was supposed to do. It was not a custom application. It was just a clear set of instructions. It solved a problem she thought needed ten weeks in ten minutes.
This happens constantly. A client brings a problem. They assume the solution must be complex because the problem is painful. But the scale of a problem does not dictate the scale of the solution.
When I look at any operational bottleneck, I classify it into one of three levels.
Level 1 is education. You already pay for the tool that fixes your problem. You just do not know how to use it yet. This applies to writing emails, drafting reports, and extracting data from standard documents. If you have Microsoft Copilot, ChatGPT, Claude, or Gemini, you have the engine. You only need the steering wheel. That means a template, a clear prompt, and fifteen minutes of training. You do not need to hire a developer.
Level 2 is integration. You have two systems that refuse to talk to each other. Your CRM holds the client data, but your back-office system needs it for billing. Your team spends Friday afternoons copying and pasting between the two. This is where automation platforms step in. Tools like Zapier, n8n, or Make can connect these systems. There is no bespoke code written from scratch. It is configuration work. It usually takes a few days to build and test. Once it is running, your team gets their Fridays back.
Level 3 is a custom build. This is real engineering. You have a complex process that requires reading unstructured documents, applying compliance logic, and making probabilistic decisions before a human reviews the output. No off-the-shelf tool does this safely. You need orchestration, API connections, and strict guardrails. This takes weeks to build properly.
The issue in the consulting industry right now is incentives. A large agency wants every problem to be Level 3. An expensive custom project pays for their overhead. A twenty-minute prompt does not. So they sell the build.
I do the opposite. I always try to solve the problem at Level 1 first.
If a prompt works, we use the prompt. If the prompt fails because the workflow spans multiple systems, we move to Level 2 and build an integration. Only when integration fails do we look at a custom build. This discipline saves money, but it does something much more valuable. It builds your internal capability. When we solve a problem at Level 1, your team learns how to use the tools they already have. They stop seeing AI as a magic box built by expensive outsiders. They start seeing it as a tool they control.
A team that knows how to write a good prompt will start automating their own minor annoyances. A paraplanner who learns how to format a report using an AI assist will soon figure out how to draft meeting agendas the same way. You do not just get a single problem fixed. You get a team that knows how to fix the next one.
My job is to make your business run smoother. Sometimes that means writing code. Most of the time, it means showing you how to use what you already have.
The honest first answer is usually the simplest one.