Quick answerYes, an AI agent should proactively surface a better-fit plan or an unused feature that would solve a problem the customer is actively describing, even when the change might reduce revenue, such as recommending a cheaper plan that actually fits their usage. Gate it narrowly: only surface the suggestion when there is clear evidence in the current conversation that the customer's actual usage does not match their current setup, not as a standing sales behavior, and always let the customer decline without friction.
Fit optimization is not the same motion as an upsell
Suggesting an upsell during an unrelated support conversation is about proposing something new the customer would pay more for. This is close to the opposite: proactively telling a customer about a plan, tier, or existing feature they are already entitled to that would better fit how they actually use the product, sometimes at a lower price than what they are currently paying. It can reduce near-term revenue, which is exactly why it needs its own governance rather than being lumped in with sales-oriented suggestions.
It is also not recommending a competitor's product, that is about sending a customer to a rival entirely. This is about optimizing fit within your own product, which is a much lower-risk, higher-trust move to get right.
Why doing this builds more trust than it costs in revenue
Customers notice when a company is willing to tell them they are overpaying or under-using something they already have access to. That noticing compounds into retention and referral behavior that a short-term revenue hit from a plan downgrade rarely offsets. The math only works, though, if the suggestion is genuinely well-targeted and not offered so often that it starts to feel performative.
When to trigger the suggestion
Trigger only on direct evidence from the current conversation, not from a standing usage-monitoring job that fires suggestions on a schedule regardless of context. If a customer describes a problem that an unused feature on their current plan already solves, mention it in that moment. If a customer describes usage patterns that clearly do not match their current tier, mention the better-fit tier in that moment, whether that means suggesting they upgrade for a real need or suggesting they downgrade because they are paying for capacity they do not use.
Do not trigger it as a general "have you considered" prompt disconnected from what the customer is actually telling the agent. That turns a trust-building moment into an ad.
How to phrase it
State the observation plainly and let the customer decide, do not frame it as a decision already made on their behalf. "Based on what you're describing, your current plan already includes X, which would solve this" reads very differently than pushing the customer toward a change. Never make the switch automatic. Always require explicit confirmation before any plan or billing change takes effect, even a downgrade the customer seems to want in the moment.
Measuring this without over-optimizing against it
If you measure agent performance primarily on upsell revenue, this behavior will quietly get deprioritized or suppressed, because a well-targeted downgrade suggestion looks like a loss on that specific metric even though it improves customer satisfaction and NPS and reduces churn. Track fit-suggestion behavior against retention and satisfaction metrics separately from revenue-per-conversation, or it will get optimized away by whichever metric is weighted more heavily elsewhere.
FAQ
Does this apply to enterprise accounts with a dedicated human account manager? The agent should surface the observation to the account manager rather than acting on it directly with the customer, since plan changes on managed accounts usually go through a relationship-based conversation, not a self-serve one.
What if the better-fit suggestion would eliminate the customer's need for the product entirely? That is a legitimate outcome and should still be surfaced honestly. Suppressing a suggestion because it might lead to cancellation is the exact trust erosion this practice is meant to avoid.
How often is too often to make this kind of suggestion? If the same customer has already declined a fit suggestion once, do not repeat it in the next conversation unless something material has changed in their usage since. Repetition after a decline reads as pressure, not helpfulness.