The batch graduated on a Tuesday. There was a sheet cake, a certificate for each of the fourteen, and a floor lead who clapped a little longer than the moment needed because she had been told the ramp was behind. Every one of them had cleared the mock-call assessment. The nesting report went out that afternoon with a single line in green: "Batch certified, 100% pass."
On Thursday the queue spiked. One of the fourteen, a quiet and careful agent who had scored near the top of her cohort, took two calls, froze on the third, and spent the rest of the shift transferring anything that looked unfamiliar. Her handle time drifted up, then up again. Nobody called it a problem, because a problem needs a baseline, and the baseline was a pass mark from Tuesday that still said she was ready. Three weeks later a quality audit sampled her calls, the client read the sample, and the question in the review was not why she struggled. It was why the program said she was ready.
The pass mark measured the room, not the floor.
An assessment is a clean, scripted, supervised exercise. The customer in it is kind and follows the plan. The queue behind it is empty. Nobody is waiting, nobody is angry, and the screen behaves. Passing it proves one thing with real confidence: this person can perform when conditions are arranged for them to succeed.
The floor is not arranged for anyone. It has a queue that does not pause, a customer who has already explained the problem twice, a knowledge base that contradicts itself in two places, and a team lead who is covering four people. The skill being tested on Tuesday and the skill being used on Thursday overlap, but they are not the same skill, and the distance between them is where most early attrition and most early quality misses live.
None of this is a secret. Every operations manager has watched a top scorer stall and a middling scorer thrive. What is rarely examined is why the report keeps treating the score as the headline anyway. It is because the score is the only number that exists on Tuesday. The real answer does not exist yet, and a report cannot be filed on a fact that has not happened.
A graduation is a date. Proficiency is a curve.
Here is the structural error, stated plainly. The program treats readiness as an event with a timestamp. Certified on this day, productive from this day. The person experiences readiness as a slope: slow, uneven, interrupted by a hard week, steepened by a good coach, flattened by a bad one. Two agents who certified on the same Tuesday with the same score can be on entirely different paths by the end of week three, and the certificate cannot tell them apart.
So the organization does what any system does when it cannot see a slope. It waits for the destination. Productivity is declared at a fixed day because the plan needs a day. When the curve does not honour the plan, the shortfall shows up as a number somewhere else: a handle time here, an escalation rate there, an attrition line in the quarter that nobody links back to a batch. The cause is a Tuesday. The evidence arrives in March. By then three more batches have graduated the same way.
Consider what the client actually bought. Not a certified agent, but a seat that performs at an agreed level from an agreed date. The contract speaks in outcomes, while the nesting report speaks in completions, and the translation between the two is performed by hope. When the seat underperforms, both sides reach for the document that supports their reading. The vendor points to the pass. The client points to the queue. Each is accurate. Neither document was ever built to answer the only question that matters to the person paying: on the day we start counting, will this seat hold, and how would we know a month earlier?
A certificate is the one document in the building that gets more confident the less anyone checks it.
Everybody sees the dip. Nobody is paid to forecast it.
The floor sees the dip in real time. Team leads know within ten days which new joiners are climbing and which are stalling. That knowledge lives in conversations and in a feeling, and it never reaches the report, because the report's author has no field for a feeling and the lead has no incentive to volunteer a bad forecast about a batch the program just celebrated.
There is a quieter cost. If the only formal measure of readiness is the pass mark, then every action taken after it is unmeasured by definition. The extra coaching, the shadowing, the quiet re-pairing with a stronger colleague: all of it is real work, none of it is in the system, and so when it works nobody learns why and when it fails nobody learns when. The organization is repeatedly surprised by something its own floor saw coming.
The same Tuesday, reported forward.
Rewrite that Tuesday report. Not "batch certified," but a statement about the next ninety days: of the fourteen, these are on a path to full proficiency by the date the client expects it, these four are not, and these two are the ones most likely to be gone before the account's next review. It is a prediction, which means it can be wrong, and that is the point. A report that can be wrong is the only kind that can be used.
Three shifts make that report possible, and none requires a new platform or a new data store.
First, judge by trajectory, not by the day-one snapshot. The score tells you where a person stood on Tuesday. The direction of their first three weeks, set against the cohort they joined with, tells you where they are heading. Handle time, quality scores and ramp milestones already exist. The slope is arithmetic on data the operation already produces, which is why nobody has to type in a number to be flattered by it.
Second, spend analysis by exception. Most of a batch will move together, and cheap signals will say so. Keep the indicator, throw the raw material away, and give the expensive look to the small number of people where the signals disagree with each other or with the cohort. The quiet agent who scored high and then stalled is exactly the kind of case where a high snapshot and a flat slope pull in opposite directions. A system that only reads the snapshot never notices her. A system that watches for disagreement finds her in week two.
Third, give readiness a price. A verified-ready seat and an unverified one are not the same product, but on most rate cards they cost the same, because readiness is booked as training spend rather than as something the account sells. Once a forecast is tied to the seats a client is actually paying for, the conversation changes owner. It moves from the training team, who must justify it, to the delivery owner, who can use it to commit with confidence or to ask for time before a seat goes live. Capability stops being a cost that defends itself and becomes part of what is on offer.
Where this breaks.
A forecast that hides its weak points is only a more elaborate green slide, so here are the weak points. Three weeks of data is a short window, and some people take off late and finish strong, which means a forecast will sometimes mark a good agent as a risk and cost them coaching they did not need. Some client environments restrict the very signals a forecast depends on, and you only find out which ones once you are inside. Trajectory needs a cohort to compare against, and a small or oddly composed batch makes the comparison noisy. And the people who admire the idea are not always the people who can authorize it. Where a method cannot name its failure points, it has not been tested against anything.
The method, not the link.
This is the work we do at SSUNDAR under the name Capability Forward. It starts with the March Bench Diagnosis: one account, one question, whether the bench you will need in March is the bench you are building today. The forecast and the benchmark stay with us, and part of the fee rides on the result the forecast shows, so some of the risk of being wrong is ours. That arrangement is the proof. A claim the author will not be paid against is only a claim.
The cake was eaten by four o'clock on Tuesday. The question it was supposed to answer was still open, and no one had asked it yet.