Case: the review season

A review season starts with the ruler, not the names. Fix the criteria before opening anyone's numbers: quality for families (reopened tickets, satisfaction), volume (tickets closed), help to the team (mentoring hours) and reliability (incidents in the team's log). Then apply them to the same period for everyone: January to July.

For each person, compute the period's rate the honest way: pooled, total reopened over total closed, not the average of seven monthly rates. Months differ in size (spring peak, two-week July), and averaging rates gives a small July the same weight as a busy April.

Start with the person whose quality is easiest to overlook. Maryna reopened 5 + 5 + 6 + 6 + 6 + 5 + 2 = 35 tickets out of 257 + 252 + 294 + 308 + 287 + 234 + 121 = 1,753 closed: 35 / 1,753 = 2.00%, the lowest on the team, with 487 of 513 survey answers positive (94.93%) and no incidents. Quiet people's quality is exactly what memory drops.

To read four people on one ruler, it helps to know the team's level on each criterion. Compute it the same way as each person's rate: pooled. Add the reopened tickets of all four (87 + 257 + 35 + 62 = 441) and divide by all tickets closed (2,160 + 2,833 + 1,753 + 1,537 = 8,283): 5.32%.

The plain average of the four personal rates (4.03%, 9.07%, 2.00%, 4.03%) is 4.78%, lower, because it gives Bohdan's 1,537 tickets the same weight as Taras's 2,833. The pooled figure is what families actually experienced: about one ticket in nineteen came back.

The baseline turns numbers into statements you can defend. Maryna's 2.00% and Olha's and Bohdan's 4.03% are below the team's 5.32%; Taras's 9.07% is well above it. Satisfaction works the same way: the team's 87.12% sits between Maryna's 94.93% and Taras's 78.67%. A baseline doesn't rank people; it shows where each one stands on each criterion.

A monthly rate built on few answers swings for no reason at all. Maryna gets about 80 survey answers a month, so a single answer moves her monthly satisfaction by about 1.26 points. A 2-point dip in one month is less than two unhappy families: noise, not a signal. In July, with only two shift weeks, she had just 35 answers, so July's rate is even jumpier.

So don't read people month by month in a review. Pool the period: over January to July Maryna has 487 positive answers of 513, 94.93%, the best on the team. The same rule protects everyone: Taras's 96% week rests on 25 answers, and a single bad month for Olha or Bohdan could look alarming on its own.

When does a monthly change matter? When it repeats, or when it's large compared with what one or two answers can move. A drop of 10 points two months running is worth a conversation; a 2-point wobble isn't.

Read on the same four criteria, each coordinator has a real strength, and each would vanish if the review looked at one column only.

• Maryna: the lowest reopen rate (2.00%), 94.93% positive answers, no incidents. Invisible in a ticket count.
• Bohdan: the most mentoring on the team, 36 hours, mostly with the spring hires. He has the fewest tickets (1,537), so a volume-only review would put him last. Olha remembers that in spring the seasonal hires went to Bohdan first.
• Taras: the most tickets closed (2,833) and a fast, calm week under the flood of 12 July. His quality (9.07% reopened, four incidents) is his area to work on.
• Olha: steady on every criterion (2,160 tickets, 4.03% reopened, 91.10% positive, one incident, 24 mentoring hours), plus her judgment on refund exceptions.

That's why the criteria are several and fixed in advance: a single metric makes one kind of contribution visible and the others disappear.

"Could we just use July? It's the freshest month, and everyone remembers it." Using the same month for everyone sounds fair, and that's what makes the shortcut dangerous.

July is a tiny, unusual slice. Because of vacations, it holds only two shift weeks per person, and for Taras one of them is the payment-page week. On July alone Taras closes the most (245 tickets) with 4.49% reopened and 87.04% positive, close to Bohdan's 3.42% and 87.88%. Over January to June, Taras reopened 9.51%. A July-only review would rate him far higher than seven months support.

That's the recency effect, written into a rule. The fix is the one already agreed: the same seven months for everyone, with July in it as two of about thirty weeks. The standout week still appears in Taras's block, as a fact about how he handles a flood.

The percentages for July can be calculated, and mentoring isn't the issue; the problem is how little of the year July shows.

Bohdan's block is where similar-to-me sneaks in. He replies slowest and pushes back on the AI reply assistant, and a lead who was fast and quick to adopt tools may write "slow and stubborn about new tools". That's a label, not a fact.

A fair block does three things. It states facts, good and bad: led the team in mentoring, 36 hours, mostly onboarding the spring hires; slowest first replies on the team; reopen rate 4.03%, two incidents. It turns a weakness into a question: "to discuss: what slows the first reply", which leaves room for the situation behind it, such as mentoring or long video-outage tickets. And it proposes a plan the person can accept or change: make onboarding a named part of his role for next spring, if he wants it.

His doubts about the assistant are a view to discuss, not a trait to grade. And ranking him by tickets alone ("fewest tickets, so the weakest") ignores the 36 hours that made the spring hires productive.

The written block is half the job; the conversation is the other half, and its order matters.

1. Their self-assessment first. You hear how they see the period before your view can anchor them, and it often brings in facts you missed.
2. Your facts on the shared criteria, next to theirs. Where you agree, say so; where you differ, show the evidence.
3. A plan built with coaching questions: goal, reality, options, from the coaching lesson, so the plan becomes theirs.
4. One next step and a review date, written down.

When a self-assessment overstates, don't argue with it; ask about it. Taras writes: "July shows what I can really do." It's partly true: the 12 July week shows real strength under pressure. Start there, ask what made that week work, and then put the six months next to it. The aim is that he leaves with the same picture you have, and a plan for his reopen rate that he chose.

Iryna asks for three or four lines per person: results on the shared criteria, one strength, one thing to work on, and the plan for next year. The evaluation part is yours to write and hers to read. The plan part is different: it stays a proposal until each person has added to it in their one-to-one. Mark it that way in the block, so nobody, Iryna included, reads it as settled.

That's the rule from the development-plan lesson, applied under deadline pressure: they draft or add, you propose, and a date to look back. A plan nobody shaped but the lead is a list of the lead's wishes. The test for all four blocks: could each person read their own and recognize themselves, strengths and all?

How it fits together: criteria first, pooled rates, a team baseline, months read together, no July shortcut, labels turned into facts, and plans built with people. That closes the course: in January the team waited for you on every hard ticket; in October it runs two weeks without you. That's a lead's real output.