AI for client service teams

Clients don't see your workload, they see response times. Your assistant keeps every request owned and deadlined, and makes handovers painless because the context lives on the card, not in someone's head.

What you'll end up with

Every client request logged as an owned card with a deadline the day it arrives.
Per-client visibility: ask "where are we with Acme?" and get a real answer.
Handover summaries written from the actual thread history, not from memory.

Before you start

Sortd set up in Gmail — a Client Work board with lists per stage (say Requested, In progress, With client, Delivered), and a client tag or custom field on cards.
Your assistant connected — two minutes, one browser approval: Connect your AI. Everything below works in Claude, ChatGPT, or any connected assistant.
The right permissions — when you connect you tick what the assistant may do — send permissions start unticked. For this guide: reading email, reading and updating the board, and preparing drafts (drafts never send by themselves).

Set it up by talking to your assistant

Each step is something you type (or say) to your assistant — copy them as they are, or change the names to match your own boards and lists. You don't need to know what happens behind the scenes, but we explain it anyway.

1. Log requests as they land

A request that isn't on the board doesn't exist — fix that first.

Check for new client emails. Make each request a card on Client Work / Requested, tag it with the client's name, set a due date three working days out unless they asked for a date, and flag anything marked urgent.

What happens: Requests become deadlined cards; the due-date view now shows this week's real commitments.

2. The account status question

The question every account manager gets asked on the way to a meeting.

What's open for Acme right now? Group by status and tell me which items are close to their deadline or past it.

What happens: The assistant reads the board filtered to that client and answers in plain language — thirty seconds before the call, not thirty minutes.

3. Chase what's with the client

Half of "late" work is actually waiting on the client — visibly chasing it is the fix.

For cards in With client where they haven't replied in 4 days, draft polite nudges referencing what we're waiting for. Leave them for my review.

What happens: Each draft quotes the actual open question from the thread — clients get specific nudges, not generic ones.

4. Handover without the meeting

When someone's off or a client moves desks, the context should move itself.

Jo is away next week. For each of Jo's open cards, add a note summarising where it stands and what the next step is, then reassign them to Alex.

What happens: Summaries are written from each card's thread and notes, then ownership moves — Alex starts informed, not archaeologising.

Put it on autopilot

Once a run works when you ask for it, schedule it — a daily morning run keeps deadlines and chases on track.

Every weekday at 8:30, log new client requests onto Client Work with client tags and due dates, list anything due today or overdue, and draft nudges for cards in With client quiet for 4+ days.

Claude (Cowork scheduled tasks) and ChatGPT (scheduled tasks) can both do this — cadences, tips and more recipes are on AI Schedule.

You stay in control

Drafts by default. Unless you granted a send permission, the assistant can only prepare replies — they wait in Gmail for a human to send.
Everything is visible. Every card the assistant creates or moves shows up on the board instantly, exactly as if a teammate did it — same statuses, same history, same audit trail.
Easy to undo, easy to unplug. Moves and statuses can be changed back on the board, and you can disconnect the assistant at any time from the same place you connected it.

More: AI for project management · every action the assistant can take, with example prompts: AI Reference · all use cases: AI guides