Agencies keep trying to move clients into their project management tool, and clients keep not going. The client portal gets three logins, ever. Meanwhile the real project — requests, approvals, "quick questions", scope changes — happens in WhatsApp and Telegram, because that's where your client already spends their day. They moved; the tooling didn't.
So stop fighting it. The problem was never chat — it's that chat and the board don't talk to each other. Here's the setup that fixes it.
Rule 1: the chat is the source of requests, the board is the source of truth
Everything the client asks lands in chat. Everything the team executes lives on the board. The failure mode is the gap between them: a request that stays in chat and never becomes a ticket, or a ticket that changes status and nobody tells the client. Whatever tooling you use has to bridge that gap in both directions — chat messages become board actions, board changes become chat answers.
Rule 2: not every message deserves a human
Sort the traffic. Most client messages fall into three buckets:
- Status questions — answerable from the board, automatically.
- New requests — need a scope check before anyone says "sure".
- Judgment calls — pricing, conflict, anything emotional. Humans only.
A team member reading every message to decide which bucket it's in is the bottleneck. The sorting itself is what you automate first.
Notice the "stayed quiet — addressed to a teammate" markers: knowing when not to answer is half the job. A bot that jumps into every group message gets muted within a week.
Rule 3: guard the scope at the message
The most expensive sentence in agency work is "sure, quick fix" typed into a chat. In a group thread it feels harmless; on the invoice it never appears. Every new ask should be checked against the agreed scope of work the moment it arrives — that check is a tool's job, not a discipline problem, because no human re-reads the SOW at 9:41pm.
Rule 4: human-in-the-loop for anything that can go wrong
Auto-send is fine for "here's the status of your tasks". For anything else, the right pattern is: the AI drafts, the reply waits with a dashed border, a person approves. The client experiences fast, consistent communication; your team experiences control.
The setup, concretely
One number/handle per client (or a group per project), connected to the projects on your board. Every conversation attached to a project. An AI project manager reads the thread, answers what it can prove from the board, flags what's out of scope, relays real questions to the right teammate, and stays quiet when the message wasn't for it.
That's exactly what Gavril does: it lives in your clients' WhatsApp and Telegram, speaks their language, and turns chat chaos into board actions — without asking the client to change a single habit.
← All posts