AI Summaries and Follow-Ups: What Gets Automated and What Shouldn't
The right line between automation and human judgment in client communication. Which follow-ups should fire automatically, and where a real message matters more than a fast one.
Automation in client communication is a balance. Too little, and you're manually scheduling follow-ups that should be systematic. Too much, and clients can tell — and the relationship cost is higher than the time savings are worth.
What should be automated
The cleanest candidates for automation are communications that are expected, predictable, and low-stakes enough that a templated message is actually appropriate:
- Post-signature onboarding welcome — expected, time-sensitive, consistent across clients
- Intake form requests — mechanical, same every time, no personalisation needed
- Weekly status update prompts to internal team — internal, systematic
- Post-delivery check-in at 2 weeks — expected, same structure each time
- 6-week renewal nudge — systematic, not personal
- Overdue invoice reminders — administrative, not relational
These are all cases where the message is essentially the same every time, the timing is predictable, and the recipient doesn't need it to feel personal — they just need to receive it.
What shouldn't be automated
The communications that shouldn't be fully automated are the ones where the relationship is the message:
- Recovery messages after a difficult delivery — these require genuine acknowledgement
- Upsell conversations — needs to reflect knowledge of the specific client situation
- Referral requests — trust-dependent, impersonal automation destroys the ask
- Post-project reviews with high-value clients — deserves a real conversation
- Messages after a client has gone quiet — the reason matters, automation assumes
The rule of thumb
If the right response depends on understanding this specific client's situation right now — automate the prompt, not the message. Let the AI surface the moment; let a human own the communication.
AI summaries: what they're for
AI-generated summaries of client history are underrated. Before a call, a one-paragraph summary of the last 90 days of interactions, project status, and open items means you're not spending the first 5 minutes of a call getting up to speed. At scale — 10, 15, 20 active clients — that compounds significantly.
The key is that the summary draws from real data: email threads, project notes, CRM activity, not just the last message. That requires a system where all of that context is in one place, which is the underlying infrastructure problem that most agencies solve last.
The balance in practice
In practice, the most effective automation strategy is layered: systematic messages fire automatically, but the system also flags moments that need a human. A client that's gone quiet gets a flag, not an auto-send. An upsell opportunity gets surfaced, not pitched. A difficult delivery gets a prompt to check in, not a templated apology.
“Automation at its best doesn't replace client relationships. It ensures you have the time and context to maintain them well.”