AI Reached Performance Reviews Before Job Descriptions

Finance leaders are optimistic about AI, and they have every right to be. Today, the technology is alleviating the manual and repetitive work that has long drained their teams for years, enabling them to focus on the work that matters. Deploying AI was treated as a technical exercise and planned as such.

However, as organizations ramped up their investments in and internal use of AI, what didn’t follow for many was a clear understanding of how it would change job roles, responsibilities, and expectations for the employees meant to be using it. Job descriptions were not rewritten, and it wasn’t announced that AI fluency would quickly become part of how employees would now be evaluated. That expectation arrived regardless.

Medius’ 2026 Financial Census surveyed over a thousand finance executives in the U.S. and found that 87% believe their own use of AI now factors into how they are evaluated as an employee. Only 33% say it’s explicit in how they are assessed, while 54% say it possibly counts but are unsure to what extent.

These are senior leaders describing their own reviews, many of them being the same people who set performance expectations for their own teams. When the standard is unclear at their level, it’s unlikely to be clearer further down.

That ambiguity will become a bigger problem as AI becomes increasingly integrated into how organizations operate. Measuring whether people use AI is easy. Measuring whether they use it well and if it’s having a positive impact on their workflows is more difficult. Sequence that work correctly and the productivity gain arrives without the AI fatigue attached.

Capability has to come before accountability

Two things must be put in place before AI enters performance conversations: a written standard and the capability to meet one.

An unwritten standard is not a neutral gap. It leaves managers working from their own interpretations, and when similar performance is scored differently from one manager to the next, the problem is structural. It also leaves those decisions difficult to defend, which matters when it could be the difference between a promotion, a pay adjustment, or a negative evaluation.

Only 30% of organizations in the U.S. describe their AI training as comprehensive, based on structured programs and dedicated time, while 54% call theirs inconsistent or limited in scope. Measure that against the 87% figure on AI as a factor in performance evaluations and the disconnect is clear. The expectation to use AI grows, while the training and support needed doesn’t follow.

Pressure builds as a result, with the same study revealing an overwhelming 75% of finance professionals say AI use has increased fatigue or burnout. Off-the-books use follows too, as 28% of leaders believe their teams use AI tools that have not been formally approved. People will find their own way to meet a standard when nobody has given them a sanctioned one, which, in a finance function means company data moving through tools nobody has vetted.

And when organizations do not build capability internally, they look for it externally. Eighty-six percent now treat AI familiarity as a core requirement or meaningful differentiator in hiring, and 49% of entry-level hires arrive already fluent. That is reshaping the pipeline, with 43% reporting entry-level hiring has risen while mid-level hiring has fallen. While hiring for fluency helps to close the gap, it does nothing for the people already in the building.

Catch more HRTech Insights: HRTech Interview with Emma Lavelle, Chief Operating Officer, UneeQ

Sequencing the work

Inconsistent evaluation and burnout look like one problem, unsanctioned tools and a thinning bench like another, and all of it shares a common root. AI reached measurement before it reached job design or training. None of this requires waiting for the technology to settle, because defining a standard and building the capability to meet it is work HR has always done. HR leaders can start now, and what has to change is the order it happens in:

• Define what good AI use looks like, role by role: This is job design rather than policy. A useful definition says which decisions stay with a person and what judgment the role is expected to apply. It will read differently for an analyst than for a controller. It also shows people how AI changes their role instead of absorbing parts of it, which is the difference between a job that has been redesigned and one that has been hollowed out. When people know what is expected, their effort goes into the work rather than into signaling that they are using AI.
• Build the capability before you assess it: That means structured programs with protected time, so learning happens inside the working day and not in the hours around it, held to the same investment standard as any other capability the business depends on. Managers need that investment too, because they cannot assess judgment they have not had the chance to develop themselves. This is what turns the time AI saves into time people can actually use.
• Measure the outcomes: Evaluate the quality of the decision and the judgment applied when AI gets it wrong. That takes longer than counting tool usage, but it tells you something worth knowing. It also shows you which uses are genuinely improving the work, so you can scale those and stop the ones that are not. Give managers a shared framework and hold them to it. Then, revisit the definition as capability grows, because a standard written for people learning AI will not fit people who have mastered it.

Done out of order, you get what most organizations have now: an expectation nobody can define, and people who are tired of trying to meet it.

Forty-two percent of organizations already have agentic AI operating inside finance processes, and another 46% plan to deploy it within twelve months. The technology will keep moving at that pace whether the people systems underneath it are ready or not.

Closing the gap between what is expected and what people have been equipped to do takes longer than deploying the technology. That is why the work has to start now, before the next wave arrives.

About Medius

Medius links invoice capture, processing, and payment to replace the worry and wondering of managing AP with calm and confidence.