A commitment without a safeguard is just a poster on the wall
Seven commitments govern how Dialoga asks, measures and reports. Each one comes with a safeguard in the code that holds it in place, because a method that depends on the user's discipline isn't a method, it's an intention.
Why classic survey practice doesn't fully apply here
The instrument stopped being a form and became a conversation, answered by people who are employed by the company and depend on anonymity to be honest. Three differences change everything that follows.
The interviewer can be audited
In a human interview, neutrality is trained and never verified. Here it is code, and every conversation is recorded. It is the only listening instrument where you can measure whether the interviewer led the answer.
The scale is applied by the system
Employees never see a scale. That removes differences in response style between people, and in exchange creates an obligation: versioning the model's calibration, because here the model is part of the instrument.
Anonymity is a condition, not a feature
It rules out the individual follow-up that classic survey practice prescribes, and forces the cycle to close at the level of the topic. It is not a setting anyone can switch off.
The seven commitments
Click to see what each one means in practice.
We audited our own instrument and found fourteen problems
In August 2026 we reviewed the whole platform against established principles of academic survey methodology, such as order effects, acquiescence bias and care with averages that lack distributions and segments that are too small. It was a review we did ourselves, not a third-party certification. It found fourteen problems, and all of them were fixed. We publish it because a method that has never been wrong hasn't been looked at closely.
There were two NPS figures, and the dashboard showed the worse one
A number called NPS was calculated from the satisfaction score for each topic. It wasn't the NPS question, the unit of analysis was wrong (someone who talked about six topics counted six times) and there was no distribution. It became a satisfaction index by topic, with a name that says what it is.
eNPS came at the end of the conversation, the worst possible spot
Because of the part-whole effect, a general question placed after specific items arrives tainted by whatever was just brought up. It was moved to right after the opening, and the agent is not allowed to repeat it at the end.
Averages without distributions, everywhere
Lukewarm consensus and a team split between extremes produce the same average and call for opposite actions. The distribution now comes with the number.
Fragile segments looked as authoritative as the rest
There are two independent reasons to require a minimum number of respondents, and they call for opposite things: anonymity protects the person, so it hides; margin of error protects the conclusion, so it labels. Hiding data for statistical fragility would systematically erase night shifts and outlying units.
The full review, with the fourteen findings, the paper that names each one and the place in the code where the fix lives, is available on request at the Trust Center.
What the risk report contains
- Inventory by risk factor and by area
- Calculated severity, with the methodology printed next to the number
- History across cycles, keeping the instrument version
- Action plan measures linked to the factor that triggered them
- Segments hidden for anonymity are declared, with the reason
Serious warning signs don't wait for the cycle
When a conversation explicitly mentions a serious risk, the system flags it and alerts leadership right away, on the channel the company set up. It doesn't wait for the end of the cycle or depend on someone opening the dashboard.
The method has a version, and it is printed
Each cycle records which version of the method produced it. That is what lets us say with confidence what can be compared between two periods, and what can't.
A method is something you can check
Talk to Laura for a few minutes and see the instrument at work. It is the most direct way to judge whether it holds up to your scrutiny.