Field NotesAI Strategy
What an AI consultant actually does.
By Rasheed · · 6 min read
For a small business, an AI consultant does four things: shows you what today's models can actually do on your own work, sets up the access and data controls so you can use them safely, builds only what's still missing, and teaches you to keep improving on your own. Here's what that looks like in practice.
What owners expect, and what happens instead
The owners I talk to are almost always overwhelmed. Many of them tried a base model of ChatGPT or Claude months ago, weren't impressed, and moved on. They tested one part of their business, didn't like the result, and decided AI wasn't there yet.
What they're missing is that performance can vary a lot depending on what you're trying to solve or automate. Choosing between model types, and controlling how much effort the model spends based on the task at hand, is new to most people. And every new GPT or Claude release brings a whole new set of capabilities, research potential and automation with it, so something that fell flat a few months ago may work now.
So early in an engagement, I'll often take something an owner asked ChatGPT months ago and wrote off, ask it again myself with the parameters adjusted, and read through the output with them.
Data security is the first real conversation
Owners are also very concerned about data security and leaks. In my view, the picture is better than most of them assume. Anthropic and OpenAI run enormous amounts of compute and guard direct access to the most advanced models, and I'd argue that gives them some of the best cybersecurity of any tech companies. As healthcare, legal and financial institutions have moved onto these platforms, controlling data and access (the harness around the model) has become a huge priority for both, and the controls have improved a lot.
The job is to use those controls properly for your business.
What that looked like at CAMCO
CAMCO, a commercial real estate investment group, is a good example from start to finish.
The managing partners were highly concerned about feeding financially sensitive data into Claude. So we restructured their Claude accounts onto Claude Team, and I explained how that would help with data retention and training.
Then came the real bottleneck. Only the managing partners could open the full financial memoranda, so when the other partners had questions, the managing partners answered them by hand. That could take hours, sometimes days.
I architected a separate SharePoint site they could share with the partners, who could then give their own Claude accounts access through the Microsoft 365 connector. Then I built an Azure Document Intelligence pipeline to take the memoranda, scrub them, and place the scrubbed copies in that shared site.
Most of my time between visits went into one loop. I started with an initial batch and reviewed the results with the managing partners, improved the pipeline, and tested again. After a few rounds, they were comfortable that the data they didn't want shared wasn't being transferred, and that the data the partners needed was accurate.
Last, I added Microsoft Purview, so the scrubbing, the transfers and the partners' access are all logged and visible to the managing partners.
Building, teaching, and deciding what not to do
Some problems do need a custom build, like the report pipeline behind The Property Inspection Pros. But the most important part of any engagement comes at the beginning: giving the owner a coherent view of what's available right now and how well it actually works.
More often than not, after a little tweaking of their Claude configuration and access controls, they start getting results they're happy with, and we end the engagement right there. Helping an owner decide what not to build is part of the job.
What you keep when I leave
One of my key motivations is passing on how to better approach and work with AI, and how to keep on top of the latest developments. You don't need to be aware of everything all the time. Even I can't do that. But I try to leave owners with the tools to keep improving their processes over time, themselves.
What you should own yourself
Two things owners often assume an AI consultant handles, which I think they should own:
- New prompts every week or month. The whole point is that I teach you to do this on your own.
- New architecture and custom builds. Often, organizing your data and configuring a harness for agents to work on that data visibly and securely is all you need.
If you want to see how we structure this work, start with our services: AI Solutions is where the building happens, and AI Enablement is the teaching. If you want to chat directly, book a free 15-minute call. If you're in or near Pasadena, book an in-person Office Hours session for you and your team.
