Own your company's context, and your own messaging channel

AI is only as good as what it knows about your company, and a lot of that gets said over text. Here is how we keep both in our own hands, with Quinn, our AI teammate, as the example.

Quinn drafts a text. Then it texts one of us to ask if it can send it.

Quinn is our AI teammate. It has its own phone line, and that line is on iMessage, so the people we talk to see a normal conversation from a number that stays the same. We built the whole thing ourselves. No texting vendor sits in the middle.

Context is what your company knows

An AI model knows nothing about your business until you hand it something. What you hand it is the context. That's your customer history, your open quotes, your call notes, how you price and what you promised someone last spring.

In your company, that's probably scattered. Some of it is in the CRM. Some is in the quoting tool. A lot of it is in text threads on people's personal phones, and it leaves when they do.

Try this. Ask yourself where the full history of your best customer lives. If the answer is a few tools and one salesperson's phone, that's all your AI has to work with too.

Software vendors have figured this out. Each one now sells an AI helper that reads the slice of your business living in their tool. You pay rent on your own memory, one slice at a time.

We'd rather hold it in one place. Ours is a CRM we host ourselves and a library of documents, both on a machine we own. Every agent we run reads from there. When Quinn drafts a reply, it has already read the whole thread, the notes on that person and the calendar.

Why the messaging channel matters

Texting is where a lot of that context gets made. It's where a customer tells you what they want, in their own words, late at night.

Most business texting runs through a vendor that rents you the number and charges by the message. Their rules decide what you can send, and your customers' replies live in their system.

So we built our own. Quinn's line runs on a computer we own, signed in to its own account. Our own code reads what comes in and sends what goes out.

How it works day to day

A reply comes in. Quinn reads it, along with everything we know about that person, and writes a draft. It texts the draft to one of us. We send it, fix it or skip it.

Nothing goes to a prospect until one of us says yes. That's slower than letting it run, and we've kept it that way on purpose.

A few rules ride along. A draft written late at night waits until morning. If the person isn't on iMessage, the same line falls back to a regular text. Every message gets logged on that person's record, so the context we started with keeps growing.

What went wrong

For about a week this summer, file attachments from Quinn's machine failed while plain texts went through. We never found the cause. It cleared up by itself, and we still send a link where we used to send a file.

We also sent a message in Quinn's voice from the wrong computer once. It delivered, from a personal number, into the wrong thread. Since then nothing goes out in Quinn's name from any machine but Quinn's.

With a vendor, the first problem would have been theirs to fix, and the second couldn't have happened. For some companies that's the better trade.

If you want your own line

We build the same setup for owners. You need a line, a machine that stays on, and somewhere for the records to live. We set those up, and they're yours. An AI teammate drafts, and a person on your team approves. Your customers' replies land in your own records.

Start with the context, though. Quinn's drafts are only good because of what it reads first.

Want to talk through a build for your company?

More articles