AI pilots to production: private AI on your own data, built, tested and handed over
The fastest way to know whether AI helps your business is to try it on your own work. I build the pilot, review its security, hand it to your IT provider or the vendor to run, and oversee it.
- Price
- Fixed-fee, quoted after a free call.
- Typical timeline
- 3 to 8 weeks
Sound familiar?
- We want an assistant that knows our own documents, without sending them to a public chatbot.
- We tried an AI tool, liked it, and have no idea whether it is safe to rely on.
- Our team spends hours on the same multi-step task every week.
- Our IT provider has never built anything with AI and does not want to start with ours.
What is included
- A private assistant on your own documents: it answers from your files and policies instead of the open internet. This is what most people mean by "training AI on our data".
- A task-automation pilot (sometimes called an agentic workflow): AI that handles a multi-step task in your tools, such as sorting an inbox, drafting intake summaries or building a weekly report, with a person approving before anything goes out.
- The privacy tier that fits your data: the AI in Google Workspace or Microsoft 365 under your business terms, a business plan of ChatGPT, Claude or Gemini whose terms state your data is not used for training, which I read and verify, or open-source models on private hardware or a private cloud tenant.
- A security review before anyone relies on it: who can reach what data, what leaves the building, what is logged, and how to switch it off.
- The production handoff to your IT provider or the vendor, with written runbooks and training for the staff who use it.
- Oversight for the first weeks in production. After that, ongoing AI oversight is part of the fractional IT director service.
What you receive at the end
- One working AI pilot on your own data, tested on your real work.
- A written security review and a recommendation to expand, change or stop the pilot.
- Runbooks your IT provider or vendor can run it from, and staff who know how to use it.
Who it suits
- Owners who want real time savings from AI and need to see it work on their own data first.
- Businesses with an AI policy in place and one routine task worth automating.
- Teams that need extra care with client or patient information.
What you need to provide
- An AI use policy, or the AI strategy and safe use engagement first.
- Admin access to your Google Workspace or Microsoft 365, or time with the person who has it.
- An IT provider or vendor willing to run the pilot once it is handed over.
- Sample documents or a walkthrough of the task, and an hour a week from the person who knows that work.
What happens after
The pilot goes to whoever runs your systems. I stay close:
- Your IT provider or the vendor runs it, from the runbooks. I never run production myself.
- I oversee it for the first weeks, then ongoing AI oversight continues through the fractional IT director service.
- If the pilot does not earn its keep, the written review says so, and you stop with nothing locked in.
Where this work happens
Pilots are built and tested remotely, by me, on your own data. The kickoff, the security review walkthrough and staff training can happen at your office in Carson City or Reno, or over video anywhere in the US.
Related services
The $1,500 technology assessment is the usual starting point, so the advice is grounded in facts. This service pairs well with AI strategy and safe use and Fractional IT director.
AI pilots: common questions
You do not do hands-on work. Why is AI different?
Because the fastest way to know whether AI helps your business is to try it on your own data, and AI pilots are new enough that your IT provider may not have built one yet. I build the pilot, test it on your work and review its security, then hand it over with documentation. I do not run it for you.
Who runs it after you hand it over?
Your IT provider, or the vendor of the tool it runs on. They get written runbooks and a walkthrough, and your staff get training. I oversee it for the first weeks, and after that through the fractional IT director service if you want. I am never the operator.
What does "your data is private" actually mean?
Three things I set up and check, not a slogan: your documents are not used to train a public model under the terms of the plan you are on, which I read and verify; access is limited to the people and systems you choose; and you can see what leaves the building. How that is done depends on the tier: business terms in your suite, a business plan whose terms say your data is not used for training, or open-source models on hardware or a cloud tenant you control. I cannot control what a vendor does, and nobody can promise nothing ever goes wrong, which is why the security review and the off switch come first.
Why open-source models?
For the most sensitive work, such as client files or patient records, an open-source model on private hardware or a private cloud tenant keeps the data on systems you control instead of sending it to an AI vendor. It costs more to set up and the models can be less capable, so it is the right tier for some work, not all of it.
What does "training AI on our own data" actually mean?
For a small business it almost never means building a new AI model. It means giving an assistant safe, private access to your documents so it answers from them. That is usually quicker, less costly and easier to control.
Talk it through first
AI pilots to production is fixed-fee, quoted after a free call. The $1,500 assessment is the usual starting point, so the advice is grounded in facts.