Updated October 10, 2026
Quick answer: Your customers' data goes wherever the vendor's setup sends it, so ask first. Check four things: what data the vendor collects and why, whether it trains the AI model, which other companies (subprocessors) can see it, and how quickly you can get it deleted. A vendor that cannot answer clearly in writing has not mapped its own data flow.
You're about to let an AI agent read your customer texts, appointment notes, or payment history — and you have no privacy officer to ask what happens to that information after it's read. Here are four plain-English questions to ask before you connect anything, so you know where your customers' data goes and who else can see it.
AI agent data privacy for a small business comes down to four checks: does the vendor say plainly what data it collects and why, is your data ever used to train the underlying AI model, which other companies (called subprocessors) can also see it, and can you get it deleted on a timeline you can actually explain to a customer who asks. Most small businesses never ask these questions before connecting a tool, not because they don't care, but because nobody hands them the checklist. Ask a vendor these four questions before you connect anything, not after.
If you run a small business, your customer list is probably your most valuable asset that isn't cash in the bank — names, phone numbers, appointment history, maybe payment details or health and financial information depending on what you do. You didn't need a privacy policy to run your business before AI tools showed up; you just didn't hand that list to software that reads and reasons over it. Once an AI agent is drafting replies to customers or triaging their messages, it's touching that same information, and the question of where it goes stops being theoretical.
Americans are not relaxed about this. In a 2023 Pew Research Center survey, 70% of U.S. adults who had heard of artificial intelligence said they have little to no trust in companies to make responsible decisions about how they use AI in their products.1 That's not a fringe opinion — it's most people. Your customers likely feel the same way about your business using AI on their information, whether or not they ever say so out loud, and your own hesitation about it is worth taking seriously rather than brushing aside.
You don't need a law degree or a compliance background to ask these. You need to know what to ask a vendor, or your automation partner, before you say yes.
This is the principle of data minimization — put simply, only collecting and keeping the information actually needed to do the job, instead of hoarding everything "just in case" it's useful later. An agent that answers appointment texts needs the appointment thread and maybe a phone number. It doesn't need your full customer database, your payment processor login, or years of old email history sitting around in its system. If a vendor's answer to "what do you collect" is a vague shrug or a 40-page policy nobody can summarize in two sentences, that's the finding, not a formality to skip past.
Some AI products feed customer conversations and business data back into the model to improve it — sometimes for everyone using that model, not just your account — unless you specifically turn that off. Others keep every customer's data walled off, used only to run that customer's own account. Neither approach is automatically wrong, but you should know which one you're getting, because "improving the model" can mean your customer's phone number or medical note becomes part of data other companies' agents are trained on. Ask directly, in writing, and get a straight answer.
Almost no AI vendor runs everything on its own servers with no outside help. A subprocessor is another company the vendor relies on behind the scenes — for example, the cloud host storing the data, or a separate AI model provider actually generating the agent's replies. A responsible vendor can name theirs and tell you where those companies are located, since that affects which privacy laws apply. If a vendor can't produce this list, they likely haven't mapped their own data flow carefully enough to control it.
Ask directly: if a customer asks you to delete their information, or you stop using the tool entirely, what actually happens to what's already stored? A responsible setup can delete a specific customer's data within a defined window, typically days, and confirm it in writing. A vague answer — "it eventually ages out" — usually means nobody built a real deletion process, which becomes your problem the day a customer, or a regulator, asks.
Take the free 2-minute readiness assessment
Say a small business connects an AI agent to draft replies to incoming customer messages, which include names, phone numbers, and sometimes payment questions. Connected carelessly, the agent might run on a plan where every conversation gets used to retrain the vendor's general-purpose model, with no visible opt-out and no list of which other companies touch that data along the way. Connected with the four checks above, the same business picks a plan where customer data stays isolated to their account, gets a named list of the two subprocessors involved (a cloud host and an AI model provider), and can request full deletion within seven days. Eight months later, a customer asks what happens to their information if they stop being a customer. In the first setup, nobody at the business can honestly answer that question. In the second, it's a two-sentence explanation, because someone asked upfront.
Yes, and you shouldn't let anyone wave it away as overcaution. You're not a tech company, you didn't sign up to become a data-privacy expert, and the fact that AI tools are moving faster than most small businesses can vet them is a real problem, not a personal failing on your part. The honest answer is that no vendor can promise a system with zero risk; what you can actually check is whether they'll answer the four questions above clearly and in writing, before you hand over anything. As one small business owner might put it, describing a realistic but illustrative scenario: "I asked our automation company point-blank whether our client data trains their AI for other customers. They said no, walked me through exactly where the data lives, and put it in the agreement. That's the moment I actually trusted the thing." If a vendor dodges that question or can't put the answer in writing, that's more useful information than anything in their sales deck — a reason to keep asking, not a reason to swear off AI tools altogether.
We don't sell your information, and we don't train AI models on it. We also say plainly where that has limits: our voice provider can keep part of the call audio and transcripts to improve its speech recognition. Every action your AI Coordinator and its assistants take is recorded, and the details are on our privacy page. That matters for trust, but it also matters for growth: businesses that can explain their data handling in one sentence to a nervous customer keep that customer and often earn a referral out of the conversation. Businesses that can't tend to lose the customer the moment the question gets asked — and word travels fast in a small local market.
If something isn't working for your business, contact us. We'll work with you until it works for your business.
Want to see what it would take off your plate? Run the hours-back calculator to see how many hours of office work would turn into minutes of talking, or set up your Coordinator. Every plan includes a 1:1 setup call, booked right after checkout.
1. Pew Research Center, "How Americans View Data Privacy," October 18, 2023 (survey of 5,101 U.S. adults, May 2023; finding covers respondents who had heard of AI and their trust in companies' use of it, not small-business AI vendors specifically).
The full pillar guide this article belongs to.
The access-and-behavior side of the same problem — what an agent is allowed to do, not just what it sees.
Once data handling is sorted, here's how to keep the agent's actions in bounds too.
The oversight side of the same problem — catching mistakes before a customer sees them.
Start here if you're not yet sure what you'd even be checking data privacy for.
It depends entirely on the vendor's setup, and you should never have to guess. Some AI tools use customer conversations and business data to improve their underlying model for every customer, unless you specifically opt out. Others keep your data walled off and use it only to run your account. Ask directly: "Is our data ever used to train your model, and can we turn that off?" A vendor that can't answer plainly is telling you something.
Security is about who can access the agent and what it's allowed to do — the guardrails around its actions. Data privacy is about what happens to the actual information that passes through it: where it's stored, whether it's used to train a model, which other companies (subprocessors) can see it, and how you get it deleted. A tool can be tightly secured and still handle your data in ways you wouldn't sign off on if you knew about them.
Ask what data it collects and why, whether your data is used to train the underlying AI model, which subprocessors can also see it, and how fast you can have it deleted. Get the answers in writing before you connect anything.
A subprocessor is another company that can also see or handle your data on the vendor's behalf, such as the provider of the underlying AI model or a voice service. You should know who they are before customer data flows to them.
No. According to the guide, Unmanually does not sell your information or train AI models on it, says plainly what its voice provider keeps, and records every action it takes.
Take our free 2-minute readiness assessment — it walks through what to automate first at your business, and what deserves a closer look at how your data would be handled.
Take the 2-min readiness assessment