Short answer: turn the vendor feature on first when it does a real job inside software you already pay for. You need something else only when the work you care about crosses systems — or when the new button talks to customers with no one in your office reviewing it.
This is the honest competitor to a managed setup. Not another chatbot company. The software you already run, with a new label on an old screen.
What the vendor button can see
One system. That is the design. Clio sees Clio. Jobber sees Jobber. AppFolio sees AppFolio. CDK sees the DMS.
A lot of useful work lives there: a matter summary from the file you already keep in the practice system, a job history from the field-service record, a resident's open work orders. If that is the job, use the feature. Do not pay a second vendor to re-ask a question your software can already answer.
What the button cannot see, unless the vendor built a separate product for it: the email where the manufacturer changed the part date, the text from the technician, the call the office missed, the PDF that never got attached. That is where things still slip.
The failure mode that looks like progress
The dangerous version is not "the AI is bad at summaries." It is "the AI now emails my customers" from inside a system that only knows part of the story, with no approval step, in a voice that is not yours.
Native features are built to keep you in that product. They are not built to protect your staff from another thing to babysit, or to keep a human in front of every outbound promise. Some vendors do this carefully. Some shipped a bot because the roadmap said so. You find out which by testing it on a real customer record, not on the demo.
A simple test
Pick one recent mess. A missed callback. A part that did not arrive. A client who asked a question you had already answered in email.
- Ask the vendor AI to reconstruct it.
- Ask a person who was there whether the answer is complete.
- Notice what it invented, skipped, or could not see.
If the answer is good enough for that job, you may be done. If the missing piece is always in a different system, a different inbox, or a phone call, the button will not grow into that. It will only get more confident inside its own walls.
The questions to ask your current vendor
- What does this feature actually read — this product, or our other tools too?
- Can it send to a customer without a person approving?
- Does our data train their model?
- If we turn it off, what of our work product disappears?
Software names on our industry pages are context for how businesses already run. They are not a claim that Normal is plugged into each one. Access is confirmed during scoping, or it is not claimed.
A realistic first step
Use the feature you already pay for for one week, on one workflow, with a person reviewing everything. Then decide. Some owners will stop there, and they should. Some will find the same gaps they had before the button existed: information across tools, no one internal to set it up, and no appetite for a bot that talks to customers alone.
That second group is who we work with. If you want to talk through which group you are in — start here. If the honest answer is "stay with what you have," we will say so.
Related: Where does a small business actually start with AI? · Your staff uses ChatGPT. That is not an AI system..