talking to your crm

A while back I built a recording tool for quotes. You speak a client meeting into it and out rolls a draft quote, and that draft goes past a human before it’s allowed out the door. That’s all it is, really. And I didn’t start it for the speed. What I was after was what ended up landing in there differently.

Because every CRM rollout has that one moment. The rep steps out of a good client meeting, flips the laptop open, and there’s a row of empty fields just sitting there. The conversation’s still fresh in their head. Rich, full of signals. And then the retyping starts. What actually makes it into the system is three lines of telegram-speak. Sometimes nothing at all.

And the industry reacts the same way every single time. I know that list by heart. More training. More required fields. A dashboard that quietly keeps score of who’s letting their admin slide, with a manager on top of it every week. And the user gets the blame, obviously. Because the system works fine, right?

Thing is, that diagnosis doesn’t hold up. That rep is doing exactly what you’d expect a reasonable person to do. It’s the form that’s the culprit.

the form is the culprit

Filling in a form is a writing task. And a writing task always loses. To the next client meeting, the next email, the drive home in traffic. People aren’t lazy. Typing something up after a meeting just feels like having that same meeting all over again, in slow motion, at a screen that says nothing back.

So people cut corners. Short summaries. Fields filled in on a guess. Notes nobody can decipher a week later. Six months after go-live the verdict is that the CRM isn’t being adopted, and, sure enough, the next round of training is already in the calendar.

And none of this is new. The biggest bottleneck in any CRM implementation is adoption. I’ve watched enterprise outfits drop hundreds of thousands on an SAP or a Salesforce, and then the thing gets used at, what, ten percent. Five, even. People just carry on in their Excel. You literally have to force them into the other tool, because they don’t see what it adds. And honestly, they’re sitting there thinking, look, whether I type it into Excel or into some little form in a prettier setting, it’s data I’m doing something with either way, so where’s the win? In Excel I can make myself a nice little chart too. You get a lot of people who just hang on to what they’ve already got. Change is a human thing, and people aren’t great at it. Yeah. I’ve watched that one play out plenty.

But it was never about willingness to share. Ask that same rep what happened in the meeting and they’ll talk for five minutes straight, no effort at all. Everything you want in the system is sitting in those five minutes. The only question is how you get it out without a keyboard wedged in the middle.

speaking is three times faster

For anything longer, speaking runs about three times as fast as typing. That’s not a marginal gain. That’s a different category altogether. A meeting write-up that takes you fifteen minutes to type, you speak in five. Out in the parking lot, right after, while it’s all still fresh.

But the part that matters more is something I only really saw once that quote tool was actually running. Thinking out loud drags up different stuff than typing does. Someone who speaks tells you what happened. Someone who types boils it down to whatever they need to offload to get away from the screen. The first gives you usable client context. The second gives you compliance.

That’s why I call dictation an interface. Not a gadget. It changes what ends up in the system, not just how fast it gets there.

The usual objection is that something like this has to be expensive or complicated. That objection’s expired. Modern transcription knows your jargon and your client names, turns speech into proper flowing sentences, and runs on your own API key for less than a euro a month. Getting started is: speak it in, transcript, paste. That’s the whole thing.

And here’s where I tie it to the MCP protocol. If you’re running an agent-native CRM that allows it, one where you can set the permissions yourself, then you could just have your leads, your visits, your activities, whatever it is, all dropped into the system by voice, with the right status on them, without entering a single thing by hand. Because the agent, that’s just agentic AI, it enters all of it for you and asks you the right questions while it’s at it. There’s an intelligence layer sitting on top that goes, hey, what status should this get, don’t you want to set the categorization, or it just sets the categorization for you, because it already knows your context, your preferences, your processes. And of course you always get the confirmation back. And if it’s a money process, invoices or quotes, a model like that is going to raise a flag a lot sooner the second it feels unsure about something.

what it doesn’t solve

The keyboard isn’t going anywhere. A short chat message you tap out faster than you’d ever speak it, and a phone number or an amount, no, you don’t fix that with your voice. Dictation wins on the longer stretches of work and thinking. Meeting write-ups, brainstorms, long emails, functional designs. Not every single mouse click.

And it won’t repair a badly set-up CRM either. If your data model is off, now you’re just getting richer text in the same wrong structure. The order stands: first a process and a model that actually hold up, then you speed up the input. Do it backwards and all you’ve done is automate your own chaos, only in higher quality now. In that quote tool of mine, by the way, nothing ever goes straight out the door. Someone always looks at it first.

Inside those limits, though, something fundamental does move. For years we treated CRM adoption as a behavior problem. People had to learn to love forms. That never worked, and it never will. The more honest route is that the form stops being the interface at all. The person talks, the way they were doing anyway after every good client meeting. The machine does the structuring.

And watch this, because this is the sneaky part. A project with barely any adoption actually looks, on the surface, like a really successful one. Because usually no problems come up. The users haven’t made a peep, since the interest was never there in the first place. They barely touch it, so nothing gets tested, and without testing you get no test results. So yeah, obviously, you’ve got yourself a beautifully implemented project. No problems, were there. That’s exactly the kind of success I want gone.

So the best CRM record isn’t some task still waiting to be done after the client contact. It’s a byproduct of talking about it. The day entering data stops feeling like admin, nobody has to enforce adoption anymore.