Measuring Deflection Honestly Against Resolution and Repeat Contact
A metric called deflection measures what percentage of customer inquiries are handled entirely by the automated system without human involvement. Organisations often celebrate high deflection rates as evidence that automation is working well. However, deflection alone is a misleading metric. A bot might deflect an inquiry (the customer does not escalate to human support) without actually resolving the customer’s problem. The customer might give up, move to a different channel, or contact the organisation again later.
True resolution means the customer’s problem is solved. Resolution differs from deflection. A customer might contact a bot about returning a product. The bot provides return instructions (deflection occurs, no human is involved). The customer attempts to return the product but encounters problems with the return process. The customer then contacts the organisation via phone to get help. The original interaction was deflected but not resolved. Measuring only deflection would show a success, but the customer has now had to contact the organisation twice.
Repeat contact is a key metric that reveals whether deflection actually solved problems. If a customer contacts the organisation about an issue, and then contacts again about the same or related issue within 30 days, this indicates that the first contact did not resolve the problem. An organisation can calculate a repeat contact rate: (number of customers who re-contacted within 30 days about the same issue) divided by (total number of initial contacts). A high repeat contact rate combined with high deflection suggests that the bot is deflecting inquiries without solving problems.
Customer satisfaction is another metric distinct from deflection. A customer might have their inquiry deflected by a bot, and might report that they are satisfied with the bot (because the bot was polite and responsive). However, if the bot did not solve their problem, the satisfaction is misleading. Many organisations now measure both satisfaction and actual outcome. A question such as “Was your problem resolved?” is more valuable than “Was the bot helpful?” or “Was the interaction smooth?”
Some organisations implement a metric called resolution rate within deflection. This asks: of the customers whose inquiries were deflected (handled entirely by the bot without escalation), how many report that their issue was actually resolved? A bot with 80 percent deflection but only 40 percent of deflected customers reporting actual resolution is failing despite the high deflection.
Organisations sometimes game metrics by designing systems that deflect inquiries without actually helping. A bot might provide information without confirming that the customer understands or can use it. A bot might offer a workaround that does not truly solve the problem but is “good enough” to satisfy a survey question. A bot might avoid escalating difficult inquiries by simply terminating the conversation without offering human support, which shows as deflection but fails the customer.
The most honest approach combines multiple metrics. An effective bot should show: reasonable deflection rate (such as 60-75 percent for most operations), low repeat contact rate (such as less than 10 percent), high actual resolution rate among deflected interactions (such as 80 percent or higher), and positive customer satisfaction. If deflection is high but repeat contact is also high, the bot is not actually solving problems. If deflection is high and repeat contact is low, the bot is genuinely deflecting resolved inquiries.
Some organisations track customer sentiment before and after bot interactions. A customer might start the interaction frustrated about a problem and end it either: resolved and satisfied (good outcome), unresolved but accepted (the bot managed expectations honestly), or unresolved and still frustrated (poor outcome). Measuring sentiment change reveals whether the interaction improved the customer’s experience.