Someone contacts support, gets a reply, and contacts support again about the same thing. This is the most expensive pattern in a support operation, and it is also the one most teams measure last. Every repeat contact roughly doubles the cost of the original interaction, and it is a much better signal of broken service than any satisfaction score, because the customer is not expressing an opinion — they are demonstrating that the answer did not work.
Call Centre Helper’s work on reducing repeat calls and Grypp’s material on repeat contacts both converge on the same starting point: you cannot reduce repeats until you classify them, because the five common causes have almost nothing in common and each requires a different fix.
Five causes, five different fixes
The answer was wrong. The reply was confident and incorrect. This looks like an agent quality problem but is frequently a knowledge problem — the agent used an authoritative internal source that was out of date. Fixing it means fixing the source, not the agent, which is more work and is why it is often skipped.
The answer was right but incomplete. The agent answered what was asked and omitted the step that makes it work. A customer told to reset their password, without being told the reset link expires in ten minutes, comes back. This is a writing problem, and the fix lives in the knowledge base rather than the conversation.
The answer was right but the customer did not believe it. The reply was accurate and the customer came back to have it confirmed by someone else. This is disproportionately common in billing and returns. It is fixed by structure — showing the policy, the date, the reference number — rather than by tone.
The issue genuinely needed two steps. The customer had to do something first, then come back. Some of these are unavoidable processes, but they should be labeled as such in the first reply, so the return contact is expected rather than a surprise.
The customer could not find the answer again. They were told something in a chat, closed the window, and had no way to retrieve it. This one is pure product work: does the conversation survive, is there anything they can refer to later.
Sorting a month of repeat contacts into these five buckets takes an afternoon and immediately tells you where to spend. In most teams the second and third categories dominate, and both are fixed by writing and structure rather than by training or hiring.
The measurement that makes it actionable
A repeat contact rate on its own is not actionable, because it does not say which conversations came back or why. The version that changes behavior is repeat contacts per resolved ticket, tracked by intent.
That means two pieces of plumbing. A stable classification of what the conversation was about — the same discipline that keeps shared mailbox rules working as a team grows. And a link between the original conversation and the follow-up, which requires matching on customer identity rather than on ticket id, because the follow-up usually arrives as a new ticket.
Track it by intent and the picture becomes specific immediately. A high repeat rate in one product line is a product or documentation problem. A high repeat rate concentrated in one language or one shift points at a translation or handover issue, not at the individual agents involved.
Four fixes that move the number
The most effective interventions are unglamorous and they all target the first conversation rather than the second.
Put the expiring detail in the answer. If a link expires, a window closes, or a price is only valid for a period, say so in the first reply. Most repeat contacts are caused by an omitted constraint, not an omitted fact.
Give the customer something to keep. A reference number, a link to the article, a copy of the conversation. The fifth cause above — could not find it again — disappears entirely when the reply ends with something retrievable.
Close the loop in writing on policy questions. When a customer asks whether something is possible, the first reply should read like a decision rather than an opinion, citing the rule. This is what stops the confirm-with-someone-else pattern.
Separate "answered" from "resolved" in your own reporting. If your agents close tickets as resolved the moment they send a reply, repeat contacts are invisible until they arrive as new tickets, by which point the connection is lost. This is the same distinction that makes first response time a misleading metric on its own.
What not to do
Do not set a repeat contact target in isolation. A team told to reduce repeats can achieve it by making the first reply longer and more complex, which suppresses follow-ups and increases effort — a classic case of a metric being optimized against the thing it was meant to measure.
Pair it with something, and pair it honestly. Effort per contact works, because it prevents the gaming above. And the share of conversations that end in a confirmed outcome rather than a silent one works, because it measures the thing repeat contacts are a proxy for.
If you only look at one number this quarter, make it repeats by intent. It is closer to the customer’s actual experience than satisfaction scores, it is harder to game than response time, and every fix it points at makes the product documentation better as a side effect.