The first step is descriptive: what happened? This is where most organizations sit — you can see the numbers, but not necessarily understand or act on them. The second step is diagnostic: why did it happen? This is where you start to understand patterns — why the experience typically dips after four months, why one department scores lower than another. The third step is predictive: what’s going to happen? Here, you use the patterns to predict how things will likely go for an individual employee, before it happens. And the fourth step is prescriptive: how do we make it happen? This is where data turns into action, automatically and proactively.

This article series has, in practice, followed the same ladder. The early articles were about measuring and describing (the index, the dialogue report). Diagnosis followed (benchmarks, dimension signals, the onboarding curve, seasonal variation). With the catalogue in the previous article, we moved toward the prescriptive step — concrete interventions, ready to be deployed. What’s missing to close the loop is the predictive step: using everything we already know to predict what’s about to happen — before it shows up as a low score.

What makes prediction possible isn’t a single data source — it’s the combination of several.

If you know the employee’s own development so far in the process: scores on the six dimensions after 1, 3, and 6 months. If you know the department’s or organization’s typical pattern: the characteristic curve, with its dips and rises, and how it usually looks compared to the rest of the organization. If you know segment data: how employees in a specific age group, professional field, or role type typically perform. And if you know — provided you’ve been structured about it — exactly which onboarding design the individual employee has actually gone through: which interventions from the catalogue were deployed, and when.

Combine those four data sources, and you have something far more useful than a single score. You can start to ask: is this employee on a trajectory that historically leads to a problematic dip in two months? And if so — what usually works in that situation?

This is where the catalogue from the previous article becomes especially valuable. If you’ve already built a catalogue of interventions, organized by dimension and phase, the AI prediction can be linked directly to action: if the system predicts an employee is heading toward a dip in the Network dimension, it can automatically suggest — or even trigger — the relevant intervention from the catalogue before the dip fully sets in.

That’s the difference between a safety net you have to remember to set up yourself, and one that’s set up automatically when the data shows it’s needed. For an individual manager, it means no longer having to manually track every signal — the system does it, and flags it when something calls for action.

If you’re overseeing onboarding across multiple teams, this is arguably where AI adds the most leverage: it’s the difference between one manager keeping a mental tally of a handful of new hires, and a system consistently tracking the same signals across every team, every hire, all at once.

There’s a data source most organizations already have on hand but rarely use for this purpose: personality tests from the recruitment process.

Most organizations that hire in a structured way already test candidates during recruitment — and, in doing so, get a picture of how the person typically operates, what motivates them, and what kind of support they need. That picture is typically used once, in the hiring decision itself, and then filed away.

That’s a paradox. You already have data on what this specific person is likely to need in order to thrive — but it isn’t being used to shape onboarding, which is exactly the moment when that knowledge would be most useful. An employee who, according to their profile, thrives with structure and clear boundaries probably needs something different in their onboarding than one who thrives on freedom and fast-paced challenges.

Connect that kind of segment insight to the other data sources — the employee’s own development, the organization’s pattern, the catalogue of interventions — and prediction shifts from generic to personal. Not ‘this age group typically needs X,’ but ‘this person, with this profile, in this role, at this time of year, has historically needed Y.’

It’s worth saying plainly: the more precisely you can predict how a specific employee is doing, the more important it becomes to use that knowledge to help — not to monitor. The purpose of this kind of AI is to catch signals early enough to act supportively, not to add another layer of control to the employee’s day-to-day. It requires transparency about which data is used, and for what — and it requires that the employee experiences the prediction as something working in their favor.

We started this series with a simple question: why is satisfaction the wrong thing to measure? Along the way, we built on it step by step — from six dimensions, to dialogue reports, to an index, to benchmarks, to the curve, to the calendar, to a catalogue of concrete interventions. Each step made the next one possible.

With AI, the loop closes: from measuring what happened, to predicting what’s about to happen — and acting on it before it becomes a problem. This is no longer just a question of understanding your onboarding better. It’s a question of building an onboarding that actively looks out for every single employee, the whole way through.

THANK YOU FOR FOLLOWING ALONG
This was, for now, the final article in the series about measuring and improving onboarding.

Improve your onboarding processes with FastTrack. FastTrack is part of HR-ON Boarding+.

You might also like: From data to dialogue: What a dialogue report can do for your onboarding