Names in the CRM. Appointments in the booking tool. Purchases in the point of sale. Conversations spread across email, text, and a phone system that records calls nobody replays. Reviews on three platforms, ad performance on two more. That’s a normal small business today, and none of that information is scarce. All of it is scattered. Almost none of it is available at the one moment it would matter: when a customer is on the phone, on the site, or standing at the counter.
What AI changed
A language model is only as useful as what it’s handed at the moment of the question. Ask it to help a returning customer and it needs that customer’s history. Ask it to answer the phone for a dental office and it needs today’s schedule, the insurance list, and the policy on emergencies. The model itself is becoming a commodity; every business will soon have access to roughly the same intelligence. What separates a useful AI system from a demo is whether the right piece of information reached it in time, with permission, in a form it could use.
So the bottleneck moved. It used to be storage: where do we keep all this? Then it was analysis: what does it mean? Now it’s delivery. The hard problem is getting the right data to the right agent at the right moment, and knowing it was allowed to be there.
The model is not the product. The route the information takes to reach it is.
How it works today, and why it breaks
Right now, the route is glue. A Zapier chain that copies a form submission into a spreadsheet. A CSV export every Monday. An integration that worked until the vendor changed something. A person who knows to check three systems before calling a customer back. We build routes like these for clients every week, and the honest truth about all of them is that they’re temporary. They break silently, they leak data into places it shouldn’t be, and nobody owns them, because they don’t belong to any one system.
Large companies solve this by hiring: a data team, an integration platform, a security review for every connection. A dental office, a gym, or a restaurant group can’t, and shouldn’t have to. They got websites without a web team. They got card payments without a banking relationship. They’ll get this without an IT department too, or they’ll be left with the glue.
Four things I think will be true in five years
- Data stays where it lives, but becomes addressable. The winning move won’t be one more place to copy everything into. It will be a way to ask for a piece of information by what it is rather than where it sits, and have it arrive.
- Permission travels with the data. Today access is decided at the door of each system. In a world where agents request information on a business’s behalf a thousand times a day, the rule about who can see what has to ride along with the record, not sit in a settings page somewhere.
- Agents request context; humans stop exporting it. The Monday CSV disappears. A phone concierge asks for the schedule. A follow-up system asks for the last conversation. A report asks for the quarter. The owner answers “who is allowed to ask” once, instead of “where is it” every time.
- Small businesses get this as a utility, not a project. Like DNS, like payments. Something you turn on, not something you build.
What to do about it now
You don’t have to bet on that future to protect yourself from it.
- Own your system of record. Your customer list should live somewhere you can export in full, today, without asking permission. A tool that won’t let you leave with your data is renting it to you.
- Choose tools with real APIs. Not “integrations available.” A documented way for other software to read and write. Every tool without one is a dead end for whatever you build next.
- Don’t let the AI vendor own the memory. The transcripts, summaries, and preferences an assistant learns about your customers are your data. Make sure they land in your system, not only theirs.
- Count the glue. Write down every export, every automation chain, every “I check three places first.” That list is your integration debt, and it’s what your next system will have to replace.
Where this is going for us
Most of what DigitDwell builds, from an AI phone concierge to a custom CRM to Atlas, comes down to the same job: move the right information to the right place at the right time, for a business that can’t hire a team to do it. That job keeps getting bigger. It’s the problem we think about most, and it’s where our next product is pointed. More on that when it’s ready.
Working on the same problem?
If your business runs on tools that don’t talk to each other, that’s the conversation we like most. No pitch, thirty minutes.