How to Choose a Private AI Partner

1. Why the choice matters

Your data holds your pricing, your clients, and your history. The system built on it will make recommendations you act on. The partner you choose decides where that data lives, how the numbers are produced, and who answers when something is wrong six months from now.

2. Three ways to run AI privately

• On your own machine. Models run on hardware in your office. Data stays in the building unless you send it out.
• In a private cloud. A dedicated environment run for you, isolated from other customers.
• A business tier of a public AI. Paid business plans from the major assistants do not train on your inputs. For drafting and general questions, this may be all you need.

Match the setup to the data. Client files and pricing rules belong in the first two.

3. Ten questions to ask any partner

1. Where does my data live, and who can access it?
2. Does any of my data go to a third-party model? When, and with whose permission?
3. Are the numbers computed in code, or written by the model?
4. Can every figure be traced to its source?
5. When data is missing, does the system show a blank or fill in a guess?
6. Who owns the software, and who owns my data?
7. Who is accountable for correctness six months after launch?
8. How is the system monitored without exposing my data?
9. If I leave, what do I take with me, and in what format?
10. Can I speak with a client and see measured results?

4. Types of providers and when each fits

• Hardware builders sell workstations and servers. They fit if you have staff to run models yourself.
• Managed service providers keep your systems running. They fit for uptime, security, and support. Ask who builds the AI.
• AI consultancies and development shops build to a specification. They fit defined projects with an internal owner afterward.
• Platform vendors sell one product that many customers share. They fit when your process matches theirs.
• Public AI business tiers are general assistants. They fit drafting, summarizing, and general questions.

5. Red flags

• A promise that the system makes you compliant with a regulation.
• No references, or results that are all projections.
• No answer to "where did this number come from?"
• A system that never shows a blank.
• No plan for who maintains it after launch.

6. Disclosure: how we answer

Dr. Data is one of these options. We build private systems for small businesses and run them after launch. Our answers:

1. On your own machine or in a dedicated cloud environment. Who can access it is set per build and written into your agreement.
2. Only with your authorization, for drafting text. We tell you which builds use a cloud model.
3. In code. The model writes the text around the numbers.
4. Yes. Every figure carries its source.
5. A blank, labeled as missing.
6. We license the software to you. Your data stays yours.
7. We are, on a Managed Intelligence plan. One-time builds include a defined support period.
8. On Managed Intelligence plans, through a monthly report of what was checked and what changed.
9. Your data belongs to you. How you take it with you is written into your agreement.
10. Yes, with clients who have agreed to take calls. Ask us.