We are drafting win-back emails with an agent. The more history we give it, the more personal the email gets, and past a point it starts to feel like surveillance. 'We noticed you stopped opening our emails on 14 November' is technically accurate and deeply creepy.
Where do you draw the line on what goes into the context?
We give it the plan, the last product used and how long since. No dates, no counts. It writes better emails with less.
Plan, product and tenure band, nothing dated. We also strip the customer's name from the context; the agent writes 'you' better than it writes 'Dear Priya'.
From the receiving end: I unsubscribe the moment an email tells me what I did. Tell me what is new instead.
From the automation side: the fewer fields the agent needs, the fewer journeys break when a field is empty. Less context is also more robust.
Update for anyone who finds this later: cut the context down to plan, last feature and tenure band. Fewer complaints, same reply rate over two sends. Small sample, but it is going in the playbook.
If anyone is measuring this, compare complaint rate, not just reply rate. The creepy version replies fine and unsubscribes a week later.