Analysing & summarising
Ask open-ended questions across your whole roster, get plain-language rundowns and check-in summaries, and have the assistant leave a suggested reply as a draft for you to send yourself - all from chat.
Your dashboard has fixed views; your questions aren’t fixed. The agent shines at the ones a built-in report never anticipated - “who’s drifting?,” “compare this quarter to last,” “what changed since her last check-in?” - and at turning a screen full of widgets into a sentence you can act on.
A rundown of your day
The single best way to start a coaching morning - one question, your whole roster triaged:
• Sarah K. - messaged about knee pain after Tuesday's session, no reply yet.
• Marcus T. - hasn't logged a workout in 11 days (was 5×/week).
• Priya N. - submitted her weekly check-in an hour ago; weight down 0.6kg, adherence 95%.
Want me to dig into any of these?
It reads the same conversations, logs, and check-ins you would - it just reads all of them at once and tells you where to look.
Analyse the roster
Ask the cross-cutting questions no single screen answers:
The numbers are Protocol’s, not the AI’s guesswork. For anything countable - adherence, streaks, trends, comparisons - Protocol does the calculation and hands the assistant the result, so the figures match your dashboard. The assistant’s job is to find and explain, not to do mental arithmetic over your data.
Summarise a check-in & write a suggested reply
Two steps that pair naturally - read the latest, then put words on the response:
Here's a reply you could send her:
"Great consistency this week 💪 For the knee, let's switch to box squats to a comfortable depth and ease off the depth - film me a set so I can check the descent. How's it feeling today?"
I've left it as a draft in Sarah's conversation in Protocol. I can't send it myself, so open her chat, adjust it if you like, and send it when you're happy.
Notice the line it doesn’t cross: it leaves the reply as a draft in the client’s conversation (with read + write access) for you to send yourself - an agent key has no way to send a client a message at any scope. Open Sarah’s conversation in Protocol, tweak the draft, then send it. If you already had a half-written draft there, the assistant won’t replace it without asking. On a read-only connection it writes the reply in your assistant instead, for you to paste. The word chat on this page always means your Claude or ChatGPT window, never Protocol’s own Chat, which is you and your clients and has no AI in it. See Safety, scopes & privacy.
Good questions to try
- “Which clients are due a program refresh - same plan more than 6 weeks?”
- “Show me everyone with a check-in submitted but not yet reviewed.”
- “Compare Mike’s main lifts this quarter to last.”
- “Who’s coming up on a renewal in the next two weeks?”
- “Write a check-in nudge for the clients who’ve gone quiet.” (it leaves each nudge as a draft in that client’s conversation; you review and send them yourself)
Reading happens in your assistant; the dashboard stays the source of truth
The agent is a fast lens over your data, not a replacement for it. When you want to see a client’s full history, a program laid out, or a billing record, the assistant will point you to it - tap through to the dashboard for the full picture and any precise edits.
Next: scopes, the hard limits, and your data - Safety, scopes & privacy →