Modifying an evaluation: the manager's perspective that completes the AI analysis

An AI-produced evaluation is not an automatic verdict: it is an objective, exhaustive base on which a human decision rests. Modifying an evaluation lets a manager adjust the score of any criterion by bringing their perspective: the context, the knowledge of the team, the specific situation. The AI analysis stays visible, the human perception is added next to it, every decision is traced, and the final score feeds the dialogue with the advisor. The human at the heart of AI quality monitoring.

A manager editing an evaluation

Key takeaways on editing an evaluation

  • An AI-produced evaluation is not an automatic verdict: it is an objective, exhaustive base on which a human decision rests.
  • Modifying an evaluation lets a manager adjust the score of any criterion by bringing their perspective: the context, the knowledge of the team, the specific situation.
  • The manager keeps the final say, and that decision feeds the dialogue with the advisor.
  • The AI analysis stays visible: the human perception is added next to it, never instead of it. Both readings coexist.
  • Every decision creates a version: who, when, why. The history is complete and nothing is lost.
  • This is human oversight at the heart of the system: the final score stays owned by a responsible person.

A score is not a verdict, it is the start of a conversation

AI conversational analysis has lifted quality monitoring's historical ceiling: you no longer evaluate a sample of a few percent of conversations, you evaluate them all, with the same grid and the same consistency. That is a base of unprecedented rigour: objective, exhaustive, reproducible.

But a quality score never lives alone in a spreadsheet. It lives between people: a manager who steers, an advisor who grows. The real question is therefore not "is the score automatic?", but "who owns the decision, and how do we make it talk to the field?". That is exactly what modifying an evaluation is for: moving a rigorous analysis from the status of a finding to that of a shared decision.

What the human perspective adds to the analysis

An AI evaluates what is said in the conversation, against your grid. A manager knows what is not in it: a particularly difficult customer that day, a recent internal instruction, an exceptional situation, an advisor's trajectory over the past weeks. This context is not a weakness of the analysis, it is the ground of management.

Modifying an evaluation means inscribing this human perspective into the score. The manager does not repair anything: they complete a machine reading, rigorous but general, with a human reading, sensitive to the specific case. The AI brings scale and consistency, the manager brings judgement and context. Together they are worth more than each one alone.

Modifying means contextualising

The whole point is to bring that perspective precisely, without redoing everything. In Raisetalk, modifying an evaluation means adjusting the score of a specific criterion, keeping the rest of the analysis intact.

The gesture is simple and explicit: the manager takes the criterion in question, chooses the value that matches their reading (met, not met, or not applicable), and explains their decision in a few words. The overall score recomputes immediately. This is the logic of hybrid analysis, pushed all the way to the decision: the automatic and the human in the same flow, with no break.

The AI analysis stays, the human perception is added

This is the point that changes everything for explainability. When a manager adjusts a criterion, the AI's original analysis is not erased: the human decision is added next to it, with its comment.

You therefore keep both readings: the analysis of the conversation as the AI produced it, and the manager's decision with its justification. For a system that must be able to explain itself, to an advisor as well as to an auditor, that is invaluable: you see the objective base and the perspective that put it in context.

Modifying is not rewriting the analysis. It is inscribing into it the perspective of a manager who knows their team and their field.

A decision shared between manager and advisor

A score is only worth something if it is understood and accepted. Modifying is therefore not the end of the chain: it feeds the exchange with the advisor.

This dialogue takes several complementary forms. The advisor has a right of reply: they can contest the score of a criterion they find unfair, and a manager arbitrates. The debrief moment formalises the coaching exchange around the evaluation. Modification, contestation and debrief together make up the same thing: a human conversation around a common analysis base. The AI gets everyone to agree on the facts; people decide what to do with them.

Who decides: a governance matter

Not everyone adjusts everything. The decision is reserved for manager profiles, and limited to their team scope.

ActionAdministratorSupervisorAdvisor
Adjust an already-scored criterion✅✅❌
Manually score criteria left as N/A✅✅❌

This split locks no one out: the advisor has a voice through contestation, the manager decides and owns it, and everyone knows who carries what. The final score is always tied to an identified human.

Nothing is lost: the version history

Every decision creates a new version of the evaluation. The history keeps, for each version, its reason (modification, contestation, debrief adjustment), its overall score, its date and its author.

You can reopen any version to see the evaluation as it was at a given moment, then return to the current one. Concretely: a complete audit trail, and the ability to explain not only today's score, but the whole chain of decisions that led to it.

The human at the heart of AI evaluation

The value of AI quality monitoring is not to replace human judgement, but to give it a reliable, exhaustive and consistent base to lean on. The manager brings what the analysis alone cannot have: the perception of context, the knowledge of people, the sense of decision.

Keeping the final say with the human, keeping the machine's analysis alongside, tracing every decision, and making the score a starting point for the dialogue with the advisor: this is what turns a score into an evaluation the teams buy into. It is also the principle of human oversight at the heart of a responsible use of AI.

Raisetalk evaluates 100% of your conversations and lets your managers keep control. To see how, discover automated quality monitoring or request access.

Blue tokens spread over a board divided into four zones, a single coral token set apart in one corner
Quality Monitoring 12 min

Evaluation disputes: what they tell you about your quality scorecard

A heavily disputed criterion is not necessarily a badly written one. Dispute rate and acceptance rate read criterion by criterion, the cause selected by the arbitrator, response time as a median: the method for knowing which criteria to review and which to explain, and how the Raisetalk Disputes report puts it into practice.

A row of blue chairs facing each other in pairs, one coral pair in the centre
Quality Monitoring 12 min

Tracking coaching: checking that debriefs take place, and what they change

Evaluating is not coaching. The number of debriefs says neither whether they were completed nor which advisors were seen. Stages, times as a median, team coverage, criteria worked on before and after: the method for tracking coaching, and how the Raisetalk Debriefs report puts it into practice.

A queue of blue call bubbles, two of them coral and joined by an arc
Quality Monitoring 14 min

First call resolution rate: measuring it on your customers' real callbacks

A customer who calls back about the same thing was not served on the first call. Advisor-entered status, instant survey, same number within 7 days: the usual FCR measures mislead. The method for a reliable rate, and how Once & Done in Raisetalk applies it to 100% of calls.

A coral speech bubble with jagged edges, picked out in the middle of a stream of blue bubbles
Use Case 12 min

Customer incivility: identifying it in calls to alert and support advisors

An angry customer is not an uncivil customer. Insults, threats, discriminatory remarks: incivility targets the advisor as a person, and most of the time it goes unnoticed. A three-step method to identify it on 100% of calls, alert the right manager and support teams, without measuring emotion.

Three screens of the new Raisetalk interface: dashboard, evaluation page, contestation panel
Conversational Analysis 7 min

The new Raisetalk interface: conversational analysis moves up another gear

Raisetalk unveils its new interface, delivered the weekend of 1 November: more modern, clearer and, above all, more productive, built on the foundations that make it strong. A dashboard for every role, every score proven in one click, one way to search everywhere, and everything you rely on still in place.

A coral label is tied by four threads to four distinct objects set side by side
Quality Monitoring 16 min

Quality monitoring: what the term really covers, and why two tools carrying that label do not do the same job

Scoring advisors, checking regulatory compliance, listening to the voice of the customer, coaching a sales force: four disciplines circulate under the "quality monitoring" label. They do not have the same users, the same scorecards or the same standards of proof. The framework for telling them apart, why a tool does not carry over from one to another, and how to write a specification that does not mix them up.

Upload a call. Read its analysis. Five minutes.

Sign up in seconds, with no credit card and no cookies. A welcome tour walks you through the key principles, then you add your first conversation.

Early access · Your feedback shapes the product