cohortes

The investigation software

Making testimonies interoperable.

You lead the investigation. The software prepares the material: it anonymises, categorises and extracts the facts with their dates and their source. The reading and the conclusion remain yours.

Three assignments, the same tooling problems

A harassment investigation, a psychosocial risk assessment and a workplace wellbeing survey are not conducted in the same way. But they all produce the same thing: dozens or hundreds of open-ended responses to read, cross-check and report back on.

How an investigation unfolds

You have run a quantitative survey? Here is what happens, step by step.

Step 1 — Import

CSV, Google Forms, Typeform, Sphinx… you drop in the export

Whatever tool did the collecting: they all export a table.

What you no longer do

Copying responses one by one into a working document.

investigations / town hall · workplace climate 2026 / import
climat-social-2026.csv · 312 testimonies

Column selected: “What would you like to say about your working conditions?”

Step 2 — Processing

Every testimony goes through the same grid

What you no longer do

Rereading the testimonies over and over to work out which ones deserve your attention.

processing · 312 testimonies
Anonymisationconsistent pseudonyms across the whole investigation
Summarythe key points of each testimony
Categorisationharassment coding framework · 7 families, 44 categories
Fact extractionwhat happened, when, and in which testimony

312 testimonies processed · 148 dated facts extracted · ready to read.

Step 3 — Reading

You move through the testimonies by theme, by person, by period, and so on.

Every testimony carries its labels. You combine what you are looking for and land on the testimonies concerned.

What you no longer do

Searching a spreadsheet to find out who said what.

testimonies · 312 · filtered: 27
Family: working conditionsPeriod: March — JuneCategoryPersonDated facts only
Witness 07Testimony 4114 March

Monday meetings tense since a change of team; contradictory instructions.

Work organisationInstructionsDated fact
Witness 12Testimony 6321 March

Workload redistributed without notice; two departures not replaced in the department.

WorkloadStaffing
Witness 07Testimony 1182 April

Remark repeated in a meeting in front of the team; no reaction from management.

Working relationshipsIn publicDated fact

Step 4 — Categorisation

Building your own coding framework

  • Inductive categorisation: themes surface from the testimonies, you set aside, you name, everything is recoded.
  • By hand: you create your families and your categories yourself.
  • Ready-made grid: the coding frameworks shipped with the software.
What you no longer do

Categorising every testimony by hand.

testimony 118 · categories
Witness 07 · 2 April
Working relationshipsIn public Dated fact+ add a category

Harassment coding framework · 7 families, 44 categories — editable.

Inductive analysis · themes outside the framework

Reorganisations losing their meaning34
Instructions that change from one day to the next21
Isolation of the evening teams18

Step 5 — Chronology

Every dated fact, tied back to its testimony

The facts are put back in order automatically, and each one keeps the verbatim it was drawn from. That is what makes the debrief defensible: nothing is asserted without a way back to the source.

What you no longer do

Rebuilding the chronology by hand, piecing the dates together one at a time.

people / person 03 · chronology
Person 039 facts · cited in 6 testimonies · March — June

14 March

Contradictory instructions given in a team meeting.

“… the Monday meetings are different now.”

Testimony 41 · Witness 07

2 April

Remark repeated in front of the team, with no reaction from management.

“… in front of everyone. Nobody reacted.”

Testimony 118 · Witness 07

19 May

A live case withdrawn, announced by group message.

“… I found out at the same time as everyone else.”

Testimony 205 · Witness 12

Confidentiality

  • Anonymisation from the first pass: everything that circulates afterwards is based on pseudonymised text.
  • Zero retention: processing runs at OVHcloud, on open source models. The content sent is neither kept nor used to train models — OVHcloud’s commitment.

An investigation under way?

The simplest thing is to watch the tool run on a case that resembles yours. Thirty minutes is enough.

Book a demoor write to us: roberto@cohortes.co