# Cross-check every interview, and see where accounts diverge > Every interview cross-checked: typed, sourced findings with a confidence level, contradictions side by side, and targeted follow-ups to resolve them. Source : https://tellmehow.io/en/features/cross-analysis-and-contradictions · Updated on : 2026-10-18 The analysis turns each validated interview into typed findings linked to their evidence. When two sources contradict each other, it puts them side by side and proposes the follow-up that will settle it. TellMeHow cross-analysis consolidates an engagement’s validated recaps, files, and external sources into typed findings, each linked to at least one piece of evidence and given a confidence level. It detects contradictions between sources and proposes targeted follow-ups, which the consultant approves before they are sent. - Every finding rests on at least one clickable piece of evidence - Targeted follow-ups of 2 to 5 questions, approved by the consultant - Consultant edits are never overwritten by the analysis ### Consolidation happens in someone’s head After fifteen interviews, the consultant cross-checks viewpoints in a spreadsheet. Noticing that accounting and sales describe different discount approvals, or estimating a volume from three rough answers, depends on memory and notes. Discrepancies found late mean calling people back. ### Typed findings, each with its evidence Each validated chapter enriches the analysis without recomputing everything. It extracts typed findings and matches them against existing ones: two phrasings of the same fact merge and pool their evidence. A finding without evidence doesn’t exist. - Types: pain point, re-keying, volume, duration, business rule, tool, key-person dependency, risk, and good practice. - A key-person dependency is phrased by role (“only the person in charge of month-end close knows…”), never by name. - Confidence depends on how many sources agree: a fact both stated and shown on screen outweighs a single statement. - Every piece of evidence points to a timestamped excerpt, a screen capture, a pin, or a page in a file. ### Contradictions, side by side When two sources describe the same fact in incompatible ways, the analysis creates a contradiction: both versions with their evidence, a hypothesis, and a proposed follow-up. For example, sales administration describes approval for every order over €5,000, while sales management says only key accounts need it. The consultant settles it, sends a follow-up, or keeps the contradiction as a finding. - The hypothesis is framed as a question to check: “do two practices coexist across branches?” - “Two practices coexist” is an audit result in its own right, not an error to fix. - An open contradiction caps the confidence of both findings involved. ### Targeted follow-ups instead of a second interview The analysis spots what keeps the map from holding together: a contradiction, a step whose input nobody described, a figure that is too uncertain, a required topic only skimmed. It proposes a short follow-up for the participants concerned. The consultant adjusts the questions and approves; the participant gets a link to a new chapter of their interview. - A follow-up has 2 to 5 questions, with an estimated length of 5 to 10 minutes. - It has its own recap, which the participant validates like any other chapter. - Once validated, it updates the findings and the original contradiction. ### External sources and protected edits An engagement can import data that doesn’t come from an interview, in a public, documented exchange format: metrics, pre-written findings, known processes. The analysis treats them as quantitative evidence that confirms, qualifies, or contradicts the interviews. An imported mailbox analysis might show 210 quote requests a month where interviews estimated 50. - Any tool can produce an external source, and firms can import their own analyses. - The consultant renames, adjusts, approves, or rejects, and those edits are locked. - A new analysis proposes changes to accept rather than rewriting; a rejected finding isn’t proposed again. ### In practice - **Inputs** : Validated recaps, the transcript excerpts and screen images they cite, pins, files, and the engagement’s external sources; never an unvalidated chapter or a test session. - **Cadence** : Incremental analysis on each validated chapter, extracted file, or imported source; full analysis on the consultant’s request. - **Confidence** : High, medium, or low, based on the number of sources, their agreement, their populations, and on-screen observation. - **Contradiction** : Two versions with their evidence, a hypothesis, and a proposed follow-up; the consultant settles it, follows up, or keeps it as a finding. - **Follow-up** : A chapter of 2 to 5 questions, estimated at 5 to 10 minutes, proposed by the analysis and approved by the consultant before sending. - **Exchange format** : Public JSON with a versioned schema, for importing metrics, findings, and processes produced by other tools. - **Safeguards** : Participant names are never passed to the analysis; any statement that assesses a person or mentions an emotion is rejected. ### Frequently asked questions **Where does each finding come from?** Every finding is linked to at least one piece of evidence: a timestamped transcript excerpt, a screen description, a pin, a page in a file, or external source data. A finding that loses all its evidence, for example after a participant removes a point, is deleted. **What happens when two employees contradict each other?** The analysis creates a contradiction showing both versions side by side with their evidence, a hypothesis, and a proposed follow-up. The consultant settles it, sends the follow-up, or keeps the contradiction as a finding, since two coexisting practices are an audit result. **Can the analysis overwrite the consultant’s corrections?** No. Every consultant edit is locked. A later analysis can propose a change, which the consultant accepts or rejects, and a rejected finding isn’t proposed again. **Can we bring in data from other tools?** Yes, by importing it in a public, documented exchange format: metrics, findings, or processes. This data confirms, qualifies, or contradicts the interviews. There are no direct connectors to the client’s systems. **Does the analysis compare employees with each other?** No. It describes processes, phrased by role, with no ratings, rankings, or individual comparisons. Participant names are never passed to it, and any statement that assesses a person is rejected. --- - Français : https://tellmehow.io/fonctionnalites/analyse-croisee-et-contradictions.md