“Next best action” has become a standard feature in sales tools. In practice, many of these suggestions get ignored. Reps get a stream of instructions — call this person, send this email — with no explanation, and quickly learn to dismiss them.
Instructions erode trust
Experienced reps know their accounts. When a tool tells them to follow up with a contact they spoke to yesterday, or chase a deal they know is on hold, every wrong suggestion costs credibility. After a few of those, even the good suggestions get skipped.
Frame suggestions as evidence plus a question
We redesigned Astra’s suggestions so each one starts with what it noticed and why it matters, then poses the action as a question the rep can accept, change or reject.
““The CFO joined Tuesday’s call and asked about implementation cost twice. No one has followed up. Share the cost breakdown?””
That format does three things. It shows the reasoning, so the rep can quickly judge whether it’s right. It leaves the decision with the person who owns the relationship. And when the rep rejects or edits it, that feedback improves the next suggestion.
Design principles we follow
- Always cite the conversation or event that triggered the suggestion.
- Suggest fewer, higher-confidence actions rather than many weak ones.
- Make dismissing a suggestion as easy as accepting it, and learn from both.
- Never act on a customer’s behalf without a human confirming.
AI that explains itself gets used. AI that only gives orders gets muted. The difference is often a single sentence of context.