At an early pilot, a MENA contact center manager paused before uploading the first call recording to intella. The question was not about integration or file formats: "How do I explain this to the team without making it feel like surveillance?"
That question changed how we approach deployment. A voice AI system can be accurate and still fail when agents and supervisors do not trust it. It may be blocked, or operate quietly while people disregard its results. In either case, the organization gains nothing. Adoption is not a secondary concern. It determines whether a voice AI rollout creates business value.
Why MENA call teams push back
Resistance to call monitoring is not unique to MENA, but local conditions can intensify it. Many contact centers already manage performance under considerable pressure. Agents may feel that minor mistakes carry disproportionate consequences, while having little experience with technology designed to support them rather than score or discipline them.
When leaders announce a new "call analysis AI" system, agents often reach the same immediate conclusion: it will be used to build a case against me. That reaction is not irrational. In many organizations, the answer to "what will this data be used for?" remains as vague as "to improve performance," which agents can reasonably interpret as "to justify firing people."
A second concern is especially relevant for Arabic-speaking teams: accuracy. Agents who speak Gulf dialects may already have seen automated systems mishandle their speech. Interactive voice response systems miss Khaleeji phrases, transcription tools return garbled text, and quality scores can appear to penalize dialect use. After those experiences, agents may expect a new AI tool to fail in the same way and fear that its errors will be used against them.
What pilots have shown us
The teams with the strongest adoption shared one practice: they answered two questions early and plainly. What does the tool measure? Who can see its output, and why?
Trust suffers when those answers stay general. "We're using it to improve customer experience" gives an agent little to work with. A clearer explanation might be: "We are examining why customers who say they are cancelling are transferred three times before reaching the right department. We want to repair that process. Your individual call scores will not enter your performance review."
That degree of clarity means management must settle these questions before deployment. If an organization launches voice AI without defining what it will and will not do with the outputs, it creates its own trust issue. The uncertainty reaches the contact center floor quickly.
The supervisor layer adds complexity
In most contact centers, agents take their cues from direct supervisors rather than senior leadership announcements. When a supervisor doubts the new tool, the team usually follows. If that supervisor treats it as a way to catch people out, the team will recognize that within a week.
We have seen pilots where executives supported the rollout while the operations floor resisted quietly, largely because supervisors were excluded from the early discussion about purpose. Supervisors who did not help define the tool's objectives can become passive blockers. They may not refuse it openly, but they neither use its insights nor explain what the system is doing to their teams.
One effective step is to involve supervisors in defining what a useful system output should look like. A supervisor who helps shape the weekly churn signal report has a stake in the result. That person is more likely to use it and describe it to the team accurately, without creating unnecessary fear.
Let agents see what the system sees
A practice that repeatedly lowers resistance is showing agents what the system says about their calls before, or alongside, management review. If a call is flagged as "high customer effort" because the caller was transferred three times, rather than because of the agent's words, concern about dialect or accent penalties begins to recede.
Leadership must decide who can access agent-level outputs. Should agents see them directly, or should access remain with supervisors? There is no single correct answer. However, giving agents no visibility into what the system says about their calls creates an information gap that can produce mistrust, even when the outputs are never used punitively.
We built intella with configurable access levels in part to address this issue. Senior leaders may need compliance and churn analytics, while agents may benefit from feedback that helps them improve. These are different views of the same data. Choosing where each view belongs is a human resources and cultural decision, not a technical one, and the platform should accommodate that choice rather than impose one model.
The accuracy question needs a direct answer
Agents in Gulf Arabic-speaking contact centers who have encountered poor speech recognition will ask plainly: does the system understand my dialect? intella's honest answer is that the system was built for this problem and reaches 89 to 93 percent transcription accuracy across Gulf and Levantine dialects on our validation set. For this use case, that improves on general-purpose ASR.
This is not about claiming perfect recognition. It acknowledges that specific speech patterns, unusual background noise, and code-switching between Arabic and English can create errors. The system is accurate enough to identify patterns across thousands of calls that human review cannot find, but individual transcripts should remain working documents, not verbatim records. That distinction guides fair use by supervisors and agents alike.
Adoption is a design issue, not a training issue
Most implementations treat weak adoption as a training challenge: improve the user guide, add onboarding sessions, or simplify the interface. Our experience is that adoption problems in MENA contact centers are usually trust problems appearing in the form of training problems. Agents who believe the tool helps them will work through a somewhat complex interface. Those who see disguised surveillance will finish required training and then stop opening the dashboard.
Trust is built before deployment, through conversations among leadership, supervisors, and the contact center floor about the tool's actual use. A platform introduced with a clear purpose and an evident benefit to agents is more likely to be adopted. One presented as a management initiative with unclear effects on individual performance is not, however well it functions technically. Define those uses and access rules before rollout, then review them with supervisors and agents.