elli

AI readiness

AI is not your biggest risk. Being unprepared is.

The readiness gap is the distance between the change heading for your organisation and how ready your people are for it today. elli measures that gap, explains where it comes from, and helps you close it.

The 70/20/10 split of value in an AI initiative Seventy per cent of the value sits in people and processes, twenty per cent in technology and data, ten per cent in algorithms. Readiness assessments today look almost only at the twenty per cent technology and data. WHERE THE VALUE SITS 70% 20% 10% people & process technology & data algorithms WHERE READINESS ASSESSMENTS LOOK TODAY not measured the 20% The 70% goes unmeasured, and that is exactly the half where it goes wrong.
BCG, Where’s the Value in AI?, 2024

In five seconds: this page is for you if

you are rolling out AI and do not know whether your people are on board
you have handed out licences and usage tails off after a few weeks
your leadership team asks "are we ready?" and you want more than a gut feeling
you have to write an AI policy and do not know where your people stand today

An AI rollout succeeds or fails with your people, not with the technology

The split is well known. According to BCG, 70% of the value of an AI initiative sits in people and processes, 20% in technology and data, and 10% in the algorithms themselves (BCG, Where’s the Value in AI?, 2024).

Yet readiness assessments are almost always about infrastructure, data quality and integrations: the 20%. The 70% goes unmeasured, and that is precisely where it goes wrong.

You can see the result in the numbers. And it was rarely the technology.

42%

of companies abandoned most of their AI initiatives in 2025

S&P Global, 2025

17%

was the figure a year earlier

S&P Global, 2025

>40%

of agentic-AI projects will be scrapped by the end of 2027, Gartner expects

Gartner, 2025

More AI is being used in your organisation than you think

Leaders estimate that 4% of their employees use generative AI for at least 30% of their daily work. Employees themselves report 13%, more than three times as many (McKinsey, Superagency in the Workplace, 2025).

That gap cuts both ways. If your people are further along than you thought, you miss the chance to roll their way of working out more widely. If they are quietly dropping out, you are building your rollout on support that is not there.

It makes sense, too: almost nobody tells their manager they are afraid of being replaced. Meanwhile 96% of CEOs see AI being used outside approved channels, and 79% lose sleep over the legal risks (Dataiku & The Harris Poll, 2026). Plenty of work is being done with AI, just not where anyone is looking.

The perception gap: 4% against 13% Leaders estimate that four per cent of employees use generative AI intensively; employees themselves report thirteen per cent, more than three times as many. Leaders estimate 4% Employees report 13% more than three times as many use of generative AI for ≥30% of daily work
McKinsey, Superagency in the Workplace, 2025
96%

of CEOs see AI being used outside approved channels

79%

are worried about legal exposure

Dataiku & The Harris Poll, 2026

AI readiness is two questions: is your organisation ready, and are your people ready too?

Most assessments answer only the first question: infrastructure, data, governance. But it almost always goes wrong on the second.

Two questions, one picture: the employee and the organisation On the left the employee with three axes: usage, engagement and experience, performance. On the right the organisation with seven dimensions. Together they answer one question: is your organisation ready for AI? Are we ready? Question 1 · the employee three axes, three moments of the same story Usage objective behaviour · source: system data weeks Engagement & experience has to be asked for · source: survey months Performance what someone delivers · source: outcome data quarters Question 2 · the organisation seven dimensions · source: interviews, document analysis, system data strategy and vision data infrastructure governance and ethics talent and skills processes culture and leadership a precondition, usually not the bottleneck

Question 1: are your people ready?

Three axes, and they measure the same story at three different moments.

Usage: what someone actually does with AI: which tools, how often, at what cost. Behaviour, not opinion. Moves in weeks.

Engagement and experience: how someone relates to the work and to the change: understanding the why, trust, a sense of autonomy. Moves in months, and you have to ask for it.

Performance: what someone actually delivers. Moves in quarters, and is almost always skipped, even though it is what decides whether usage and commitment produce anything at all.

Question 2: can your organisation carry it?

Seven dimensions decide whether AI can land in your organisation: from strategy and data to governance, talent and culture. Established through interviews and document analysis, topped up with the system data you already have.

Rarely the bottleneck, always the precondition: without governance, usage becomes invisible rather than absent.

Why you measure usage, engagement and performance together

The three axes do not move in step, and that is exactly the point. The people who use AI most do not necessarily perform best. Measure usage alone and you label critical but productive people as drop-outs. Measure experience alone and you do not know why someone has stalled. Measure performance alone and you do not know how durable that result is.

And one strong organisational score says nothing about what sits underneath it. A 68 out of 100, strong on talent and weak on governance, is useful for the leadership team but useless as the basis for an intervention.

Three axes, three speeds Usage responds in weeks and runs ahead; engagement responds in months; performance follows in quarters. today weeks months quarters Usage signal after weeks · runs ahead Engagement signal after months Performance follows in quarters

Usage is the only axis that runs ahead of the other two. It is continuously measurable and changes in weeks. That makes it the only signal that moves before the loss has happened, and so the only one that lets you intervene in time.

Prefer the full story?

The white paper Human-ready is AI-ready works out the method behind this page in full.

Download the white paper

One average score says nothing: look per team and per profile

A readiness score is an average, and an average describes a reality nobody actually experiences. Underneath an average of 68, IT can sit at 95 and Finance at 55; that difference decides where an intervention pays off and where it is waste.

Cross engagement with performance and you get nine segments, each with its own approach. Add usage as a third axis and those same segments exist in three layers: an employee is not in a box, but at a coordinate. That is how you tell apart profiles that look alike on the flat plane but ask for the opposite.

From average to coordinate: nine segments in three layers of usage Three three-by-three matrices side by side, engagement against performance, one per level of usage: low, medium and high. The same box in another layer is a different profile: an employee sits at a coordinate, not in a box. LOW USAGE performance → engagement → MEDIUM USAGE HIGH USAGE The same box, a different layer: a different profile. An employee is not in a box but at a coordinate.

Two examples of what that gives you.

The quiet drop-out

Still performing at level and not scoring alarmingly on engagement, but usage has been sliding for weeks. The earliest reliable signal that someone is checking out mentally.

The shadow user

Says they use a lot of AI while measured usage is low. That difference is itself the signal: this group proves there is demand, not reluctance.

So the same low score calls for a different conversation. Without that detail, everybody gets the same training.

A segment shows where it hurts, the drivers explain why

A segment shows where it hurts, not why. Two employees can end up in the same box for opposite reasons, and that reason is not in the KPI itself.

So we correlate the KPIs with the drivers behind them: perceived usefulness, autonomy over the tool, trust in the policy, workload, leadership support and psychological safety. The same logic as classic engagement models (Job Demands-Resources, Self-Determination Theory), carried through to AI.

That distinction decides the intervention: the KPI points at the place, the correlation at the cause.

KPIs correlated with drivers KPIS Usage Engagement Performance correlate DRIVERS perceived usefulness autonomy over the tool trust in the policy workload leadership support psychological safety

A low usage score × trust in the policy
= a communication problem

That same score × perceived autonomy
= a different kind of tool or a different role

From insight to adoption: every team blocks on one point

Measuring is not yet adoption. So we use ADKAR as the backbone: five phases, each with its own failure signal.

ADKAR: five phases with their failure signal Awareness, Desire, Knowledge, Ability and Reinforcement as a sequence. Each with a failure signal. One phase is highlighted: this team blocks on Desire. Awareness knowing why FAILURE SIGNAL "another tool" Desire wanting to join in FAILURE SIGNAL quiet resistance THIS TEAM BLOCKS HERE Knowledge knowing how FAILURE SIGNAL fades after training Ability able to in the real work FAILURE SIGNAL knowledge, not usage Reinforcement keeping it up FAILURE SIGNAL relapse For a team, adoption usually fails on one of those five, not on all five. The measurement points at which one. Then you repair that letter, and not the whole programme.

That happens in ninety-day waves: a baseline per team, the why and the ground rules, role-specific labs, practice with support, reinforcement, and then a re-measurement. By then the blocking point has moved, and the next wave starts there. Not a project with an end date, but a rhythm.

The ninety-day wave as a recurring rhythm 90 days a rhythm, not an end date baseline why & ground rules labs practice with support reinforcement re-measure

How a programme runs

Week 1
As-is scan. What is already there: policy, governance, existing data and, where available, usage data.
Week 2-3
Survey. Short and targeted, in Dutch, French and English, including for people without a work mailbox.
Week 4
Analysis. Scores per dimension and per team, segments, and the correlation with the drivers.
Week 5-6
Board-ready readout. Where you stand, where it hurts, why, and what the first wave should be, with an owner and an intended result for every recommendation.
After that
Guidance and implementation. The ninety-day wave, and a re-measurement that shows whether the intended segment really moved or only the average went up.

That last stretch is where this differs from an assessment. Most readiness exercises end at the report. We stay until the behaviour changes.

What you get in your hands

  1. One score with a maturity level, from wait-and-see to embedded, plus a profile per dimension so you can see which dimension is the brake.

  2. Readiness per team, because one average steers you wrong.

  3. Segments and profiles with the recommended approach per group.

  4. The root-cause analysis: which driver explains the score most strongly in which segment.

  5. An adoption plan in waves, with an intervention area, an owner and an intended result for every recommendation.

  6. A re-measurement at fixed intervals, so you know whether it worked.

Privacy

Reporting is always aggregated, never on a group smaller than five, GDPR- and AI Act-compliant. That choice is not merely legal: a measurement that undermines trust makes exactly the problem it sets out to solve worse.

Why elli

We measure the half others leave out.

The organisational model follows the common standard: seven dimensions, weighted scoring, five maturity levels. Alongside it comes a profile per employee on three axes. That second half is what readiness assessments usually do not do.

We do not stop at the report.

Measure, analyse, advise, guide, implement, and then re-measure. The guidance is not an afterthought: it is where the result comes from.

Continuous instead of once.

Readiness crumbles the moment a rollout reaches teams you never measured. Experience moves in months, usage in weeks. One measurement a year is always too late.

Built in the EU.

Reported in aggregate, with a privacy threshold that is not up for negotiation.

Start with a free baseline

The first measurement is free: one survey among your employees and a readout showing where you stand today and which dimension is your brake. It doubles as the baseline you measure everything that follows against.

From 25 employees. You send the survey to your team yourself; we do the rest.

Request your free AI-readiness check

Leave your details and we will get in touch to schedule the baseline.

env: preview