Draft thoughtful follow-ups from your CRM, not from guesswork

Follow-up messages are a useful place for AI assistance because the raw material often exists in a CRM or meeting record. They are also easy to get wrong: a fabricated promise, an invented detail, or an overly familiar tone can damage trust quickly.

Use a source hierarchy

Decide which records are allowed to inform a draft. A signed proposal, approved CRM field, and confirmed meeting note may be trusted for different purposes. Treat a model-generated summary as a draft source, not as a new fact.

Keep sensitive details out unless the workflow and the person using it are explicitly authorized. A shorter email with one verified next step is better than a comprehensive message assembled from every record.

Draft around the next step

Gather the contact, purpose, last confirmed interaction, open question, and approved next step.
Ask AI for one concise draft using only those fields.
Highlight statements that are commitments, dates, prices, or claims.
Have the relationship owner edit and send the message.
Record the final outcome separately from the generated draft.

Personalization is not the same as more detail. One accurate reference to the person's goal usually feels more thoughtful than five guessed details.

Draft prompt · verified context onlyWrite a concise follow-up email using only the approved context below. Requirements: - acknowledge the last confirmed topic; - state the single next step that is already approved; - ask at most one useful question; - do not add a date, price, feature, promise, or personal detail not in the context; - mark any sentence that requires the sender to verify it; - do not send the email. Approved context: {{crm_context}}

Make review quick and meaningful

Show the source fields next to the draft. A sender should be able to see where each claim came from and remove a sentence without fighting the tool. If a message needs legal, security, or pricing review, route it to the right person rather than hiding that requirement in a prompt.

Learn from sent messages carefully

Measure reply quality, edits per draft, and follow-ups that required correction. Do not treat sending more email as success. Ask whether the workflow helped a person remember the right context and move a real conversation forward.

Every factual statement has an approved source.
Commitments and sensitive claims are highlighted for review.
The human sender owns the final message.
CRM corrections are not silently overwritten by generated text.

The blueprint: gather verified context → draft one useful message → inspect every promise → send deliberately → record what actually happened.

Back to all articlesBuild thoughtfully · Review often