Personal AI You Can Trust
What are you not telling your AI? This week with Mark Suman from Maple on privacy, candour, and the work that actually moves a business.
What You're Not Telling Your AI.
Many people will say they care about privacy, and they mean it, but the day to day reality of running a business often pushes us toward whatever moves faster and gets the job done.
Customers need answers, staff need direction, cash has to be watched and problems have to be solved, so the business ends up using the tools that are easiest to reach, even when there is some discomfort about what those tools might be doing behind the scenes.
That is why many privacy arguments fail to land inside a small business.
They ask owners to care about privacy as a principle and to act as though an abstract concern should matter more than an immediate operational need, which is often a luxury that the day to day reality of running a small business doesn’t allow for.
AI changes that calculus because it moves the privacy question much closer to the centre of the business.
The problem with AI inside a small business is that the useful information is often the sensitive information.
These are the use cases that require client names, commercial terms, project history, contract details, employee information, customer records, pricing, margins, and the internal decisions that have not been made yet.
These are exactly the categories of information that businesses already work hard to restrict through permissions, contracts, confidentiality obligations, access controls, and internal processes.
That is what stood out in this week's conversation on The Good Stuff with Mark Suman from Maple.
Mark made the point that people have tolerated data extraction for years. They knew their email was being scanned, that social platforms were tracking them, and that the internet was gathering more than it admitted, but most carried on anyway because the trade-off felt distant enough to ignore.
AI brings that trade-off into the middle of your work.
A business owner might use it before they have decided what to do, trying to work out what to say to a client, how to handle a staff issue, whether a project is still profitable, or how to respond to a problem that has not yet been resolved. To be useful, the model needs the context around the decision, and that is where the privacy problem becomes more serious.
For small and medium sized businesses, this matters because a lot of the commercial value of AI becomes dependent on context.
A generic prompt can help with generic work, but the valuable use cases usually require information the business is already responsible for protecting.
You might be trying to understand why sales have slowed down, or whether a staff member is still right for a role, or whether the next hire is still financially feasible, or how to respond to a difficult customer without making the situation worse.
The more context you give the AI, the more useful the tool becomes, the better the reasoning can be, and the more specific the situation, the less generic the response.
This is where privacy becomes more than a defensive issue and becomes about performance and output.
When a business doesn’t feel safe being candid with its AI tools, sensitive information might leak, be stored, be used for training, or become subject to rules the business doesn’t control.
But you can also start editing or self censoring the context before the conversation begins, leaving out the detail that would actually change the quality of the output, like removing names, or describing the situation in a way that feels safer to share.
Then your AI gives an answer to that version of the problem rather than the real one.
This is why a watched AI is potentially a worse thinking partner. The model might be powerful, the interface smooth, the answer intelligent-sounding, but the person using it has become less truthful with it.
That is a limitation because the real promise of AI for SMEs is not that it can produce more words, but that it can help a small team reason, decide, document, automate, and execute with more leverage than they could before, and that promise depends on the business being able to bring the real problem into the room.
Many small businesses already trust big brands by default because everyone else uses them, and that kind of trust is mostly social proof.
The trust that makes AI genuinely useful in a business is something more specific. It’s what lets a business owner say the actual thing and bring the sensitive customer context, the awkward staff issue, the worrying cash position, or the uncomfortable strategic thought into the conversation without holding back.
If the safest version of an AI workflow is one where the owner never tells the tool anything important, then the business is leaving a lot of value on the table.
The tool can help with surface level work, drafting emails, summarising documents, generating first drafts of things where the details do not really matter.
The deeper work requires the details you’re withholding.
The most valuable questions are sometimes the ones we feel least comfortable asking in a place you don’t fully trust.
AI turns that into a product problem.
The best tools won’t simply be the ones with the strongest models, they’ll also be the ones that create the conditions for candour, whether that comes through stronger privacy, local control, open source infrastructure, or clearer boundaries around how memory and context are handled.
The exact implementation will vary but the principle is the same - the AI has to be somewhere the business can tell the truth.
This is easy to miss because so much of the AI conversation is still dominated by demos. A demo can ask a polished question because a demo is designed backwards from the answer. The context is clean, the data is fake, the problem is contained, and nothing serious is at stake.
That is why so many AI demos feel impressive and slightly unreal at the same time.
They show what the tool can do when the world has already been arranged to make the tool look good. A real business doesn’t work like that.
You’re not simply asking the AI to draft an email or summarise a document. They are trying to work out what the situation actually is before deciding what to do about it.
Last week, we talked about whether businesses are giving Anthropic, OpenAI, and Google an off switch on their business.
This week's conversation with Mark points to another version of the same underlying issue.
Even before a platform can switch you off, you may already be switching yourself off inside it, withholding the information that would make the tool genuinely useful, using AI only for the safe outer layer of the business, asking polished questions because the real ones feel too sensitive to type.
The work that actually moves the business forward requires that real context.
If you’re thinking about privacy in this context inside your business, we’d love to chat.
We got into this and more this week in the Big Episode 57 of The Good Stuff with Mark Suman.