Measuring agent onboarding means tracking whether a new hire is actually becoming competent, rather than tracking whether they finished the training. The useful signals are trajectory and consistency rather than the score itself, a written definition of what ramped means, and the pattern of which criteria the hire fails and in what order. Read those across a whole intake and you learn something a curriculum cannot tell you, which is whether the training worked.
In short
- A new hire’s first QA scores are close to meaningless. Direction and spread carry the signal, not the number.
- Most teams cannot say what ramped means. Replace the 90-day calendar date with a four-condition exit test the hire can read on day one.
- New hires fail criteria in a predictable order, from procedure to judgement to tone. A failure out of order tells you where the training leaked.
- Score during nesting, but label those scores and keep them out of the ranked record. Ambiguity about which scores count is what damages trust.
- When several hires from one intake fail the same criterion, that is a training defect, not a hiring defect, and it is invisible on individual scorecards.
- At two reviews a week, week one is two data points. You cannot see a ramp curve in two points, so teams substitute a feeling for it.
Why a new hire’s first QA scores tell you almost nothing
Every onboarding programme produces quality scores from about week two, and almost every manager reads them the way they read a tenured agent’s: as a level. That reading is wrong for the first month, for three reasons that push in different directions.
The queue is not the same queue. New hires get routed the simple contact types on purpose, so their score is inflated relative to the work they will be doing in three months. A hire sitting at 88 on password resets has demonstrated nothing about how they will handle a billing dispute.
The sample is too small to mean anything. At a conventional cadence of two reviews per agent per week, a new hire’s week-one score is an average of two conversations. Two. That number is closer to a coin flip than a measurement, and it will move ten points either way next week for reasons unrelated to learning.
Reviewers go easy, and do not know they are doing it. Knowing a conversation belongs to someone in their second week changes how a reviewer grades borderline criteria. It is human and mostly well intentioned, and it means an early score is partly a measurement of who reviewed it.
What to look at instead
Direction. Plot each hire’s conversation scores as a series rather than collapsing them into a weekly average. A hire who runs 72, 74, 79, 83 is learning. A hire who runs 88, 71, 90, 68 is not ahead of them, whatever the averages say. The second is doing well on the contact types they were drilled on and falling apart on everything else.
Spread. Track the gap between a hire’s best and worst conversation each week. In a healthy ramp that gap narrows before the average rises. Consistency arrives before competence, and it is the earlier and more honest signal.
None of this works if the criteria are vague, so check that the rubric you are scoring against is specific enough that two reviewers would grade the same conversation the same way before you read trajectories off it.
What does ‘ramped’ actually mean? Write the exit test down
Ask five support managers when a new hire is ramped and four will say ‘about 90 days’ and the fifth will shrug. Ninety days is a calendar, not a standard. It tells you when to stop paying attention, not whether the person is ready. That vagueness is not a local problem: Gallup’s research on why the onboarding experience drives retention found only 12% of employees strongly agree their organisation onboards new people well.
An exit test is better and takes an afternoon to write. Four conditions, all holding at the same time, for two consecutive weeks:
- Quality floor. Scores at or above the 25th percentile of the tenured team, not the team average. Average is the wrong bar, because half your experienced agents sit below it by definition and they are not on a performance plan.
- Consistency. No single criterion failed more than once inside the window. One bad conversation is normal. Failing verification three times in a fortnight is a habit, not a bad day.
- Independence. Escalation and internal-help rates within a reasonable band of the team’s. This is the condition everyone forgets, and it catches the hire who looks ramped only because a senior colleague is quietly doing the hard part of every conversation.
- Coverage of contact types. They have handled, unaided, the contact types that make up most of your volume, not a sample of the easy ones.
Publish the test and give it to the hire on day one. Someone who knows what they are aiming at will aim at it, and you remove the most common complaint about onboarding, which is not that it is hard but that nobody will say what finished looks like. Never apply it retroactively. If someone was signed off in June under a looser test, they are signed off.
Expect the first version to be wrong. Run it against three hires you already consider fully ramped. If any of them fail it, the test is too strict rather than the agent being secretly bad.
Which criteria new hires fail first, and what the order tells you
New hires do not fail randomly. They fail in a sequence, because criteria differ in how much live exposure they need before they can be satisfied. Failures start with things that can be taught in a room and move toward things that can only be learned on live work. Knowing the expected sequence lets you tell a normal ramp from a broken one, and moves the conversation from ‘this person is struggling’ to ‘this specific thing did not get taught’. The failure that arrives out of sequence is the interesting one.
| Criterion family | Usually fails | Why it fails then | What it means if it is still failing at week 8 |
|---|---|---|---|
| Procedure and process adherence (steps followed, fields completed, correct macro or template) | Weeks 1 to 2 | Pure recall under time pressure. There is nothing to understand, only to remember, and remembering is hard while everything else is also new | Not a person problem. Your documentation is hard to find or out of date, or the process genuinely has more steps than a human can hold. Fix the process or the docs |
| Tool and system accuracy (right record, right field, clean notes, correct disposition) | Weeks 1 to 3 | The interface is unfamiliar and the hire is watching the customer rather than the screen | Training used a sandbox that does not match production, or the hire never got enough unsupervised reps before going live |
| Policy judgement (when to make an exception, when to escalate, when to say no) | Weeks 3 to 6 | Cannot be taught from a slide. It needs enough edge cases to have built a sense of where the line sits | Normal to still be wobbly. Worrying only if the errors are all in one direction, which usually means the hire is scared of one specific outcome |
| Completeness of resolution (root cause addressed, next steps set, no predictable repeat contact) | Weeks 4 to 8 | Requires knowing what happens after the conversation ends, which a new hire has not seen yet | The hire is closing conversations to survive the queue. Check whether they were put under handle-time pressure too early |
| Tone and empathy under pressure | Weeks 6 to 12 | Only shows up once the hire gets genuinely difficult conversations, which the routing rules were protecting them from | Coach it, but check the routing first. If they got hard contacts in week two this is out of sequence and the ramp plan is the problem |
Should you score during nesting, and does it count?
Nesting is the phase after classroom training where a new hire handles live conversations under close supervision, before joining the general pool. Some teams call it incubation, guided practice or on-the-job training. It is the first time real quality data exists for that person, and there are three things you can do with it.
- Do not score at all. Common, and it throws away the only window where you can observe a ramp happening. You then spend week twelve guessing about week three.
- Score it and count it in the record. Worse. You are grading people on the days they are least equipped, with the least reliable data you will ever collect, then carrying that number into their first review. Agents notice, and conclude that quality scores are a stick.
- Score it and label it. Same rubric, same reviewers, tagged as onboarding, excluded from ranked performance and from anything attached to pay or promotion. This is the right answer, not a compromise.
It is not the scoring that damages trust, it is the ambiguity. A hire told on day one ‘we score everything from your first conversation, none of it counts toward your review, here is what it is for’ will engage with it. A hire scored without that conversation will assume the worst, and be partly right.
Two things that make nesting scores worth collecting
Same-day feedback, not weekly. A habit formed in week two costs an order of magnitude more to unlearn in week ten, because by then it has been reinforced a few hundred times. The value of nesting data is mostly in how fast it comes back, which makes the mechanics of quality feedback matter more here than anywhere else in the programme.
Calibrate reviewers on new-hire work specifically. Agreement is worse on new-hire conversations than on tenured ones, because the errors are more numerous and reviewers differ on how much slack to give. Almost nobody checks this. Run one calibration session on nesting conversations only before you make a decision off the data, or you will read reviewer variance and call it a ramp curve. The mechanics are the same as any other calibration session you run, only the sample changes.
What to measure at week 1, week 4 and week 12
Three checkpoints is enough, and more than three means nobody runs them. Each asks a different question, and the common mistake is asking the week-twelve question in week one. Week 1 asks whether anything is mechanically broken. Week 4 asks whether the person is learning and the training is holding. Week 12 asks whether they are ramped by the exit test, and if not, whether that is about them or about us. Read the table down the last column, because the unhealthy patterns are what you are scanning for.
| Checkpoint | What to look at | Healthy pattern | Unhealthy pattern, and the read |
|---|---|---|---|
| Week 1 | Volume handled, procedure and tool criteria only. Ignore the composite score entirely | A low but rising score. Procedure failures present and repetitive. The hire is asking a lot of questions | A high score. It almost always means the queue is too easy or the reviewer is being kind, and you will pay for it in week six. Also unhealthy: very few questions, which usually means the hire does not yet know what they do not know |
| Week 4 | Trajectory across the four weeks, spread between best and worst conversation, criterion-level fail rates, escalation and internal-help rate | Line rising, spread narrowing, procedure failures fading, judgement failures appearing. Escalation rate still above the team’s but falling | Flat line: the feedback loop is not reaching them, check whether reviews are being delivered at all. Sawtooth with a wide spread: they are strong on drilled contact types and lost on everything else, so widen exposure. Escalation rate already at the team average: they are guessing rather than asking, which is the most expensive failure on this list |
| Week 12 | The full exit test: quality floor, consistency, independence, contact-type coverage. Plus cohort-level criterion fail rates against the tenured team | All four conditions met for two consecutive weeks. Any remaining failures scattered across different criteria rather than concentrated in one | One criterion failing repeatedly while everything else is fine: a specific teachable gap, coach it, do not extend the whole ramp. Quality floor met but independence not: they are being carried. Several hires from the same intake failing the same criterion: stop looking at the individuals, the training is the defect |
The cohort read: training defect or hiring defect?
This is the part individual scorecards physically cannot show you, and it is the highest-value thing on this page.
When one hire fails a criterion, you coach the hire. When four of six hires from the same intake fail the same criterion, coaching four people separately is not just inefficient, it is a misdiagnosis. You have a training defect, and you will reproduce it in the next intake unless you find it.
How to run the comparison
For each intake, build a criterion-by-criterion fail rate. One row per criterion, one column for the cohort’s fail rate, one for the tenured team’s rate on the same criterion over the same period. Then look for criteria where the cohort rate is roughly double the tenured rate or worse.
A general gap is expected, because new hires fail more and that is what new means. What you are hunting is the criterion whose gap is disproportionate compared with every other criterion in the same cohort, because that is the module that did not land.
Then close the loop, which is the step almost no team takes. Change the training, run the next intake, and check whether that criterion’s gap came down. Training changes are usually made on instinct and never verified, which is why the same module fails for years.
Two caveats. Cohorts are small, so four of six is a reason to go and read those conversations, not a statistic. And a cohort gap can be a routing artefact, so confirm the two intakes handled comparable work first. Prolonged failure on one criterion is also among the more reliable early signals that a hire will not stay, which connects this to retention in customer support and makes the cohort read worth running even when every individual looks acceptable. What comes out of it is a coaching plan grounded in evidence rather than impression, the same discipline described in turning QA data into coaching, applied to a group instead of a person.
What changes when every conversation is scored from day one
Everything above is runnable by hand, and considerably easier to run when the underlying data is not two conversations a week. Under QA sampling, a new hire’s first week produces two reviewed conversations. You cannot see a curve in two points, so managers substitute a feeling, weighted heavily by whichever conversation they read last. That is the real state of most onboarding measurement, including at teams who would tell you they measure it carefully.
With 100% coverage revealing trends that 3% sampling never could, week one is every conversation the hire handled. The mechanics of scoring every conversation are covered separately. Three things become possible:
- Day-level trajectory instead of week-level. You catch a wrong habit in 48 hours rather than at the fortnightly review, which is the difference between a correction and a retraining.
- The cohort comparison becomes readable. Criterion-level fail rates across six hires need volume behind them. At two reviews a week they are noise.
- Contact-type coverage becomes verifiable. You can confirm the hire handled the contact types the exit test requires rather than assuming it from the rota.
There is a real cost, and it is worth naming. Scoring everything raises the bar on defensibility rather than lowering it. A hire in week three told they lost points needs the specific moment in the specific conversation, not a total, or you have automated the thing that erodes trust fastest. Every deduction has to name the criterion, point at the exchange that triggered it, and be disputable. That traceability matters more than the coverage does.
Kaizo’s Auto QA scores conversations against your own scorecard and keeps the reasoning attached to each one, so a nesting-week deduction can be traced back to the exchange that caused it rather than argued from memory. Cadence is the other half: the constraint on same-day nesting feedback is rarely willingness, it is the hours a manager spends assembling evidence before a coaching conversation, which is where coaching support earns its place. EverHelp cut coaching prep time by 90% on that basis, and prep time is what stands between a weekly nesting review and a daily one.
Five ways onboarding measurement goes wrong
In rough order of how often they show up.
- Using average handle time as the ramp signal. Handle time falls as a hire gets faster, and it also falls when they start closing conversations without resolving them. On its own it cannot tell those apart, which makes it useless as a proficiency measure and harmful as an onboarding target.
- Benchmarking the cohort against your best agent rather than the 25th percentile, which sets a bar most of your existing team would fail and makes every new hire look broken.
- Changing the rubric mid-cohort. It makes the trajectory unreadable, because you can no longer tell improvement from a scoring change. Version the scorecard and start new versions at intake boundaries.
- Treating day 90 as a decision date. If the exit test was met at week seven, sign them off at week seven. If it was never met, week thirteen will not fix it.
- Never comparing cohorts. Without an intake-over-intake view you cannot tell whether last quarter’s training changes did anything, and the quality trend you report upward is hiring luck and training effect mixed together.
Frequently asked questions
How long does it take to ramp a new call center agent?
Most support teams land between six and twelve weeks to full independence, but the number is a property of your work rather than of the hire. The range of contact types, how much policy judgement the role carries and how good your documentation is move it more than anything about the individual. The better question is not how long it takes on average but whether you have written down what ramped means, because a team that cannot define it cannot tell you whether the hire in front of them has got there.
What is nesting in a call center?
Nesting is the phase between classroom training and full independence, where a new hire handles real conversations under close supervision, usually on a restricted set of simpler contact types with a trainer or senior agent immediately available. It is also called incubation, guided practice or on-the-job training. It matters for measurement because it is the first period where genuine quality data exists for that person, and how you treat those scores sets the tone for how the team reads quality data afterwards.
Should new hire QA scores count toward performance reviews?
No. Score during onboarding, using the same rubric and reviewers, but tag those scores as onboarding and keep them out of the ranked record and out of anything attached to pay or promotion. The data is valuable for spotting where training leaked and unreliable as a measure of the person. What damages trust is the ambiguity rather than the scoring, so tell the hire on day one which scores count and which do not.
What does time to proficiency mean, and how do you measure it?
Time to proficiency is how long it takes a new hire to perform at the standard expected of a fully productive agent. Measuring it requires defining that standard first, which most teams have never done. Replace the calendar date with an exit test of four conditions that must hold together for two consecutive weeks: quality scores at or above the tenured team’s 25th percentile, no single criterion failed more than once, escalation and internal-help rates near the team’s, and hands-on coverage of the contact types that make up most of your volume.
How many conversations should you review for a new hire each week?
As many as you can, and considerably more than for a tenured agent. At two reviews a week, week one is an average of two conversations, which cannot distinguish learning from luck, and the criterion-level patterns that show what training missed never become visible. If you are limited to manual sampling, weight it heavily toward new hires in their first month and prioritise reading whole conversations over thin coverage across the whole intake.
What are the stages of call center onboarding?
Most programmes run four: orientation and systems access, classroom or virtual training on product and process, shadowing where the hire observes experienced agents, and nesting where they handle live conversations under supervision before joining the general pool. The stages are well covered elsewhere and they are the easy half. The half that decides whether onboarding worked is the measurement laid over the top of them, which is what this guide covers.
Related terms
See the ramp curve instead of guessing at it
Score every conversation a new hire handles from their first day, with the reasoning attached so a week-three deduction can be traced to the exact exchange. See what a real ramp looks like on your own data.
