Chris M.Updated
Private AI vs public chatbots: where does your project data actually go?
It depends on the tier. On a public consumer account your input may be retained and may be used to improve the provider's models. On an enterprise tier there are contractual commitments not to train on your data. On a private deployment the data never leaves your environment at all. Most construction businesses are using the first tier for work that belongs in the second or third.
Somewhere in your business this week someone pasted part of a subcontract, a client email or a tender figure into a free chatbot. Not maliciously. They were trying to get their work done faster, and the tool helped.
What are the three tiers, and what is each one for?
| Tier | What it is | Safe for | Not safe for |
|---|---|---|---|
| Public and free | Consumer accounts signed up with a personal email | General questions about how a standard-form mechanism works | Margins, client information, pricing, personal data, anything commercially sensitive |
| Enterprise | Paid business versions with admin controls, audit logs and contractual data commitments | Day-to-day drafting, summarising and analysis of your own material | Material a client's confidentiality obligations forbid leaving your environment |
| Private deployment | The model runs on your own server or cloud tenancy, data never leaves | Assistants over your project record: contracts, correspondence, commercial positions | Nothing, but it costs more to stand up and maintain |
UK government guidance draws exactly this line for its own staff: nothing sensitive goes into public tools unless it is already public.
The uncomfortable observation is that plenty of firms are paying for personal-tier convenience with company data, because nobody set up the enterprise account. That is not a policy failure. It is a procurement gap that policy then gets blamed for.
Open-source models have made the third tier genuinely affordable, which is why it is now the foundation for private commercial intelligence over a project record rather than something only large organisations can consider.
Which questions actually matter when choosing?
Five. You do not need to become a security specialist, and you do not need to read a vendor's whitepaper.
- Is our data used to train the provider's models, and is that commitment contractual rather than a statement on a webpage?
- Where is it processed and stored, and for how long is it retained?
- Who inside our business can see what others have put in?
- Can we switch it off, export our data, and prove what was shared?
- Does this arrangement survive our clients' confidentiality obligations and the UK GDPR?
If a vendor cannot answer those in writing, that is your answer. The willingness to put it in writing is the test, not the content of the answer. On the last question in particular, the ICO's guidance for organisations is the reference point rather than any vendor's own summary of it, and its data sharing guidance covers the case that actually applies on a project: personal data moving between parties who each have their own obligations.
Do bans work?
No, and the failure mode is predictable enough to plan around.
The instinctive response to all this is to ban the lot. What follows is shadow use: people quietly using personal accounts on personal phones, with zero visibility and zero control. The information leaks anyway, and now you cannot see it happening, which is strictly worse than where you started. This is why controlled adoption beats a blanket ban in practice as well as in principle.
The working alternative is boring. Give people a sanctioned tool that is genuinely good, on an enterprise or private tier. Publish red lines everyone can remember: client data, pricing and personal data stay out of public tools, no exceptions. Train by role, so the QS knows what is safe with a contract and the bid team knows what is safe with a tender, which is the step most rollouts skip. Then route the genuinely sensitive work to a private deployment, where the question stops being a judgement call.
Your people are already keen, which is the best adoption problem a firm can have. Give that enthusiasm a sanctioned home, on tools you control and under rules you wrote, and the energy your team is currently spending in the shadows becomes the fastest capability upgrade available to a construction business today.