Grund InstituteAI is now a strategic priority in most organizations. Executives buy platforms, approve enterprise licences, set adoption targets and announce ambitious goals for productivity and innovation. From the top, progress looks good. The tools are deployed, training is complete, usage is climbing and departments are reporting new applications.
On the ground, the story can be different. Some people find real gains. Others struggle with tools that do not fit, policies that are unclear, extra checking, and processes that never adapted. A manager believes AI is reducing workload while the team has quietly taken on new work: correcting generated output and supervising automation.
Both views can be accurate. The trouble starts when decisions rest on one of them and the other stays invisible.
The gap is not proof of poor leadership or of employee resistance. It signals that the organization lacks a shared, evidence-based picture of how AI is changing its work. This briefing opens the second theme of the series, on why AI adoption fails inside organizations.
What leadership sees
The reporting view
- Licences deployed
- Training completed
- Usage rising
- Use cases live
- Reported time savings
Adoption is progressing
What employees experience
The working view
- Checking output
- Workarounds
- Manual hand-offs
- Unclear policy
- Exceptions to resolve
The work has changed shape
§Strategy and operational reality
AI adoption is usually judged on what management can easily see: licences, training numbers, usage, implemented use cases and reported productivity gains. These show that adoption activities are moving. They cannot show whether people can work AI into their responsibilities, or whether the expected benefits are arriving.
An employee uses AI every day and spends much of that time verifying its output. A team adopts the approved tool and keeps using another because the approved one lacks something essential. A manager reports a process as automated while employees still resolve the exceptions by hand and move data between disconnected systems. None of this appears in standard reporting.
Research published by Great Place To Work in 2025 found differences between how leaders perceive their AI communication and how well employees understand the organization's AI direction. Communicating a strategy does not guarantee that people know what it means for their work.
An organization can have a clear AI strategy at the top without a shared understanding of it anywhere else.
That matters most when adoption figures drive investment, scaling or workforce planning decisions.
§Five gaps, not one
Differences between managers and employees come from distinct sources. Treating them all as resistance to change hides the operational causes.
The value gap
Management and employees are often measuring different stages of the same workflow. The disagreement is less about the technology than about where the counting stops.
- Leadership sees
- AI-generated reporting as a major efficiency gain, because preparation time has collapsed.
- Employees see
- The preparation saving partly consumed by fact-checking, correction and coordination. Briefings 02 and 03 trace where that time goes.
The workflow gap
AI improves one activity without fixing the process around it. A tool can look like a success while the end-to-end workflow stays inefficient.
- Leadership sees
- An AI-enabled process, successfully implemented.
- Employees see
- Fragmented tools, manual hand-offs, duplicated steps and approval bottlenecks. Briefing 05, AI on Top of Broken Workflows, follows this next.
The capability gap
General familiarity with AI is not the same as applying it well at work. People also differ in interaction style, need for structure and expectations of verification. These are implementation requirements, not deficits in competence or motivation.
- Leadership sees
- A trained workforce, ready to use AI effectively.
- Employees see
- Missing domain-specific guidance, practice and confidence for their real tasks. Briefing 07, Cognitive Fit, explains why one approach does not suit everyone.
The governance gap
A policy can exist on paper without working in daily operations. The result is inconsistent practice, unnecessary workarounds and unapproved AI use.
- Leadership sees
- Clear rules on which tools are approved and how information is handled.
- Employees see
- Ambiguous policy, unclear accountability and approved tools too restricted for everyday tasks. Briefing 10, Shadow AI, looks at what happens next.
The expectation gap
When expectations outrun implementation conditions, management reads slow progress as weak adoption and employees read it as unrealistic targets. Both readings need testing against evidence.
- Leadership sees
- Rapid efficiency gains expected soon after deployment.
- Employees see
- Gains that depend on process redesign, integration, reliable data and time to adapt.
§Why the gap stays hidden
Five structural factors keep organizations from seeing the distance between expectation and experience.
- Top-down implementationStrategy is designed and rolled out with little input from the people who know the workflow, so dependencies, exceptions and requirements are missed.
- Reliance on adoption metricsDashboards show how often people use AI, not whether it improves outcomes. Teams look highly engaged while working through unreliable processes.
- Fragmented feedbackProblems are scattered across managers, IT support, individual teams and corridor conversations. Without structure, recurring issues never surface.
- Pressure to show progressWhen AI is a strategic priority, everyone feels pressure to report good news. No one has to misrepresent anything; optimistic assumptions and loose definitions are enough.
- No comparable evidenceManagers and employees are asked different questions, at different times, with different definitions of success. Their answers cannot be compared.
§What the gap costs
The cost is not the disagreement itself. It is the chance of making major decisions on an incomplete picture.
| Decision area | What the gap puts at risk |
|---|---|
| Investment | Programmes expand on positive adoption figures before anyone knows whether the first implementation paid off. Ineffective practice gets locked in and becomes costlier to undo. See Briefing 12, The Missing AI ROI. |
| Productivity | Reported gains are overstated when management counts task speed and employees absorb the downstream work. |
| Trust | Employees grow skeptical when reported benefits do not match their day. Leadership misses real employee innovations that formal reporting never captures. |
| Governance | The distance between formal policy and actual use creates uncertainty about responsibility and data handling, often unnoticed until an incident. |
| Scaling | A workflow that works for one team is rolled out to others before its assumptions have been tested in their context. |
§Alignment is not agreement
An AI assessment should not try to make managers and employees report the same experience. Their roles give them different information. Managers see budgets, priorities, resource allocation and overall team performance. Employees see workflow interruptions, tool limits, verification effort and exceptions. The two views complement each other.
A manager can rightly observe that the team delivers more since AI arrived. Employees can rightly report that the extra output depends on extra checking and coordination. Both are true. And a manager and a team can agree that adoption is going well while both miss a problem landing on another department.
A shared view, confirmed by workflow evidence. Safe to act on.
Everyone agrees, and everyone is missing something. The most dangerous quadrant.
The difference is genuine. Find its operational cause.
Different visibility or different situations. Gather evidence before concluding anything.
§How to diagnose the gap
We recommend a structured comparison of managerial expectations, employee experience and operational context, in six steps.
- Define the workflow and the objectiveWhat work is done, where AI is involved, what improvement is expected, which teams and roles take part. Broad questions about AI satisfaction say little without this scope.
- Gather both perspectives independentlyManagement describes objectives, expected benefits, decisions, approved tools and perceived results. Employees describe actual use, workflow fit, capability, verification effort and outcomes. Keeping them separate stops one view shaping the other.
- Compare on common dimensionsMap both onto the same dimensions: business value, workflow integration, cognitive fit, governance and role capability. Keep the differences; do not average them away.
- Separate disagreement from uncertaintyNot every difference is a contradiction. One group may lack visibility, or the evidence may be thin. Distinguish supported alignment, supported divergence and insufficient evidence.
- Find the operational causeAccess, expectations, training, fragmented workflows, unclear policy or poorly defined success. Investigate the explanation; do not assume it.
- Turn findings into decisionsProduce priorities for action, not just an alignment score: workflow redesign, revised governance, role-specific support, better tools, clearer responsibility, better measurement. Each with an owner and a way to check it worked.
§One implementation, two views
Take a fictional organization that introduces AI into customer service. Management expects faster response preparation, more consistent answers and more team capacity. After launch, usage looks strong and the manager reports more requests handled. Employees describe a different picture. Generated replies often need correcting for customer-specific details. Account information has to be checked in separate systems. Some replies read well but do not fully solve the customer's problem.
| Management perspective | Employee experience |
|---|---|
| Response drafts are generated faster | Drafts often need correction |
| AI usage is increasing | AI is used, but confidence varies by task |
| More customer requests are processed | Extra verification is needed |
| The implementation looks successful | Some workflow problems remain unresolved |
Illustrative example. It does not represent findings from a GrundMind engagement.
A shallow reading says management and employees disagree about AI. A useful investigation asks whether the extra verification is proportionate, whether customer outcomes have improved, and whether better system integration would remove the recurring corrections. The answer might be that AI creates real value and one part of the workflow needs redesign. Or the evidence might show that the productivity gain disappears once all human effort is counted. The assessment exists to find out which.
§The GrundMind view
GrundMind treats implementation alignment as a diagnostic problem, not a communication problem. Our AI Implementation Alignment (Team Gap) diagnostic compares management expectations with employees' actual experience of AI implementation.
Neither managers nor employees are treated as the definitive source of truth. Both perspectives are brought into one evidence structure and mapped onto the same five dimensions the GrundMind diagnostic uses: business value, workflow, cognitive fit, governance and trust, and role capability. The output shows where expectation and experience align, where they diverge, where evidence is too thin to say, and what is causing each difference.
Results are reported at team level. The diagnostic shows patterns across a team, not who said what, so people can describe their real experience without managing how it reflects on them.
The perception gap also connects the rest of this series. Workslop, the productivity illusion and the verification tax are all, in part, things employees see and leadership does not. Closing the gap is often the first step to measuring them.
§Five questions for the leadership team
- Do we know how our people actually experience AI at work?Usage figures do not answer this.
- Are managers and employees measuring the same stage of the workflow?Many disagreements disappear once both count the same work.
- Where do our AI policies work on paper but not in practice?Workarounds and unapproved tools are the signal.
- Which of our agreements have we actually tested?Shared optimism is not evidence.
- Are we about to scale something we have only seen from one side?Test the assumptions in each context before expanding.
The gap is information. Treat it as evidence, not resistance.
Leaders and employees see AI from different positions, and both see something real. The risk is not that they disagree. It is that decisions on investment, scaling and workforce planning are made from the reporting view alone, while the working view stays invisible.
Organizations that scale AI well will compare both perspectives systematically, on the same workflows and the same dimensions, and act on what the evidence supports.
The question is not whether leaders and employees agree about AI. It is whether the organization knows what is actually happening.
Book an AI adoption diagnostic pilot.
See where leadership expectations and employee experience of AI align, where they diverge, and what is driving the difference, before you scale.
About this briefing
The GrundMind Adoption Briefings are short, leadership-facing notes from the Grund Institute on the measurement and design of AI adoption in organizations. The series accompanies the longer-form Grund Institute white papers on cognitive fit, the Second Mind framework, and the GrundMind diagnostic.
Sources
- Great Place To Work, research on leadership and employee perceptions of AI communication, 2025.
- Grund Institute, Bridging the Cognitive Gap in AI Adoption, 2026.