Ready to change.Rewarded for staying put.

Microsoft’s Work Trend Index 2026 finds 65% of surveyed AI-using knowledge workers fear falling behind without new skills. 45% feel safer focusing on current goals than reinventing work. Only 13% say they are rewarded for AI reinvention even when it does not meet results expectations. [Microsoft]

These are answers to different questions, not mutually exclusive groups or a conversion funnel. The survey does not establish that incentives caused a particular business outcome.

Still, the tension is a useful management question. If a team member has a live queue, a delivery target and little room for mistakes, choosing the familiar process may be sensible. Calling that person resistant to change avoids examining the system that made the decision sensible.

Imagine a hypothetical operations specialist who sees a way to shorten onboarding. The improvement requires access to sample records, an afternoon with someone who understands the integration and a manager willing to review exceptions.

None of those resources appears simply because the company bought AI licences. So the specialist experiments after hours, cannot access representative data and produces a promising demonstration that never becomes an operating process.

The management response is another request for innovation. The employee responds with another demonstration. Everyone remains busy. Little changes.

FreeMalta interpretation: this is a permission and capacity problem. Give the work a sponsor, a bounded scope and scheduled time. Treat implementation as work, not as a hobby staff are expected to practise between responsibilities.

A blanket ban encourages workarounds. Blanket permission encourages mistakes. A useful policy distinguishes low-risk assistance from actions with material consequences.

For example, preparing a draft from approved public information is not the same as sending customer data to an unapproved service. Suggesting a refund is not the same as issuing one. Reading a record is not the same as changing it.

Specify allowed data, approved tools, review requirements and escalation routes. Keep the rules understandable enough that someone under time pressure can use them. If interpreting the policy requires a second policy, the first one may need editing.

Provide a safe place to report problems. A person who notices that the system misclassified a case has supplied valuable evidence. Hiding that failure to preserve an impressive dashboard would be the expensive behaviour.

KPMG’s September 2026 Global AI Pulse reports that 53% of surveyed organisations place accountability for AI-informed decisions at C-suite level or above. Its research links maturity with stronger coordination and controls; it does not make a universal causal promise. [KPMG]

For a smaller Malta business, senior ownership need not mean a new committee. It can mean one named person responsible for the objective, the resources and the decision to continue.

Below that level, assign an operational owner who knows the actual process. That person should have enough authority to inspect exceptions and change the implementation—not simply be asked to explain why the tool disappointed.

People can experiment more confidently when the boundary is explicit: this is what you may change, this is what requires approval and this is who will help if something goes wrong. Accountability and autonomy are useful partners. Vagueness is a poor substitute for either.

Keep delivery metrics, but add learning and quality measures appropriate to the pilot. Did the team test the difficult cases? Did it uncover a source-data problem? Can a colleague reproduce the result? Does the full workflow improve after human review?

Do not reward the number of prompts, generated documents or hours claimed as saved without checking what happened to the work. Activity metrics can rise while customers wait just as long.

Separate an unsuccessful hypothesis from careless execution. A properly bounded test that shows a use case is unsuitable can prevent a costly rollout. It deserves a different response from deploying an untested system into a consequential process.

Make credit visible across roles. The person who cleaned the records, the specialist who spotted the error and the manager who removed a pointless approval step may contribute more than the person who presented the demo.

Our suggested pilot agreement fits on a page: the problem, the owner, the allotted time, the permitted data, the actions allowed, the quality standard and the review date. Add a stop rule before anyone becomes emotionally attached to the prototype.

At the review, decide among three options: continue, redesign or stop. Each should be a legitimate outcome. If every pilot must be declared successful, the review is a ceremony rather than an evaluation.

For Malta’s compact teams, visibility can be an advantage. The decision-maker and the person handling the awkward cases may be close enough to review the workflow together. Use that proximity deliberately rather than assume a small company automatically communicates well.

The invitation to innovate becomes believable when the organisation changes what people are allowed, equipped and rewarded to do. Until then, the old process has a powerful defence: it is still the one paying everyone’s performance bonus.