How do we tell closed from solved?

By what happens next rather than by what the record says. Repeat contact within a few days, whether the customer completed the thing they were attempting, and any refund or complaint that follows are all generated without asking and cannot be adjusted by whoever closed the case.

Do satisfaction surveys answer this?

Partly, and they are weaker than they look. Surveys are answered by the very pleased and the very annoyed, so they describe the ends of the distribution well and miss the large middle who said nothing and quietly took their business elsewhere.

What breakdown is worth having?

By case type, always. An aggregate rate averages trivial enquiries together with the ones that cost real money, and the only figure worth acting on is how the system performs on the expensive category rather than across everything at once.

Glossary

Resolution rate

The share of cases genuinely settled rather than simply marked finished, which are different things decided by different people.

In plain terms

How many problems actually got fixed. The catch is that somebody has to decide a case is finished, and the person deciding is usually the one being measured on it, not the customer who had the problem.

Why it matters

Because everything downstream is built on it. Staffing, automation business cases and judgements about whether a deployment worked all rest on a count of solved cases, and if closing is what is really being counted, every conclusion drawn from it inherits the error.

How it works

Closing and solving are separate events performed by different parties. A case is closed by whoever handles it and solved only if the customer's problem went away, and nothing in most systems forces those two to agree.

Automation widens the gap rather than creating it. Human agents close cases optimistically too, and a system doing the same at volume, with no instinct for when an answer felt thin, turns an occasional error into a steady rate.

The reliable check is behaviour afterwards. Whether the same customer contacts again within a few days, whether they completed what they were trying to do, whether a refund or complaint follows: those happen without anybody being asked and cannot be adjusted by whoever closed the case.

Asking the customer helps and is weaker than it looks. Surveys reach the people willing to answer surveys, which skews towards the very satisfied and the very angry, and misses the large middle who quietly went elsewhere.

A rate that is too high is a warning rather than a triumph. Some cases genuinely cannot be settled in one pass, so a system reporting that almost everything resolved is more likely to be describing its closing habits than its performance.

Who is measured on it decides how it behaves. Where closing rates are reported per agent or per system and nothing downstream is reported alongside them, the number stops describing the work and starts describing the incentive, which happens without anybody choosing to be dishonest.

It interacts badly with time targets. A team pushed on speed and on resolution at once will close faster, and closing faster is the one action that satisfies both measures immediately while making the second contact more likely rather than less.

Breaking it down by case type is where the useful information is. An aggregate figure averages the trivial and the hard together, and the number worth acting on is how the rate looks on the cases that actually cost the business something.

Where the two versions of the number separate

Reporting almost always stops at the third step, and everything that decides whether the work was any good happens at the fourth. That is not laziness: the third step produces a clean event with a timestamp, and the fourth produces a signal that arrives days later, attached to a different record, sometimes in a different system entirely. Closing the loop means joining those two, and the joining is the work most support reporting never does. Where it does get done, the finding is reliably uncomfortable and reliably useful: a modest share of closed cases generate a second contact, that share is much higher for some case types than others, and the types where it concentrates are usually the ones somebody already suspected. None of that requires new tooling or a survey programme. It requires deciding that the number reported at step three is provisional until step four has had a few days to happen, which is a reporting convention rather than a project.

Seen in the wild

Cases marked solved that reappear from the same customer two days later.
Tidio Lyro
An automated agent closing a case with an answer nobody confirmed worked.
Sierra
Internal questions marked answered while the same query keeps being searched.
Glean

Common misconceptions

People assume

A closed case is a solved one.

In fact

Closing is done by whoever handled the case and solving is decided by whether the customer's problem went away. Most systems never make those agree, so the count of closures is what gets reported.

People assume

A very high rate is a good sign.

In fact

Some cases cannot be settled in one pass, so a rate approaching everything is more likely to describe closing habits than performance. It is a prompt to check repeat contacts rather than a result to celebrate.

Questions

How do we tell closed from solved?: By what happens next rather than by what the record says. Repeat contact within a few days, whether the customer completed the thing they were attempting, and any refund or complaint that follows are all generated without asking and cannot be adjusted by whoever closed the case.
Do satisfaction surveys answer this?: Partly, and they are weaker than they look. Surveys are answered by the very pleased and the very annoyed, so they describe the ends of the distribution well and miss the large middle who said nothing and quietly took their business elsewhere.
What breakdown is worth having?: By case type, always. An aggregate rate averages trivial enquiries together with the ones that cost real money, and the only figure worth acting on is how the system performs on the expensive category rather than across everything at once.