This agency stopped typing into its CRM. Logged deals jumped from 40% to 95%.
Client: An agency
What if you could talk to your CRM in your own language — and it talked back?
That’s not the pitch. That’s the daily routine now. No new dashboard. No new discipline. The reps just started talking.
Here’s the before:
A lead writes on WhatsApp. Silence for days. A prospect asks for a meeting — it never reaches a calendar. A great call ends, and the note that should hit the CRM never does. On paper, every rep owns their pipeline. In reality? Entries land late — or never. Follow-ups slip for weeks.
The agency was losing deals it had already won. Not to competitors. To friction.
Third time’s the promise
This is Metatron 3.0 — and the number matters. Version one was a WhatsApp reader that fished sales opportunities out of group chats: useful, narrow, nowhere near an assistant. Version two bolted a JavaScript engine onto it, so code could take over everything GPT-3.5 couldn’t yet do — more capable on paper, but so clunky that nobody used it for anything that mattered. Both taught the same lesson: the hard part isn’t intelligence — it’s an assistant a salesperson actually uses.
The market didn’t have the answer either. The heavyweight agents out there — Hermes, OpenClaw — practically demanded a PhD just to install. Power you can’t hand to a person is power you don’t have.
The hinge wasn’t something we built — it was something that changed. Models crossed a threshold: agents finally had enough intelligence to understand what’s being asked of them and to try again when something doesn’t come out right the first time. That is precisely the limitation version two’s JavaScript engine was trying to engineer around — and couldn’t. Once the model itself could do it, everything the first two versions had fought for became simple.
So version three got one brutally simple bar: if a rep can’t use it in their first hour, it doesn’t ship.
What Metatron is
An always-on personal sales assistant that lives where the work already happens: WhatsApp and the company’s internal Matrix chat. Voice or text. You talk to it like a colleague:
“Remind me on September 8th to book a 1-on-1 with Mihail from BNI.”
“Add a note to opportunity X — wants to finish with the old supplier before switching.”
“Which opportunities haven’t I talked to in a while?”
It writes the CRM entry, sets the reminder, resurfaces the cold opportunities you forgot. And two rules keep it trustworthy: it’s locked to its own user — it never reads anyone else’s messages — and it does nothing on its own initiative. Every action starts with you; the only things it runs unattended are the ones you scheduled yourself, recurring or on a date you set.
Why a Zapier can’t do this
A rules engine moves data. It doesn’t know you.
Metatron runs three layers of memory: a database of every message it has ever received; live context that never shuts off, because it never shuts off; and memory files it writes and re-reads on its own, in its own folder — notes it takes about you, for itself. The longer you work with it, the more it works like you.
You can even give it a personality, and it keeps it — one of our colleagues runs his as Cthulhu, permanently grumpy. Still files every note perfectly.
A workflow follows steps. Metatron learns you.
The numbers — and where each one came from
None of these numbers came from willpower. Each has a mechanism behind it:
- Response rate doubled because of reminders. “Remind me tomorrow at 12 to talk to Ion Popescu” takes three seconds to say — so people actually say it. The follow-up stopped depending on memory.
- The biggest revenue lever was one question: “Which opportunities haven’t I talked to in a while?” Deals that don’t close instantly used to go cold in silence. Now they resurface on demand.
- Logging jumped to 95% because an entry costs ~5 seconds. Nobody skips a 5-second task. Everybody skips a 2-minute one.
- The multiplier under everything is voice. Dictating in Romanian runs 3–10× faster than typing, with 90%+ comprehension of completely natural speech — no robotic diction, no repeating yourself. If it misses something important, it asks.
The honest part
It can fumble a name that isn’t saved in WhatsApp contacts, and dictating an email address full of odd characters gets clunky. We’d rather tell you that than pretend.
Launched at 3 AM. Mistaken for a colleague by 9.
The launch happened at 3 AM, while everyone slept. The assistant arrived in the team chat under a human name and avatar — Vasile. The first rep to come online assumed someone had hired a new colleague overnight, and confronted him, confused about what was going on. Vasile politely explained what he was. Minutes later, the rep was throwing tasks at him just to see what the new “colleague” could do.
The lesson
The technology wasn’t the surprise. The people were. Nobody hesitated. One of the first reps to get it put it best: “I love playing with it — I already have it write most of my messages.” The moment reps could offload the work they hated — in three seconds, out loud — they did it instantly. The demand was always there. It was just waiting for a door.
Steal this: the 5-Second Audit
You don’t need Metatron to use the biggest lesson in this story. Run it on your own team this week:
Metatron is opening to a small group of early adopters — the people who’ll help us sharpen it fast. If you want a CRM that finally listens, claim your seat.
→ Talk to your pipeline. Let it answer back.
