Stop measuring AI adoption. Start measuring what disappeared.

There is a particular type of metric we love when we talk about AI transformation: 50,000 AI interviews completed, 100,000 candidates assessed, 90% adoption, 47 markets live. These are the AI hiring metrics we reach for first. They're impressive numbers, and they're not meaningless. They tell us that a technology has been deployed, that people are using it and, in global organisations particularly, that something has successfully moved beyond a small pilot.

But they don't answer the question I increasingly think matters most: what actually got better?

If 100,000 candidates went through an AI-enabled hiring process but recruiters are still spending the same amount of time screening, candidates are still waiting a week for a response and hiring managers are still conducting interviews that add very little signal, have we really transformed anything?

We've become very good at measuring how much AI we're using. I think we need to get much better at measuring what disappeared because of it.

Adoption metrics are not AI hiring metrics

Technology usage isn't the same as transformation. This isn't unique to AI. We've always had a tendency to measure new technology by its usage because usage is easy to see. We can measure logins, transactions, completion rates, number of users and number of interactions, put them on a dashboard and watch the numbers go up.

And adoption does matter. Technology that nobody uses isn't going to transform very much. Measuring AI adoption tells us the technology is there. But adoption is an input, not an outcome. A study of more than 3,000 managers by MIT Sloan Management Review and BCG found that only about one company in ten reports significant financial benefits from AI.

When I look across some of the AI hiring programmes we've supported at Maki, the numbers I find most interesting aren't necessarily the biggest ones. Of course, operating at significant candidate volume tells us something about scale. But I'd rather talk about one organisation giving 232,000 hours back to recruiters, another reducing screening time by 90%, or another making its overall hiring process 26% faster. The named versions are in our customer stories.

Those numbers tell us something different. They tell us what changed because the technology was there.

The questions become less about how many people interacted with AI and more about what was removed from the process as a result. How much recruiter screening disappeared? How much candidate waiting time disappeared? How many unnecessary steps disappeared? How much manual administration disappeared? How much inconsistency disappeared? And, further down the line, did some of the poor hiring outcomes we were trying to solve start to disappear too?

Those are the AI hiring metrics that tell us whether transformation actually happened.

Did recruiter work disappear, or just move somewhere else?

Efficiency is probably the most obvious promise attached to AI in hiring, but I think we need to be much more precise about what we mean by it. It's easy to say that AI has automated a screening stage. It's harder, and much more useful, to ask what happened to the work that used to sit around that stage.

If a recruiter no longer conducts a 20-minute screening call but now spends 15 minutes reviewing the output, checking another system and manually progressing the candidate, we haven't removed 20 minutes of work. We've mostly moved it somewhere else.

The real opportunity is to look at the process end to end. How much human time was required to move a candidate from application to a meaningful hiring decision before? How much is required now? Which activities genuinely disappeared, and which simply changed shape?

We've seen what happens when that work genuinely does disappear. Across Maki programmes, we've seen 232,000 recruiter hours returned in one global hiring transformation, a 90% reduction in screening time in another, and at least 30% time savings in screening elsewhere.

That's a very different way of thinking about efficiency. Rather than measuring the volume of work AI completed, we're measuring the human capacity that was created because AI completed it.

And that, ultimately, is the opportunity. The goal isn't to automate recruiter work so that recruiters can become busier somewhere else in the funnel. It's to give people back time for the parts of hiring where their involvement has much greater value: building relationships, understanding candidates, partnering with hiring managers and making good decisions.

How much waiting disappeared for the candidate?

There's another efficiency metric we don't talk about enough, and that's candidate time. A huge amount of traditional hiring involves waiting: waiting for an application to be reviewed, waiting for a recruiter to become available for a screening call, waiting for an interview to be scheduled, waiting for feedback and waiting to find out whether you're moving forward.

Internally, those gaps can look like normal process. For the candidate, those gaps are the experience.

AI gives us the ability to remove a surprising amount of that dead time. Screening doesn't necessarily need to wait for a free space in someone's calendar. Candidates don't need to sit in a queue while recruiters work through hundreds or thousands of applications. As more of the surrounding workflow becomes intelligent and automated, administrative delays can start disappearing too.

We've seen this translate into a 43% reduction in time to hire in one global programme and a 26% improvement in hiring speed in another. Those numbers matter not just because the organisation is operating more efficiently, but because days of waiting have disappeared from someone's experience of trying to get a job.

What's particularly important is that speed and automation don't have to come at the expense of candidate experience. Across different programmes we've supported, we've seen 98% of candidates say the experience improved their perception of the employer, while another programme achieved an average candidate satisfaction score of 8.4 out of 10.

That's an important distinction. The ambition shouldn't be to make hiring faster for the organisation by making the experience colder or more transactional for the candidate. The opportunity is to use technology to remove friction for both.

Maybe we don't need to automate every step. Maybe some steps shouldn't exist.

There is a temptation when introducing AI to take the hiring process we already have and automate each part of it. But I think some of the most interesting conversations happen when we stop asking how to automate a step and start asking whether that step needs to exist at all. We made the same point in AI doesn't fix broken processes, it just runs them faster.

If we can create strong, structured evidence about a candidate earlier in the process, do we still need every interview that traditionally came afterwards? If screening and assessment can happen together, do they always need to exist as separate stages? If information can move automatically between systems, do we need the same administrative hand-offs?

The best transformation projects don't just make an old process faster. They create an opportunity to question why the process looked that way in the first place. This is where I think AI becomes particularly interesting in hiring: not because we can automate ten steps, but because we might discover that we only need six.

Every unnecessary interview that disappears gives time back to a recruiter or hiring manager and to a candidate. Every hand-off that disappears removes an opportunity for delay. Every redundant stage that disappears makes the experience simpler. Sometimes the most valuable thing technology can do isn't perform a task faster; it's make the task unnecessary.

Did inconsistency disappear too?

Not everything worth removing can be measured in hours. One of the less visible problems in hiring is inconsistency. Two recruiters can interpret the same CV differently, two interviewers can ask completely different questions, and hiring managers can apply different standards to candidates applying for the same role.

Human judgement will always have an important role in hiring. Maki's own approach is designed around AI supporting rather than making final hiring decisions, with structured criteria providing evidence for recruiters to use alongside their own judgement. But keeping humans in the process doesn't mean every part of that process benefits from being unstructured.

Used well, AI can create greater consistency in how candidates are screened, assessed and interviewed. The same criteria can be applied more systematically, evidence can be captured in a structured way and hiring teams can make decisions with a more consistent foundation. That's also central to how we think about assessment at Maki: structured, role-specific evaluation designed to provide more consistent evidence for hiring decisions.

Again, though, the interesting measurement isn't simply whether candidates completed an AI assessment. In one programme we've supported, candidates with strong assessment results were 50% more likely to be hired. In another, changes to the hiring process contributed to a 7% improvement in quality of hire.

That's where the conversation starts moving from automation into decision quality. The question isn't simply whether AI participated in the process. It's whether unnecessary variation disappeared from it and whether the signal we created actually meant something later in the hiring journey.

And eventually, did bad outcomes disappear?

This is where measuring AI gets harder, because some of the most important outcomes happen long after the AI interaction itself. Time to hire is relatively easy to track, recruiter hours can be modelled and candidate experience can be measured quickly. But if we're serious about measuring hiring transformation, eventually we have to look at what happened to the people we actually hired.

Did early attrition fall? Did quality of hire improve? Are more people succeeding in the role? Did we reduce the number of candidates who looked strong during recruitment but weren't actually a good match once they started?

In one global programme we've worked on at Maki, employee attrition in the roles using the new hiring process decreased by 25%. In another, we saw that 7% improvement in quality of hire.

For me, this is where the business case for AI in hiring becomes much more interesting. Saving thousands of recruiter hours is valuable. Hiring faster is valuable. Giving candidates a better experience is valuable. But if the people coming through that process are also more likely to succeed and stay, we're no longer just talking about recruitment productivity.

We're talking about whether the hiring process itself is producing better outcomes.

Five AI hiring metrics that measure what disappeared

Reduced to a dashboard, these are the five AI hiring metrics I would put on it.

• Recruiter hours returned. Human time no longer spent on screening and administration, measured end to end: the 232,000 hours above.
• Candidate waiting time removed. Days from application to a meaningful decision: the 43% and 26% above, read alongside candidate satisfaction (98%, 8.4 out of 10).
• Steps removed. Interviews, hand-offs and stages the process no longer has. Ten steps becoming six is a transformation.
• Consistency gained. Whether early structured evidence predicts later outcomes: the 50% above.
• Attrition and quality of hire. The slowest metrics and the ones that matter most: the 25% and 7% above.

Measure what changed, not just what happened: the AI hiring metrics that matter

None of this means we should stop measuring adoption. We need to know whether people are using the technology, how many candidates are moving through it, whether completion is healthy and whether a programme is scaling successfully. Those numbers give us important information about reach and adoption.

But they should be the beginning of the measurement conversation, not the end.

For every adoption metric, I'd like to see an outcome sitting beside it. If 100,000 candidates completed an AI interview, what happened to recruiter capacity? If 20 markets are live, are candidates moving through those processes faster? If 90% of recruiters have adopted the technology, which manual tasks are they no longer doing? If an assessment is being used at scale, is it helping us identify people who are more likely to succeed? You can size those levers for your own organisation before a programme starts.

Because ultimately, the goal of AI transformation shouldn't be to insert as much AI as possible into hiring. It should be to remove the things that make hiring slower, harder, more inconsistent and more frustrating than it needs to be.

Maybe that's recruiter hours spent on repetitive screening. Maybe it's days of candidate waiting. Maybe it's an interview that never added much value in the first place. Maybe it's inconsistency in how people are evaluated. Or, ultimately, maybe it's early attrition and poor hiring decisions.

So yes, measure adoption. It tells us whether the technology is being used.

But if we want to understand whether AI is actually transforming hiring, start measuring what disappeared.

Frequently asked questions

What is the difference between AI adoption metrics and AI hiring metrics?

Adoption metrics count usage: interviews completed, candidates assessed, markets live. AI hiring metrics measure what changed as a result: the five above.

Which AI hiring metric should a talent team track first?

Recruiter hours returned, measured end to end, because it is the fastest to see and the easiest to fake by moving work around. Agree the baseline for attrition and quality of hire before the AI goes live; they take months.

‍