Our approach

QualCheck doesn’t replace reviewers.

Through structured AI-powered analysis, it enables safer, faster, and more consistent human judgement.

In four stages, QualCheck independently analyses your evidence, verifies claims in the draft against it, surfaces content requiring closer human review, and turns the findings into prioritised recommendations — with a quality view across Coherence, Rigour, Utility, and Credibility.

How it works

A walk-through of how QualCheck reviews a report, stage by stage.

You upload
  • Draft report
  • Source evidence
  • Terms of Reference
QualCheck returns
  • Evidence findings
  • Claim checks
  • AI & integrity risk signals
  • Recommendations
  • Quality scorecard
  • Exportable summary
1

Evidence Analysis

The source data is analysed independently — without reference to the draft report — to generate findings relevant to the project’s Terms of Reference. Each finding is assessed for confidence based on data coverage and source strength.

In practice

Independent analysis of interview transcripts finds yield increases reported by 4 of 12 sampled farmers. The remaining 8 reported no change or didn’t comment on yield specifically. This becomes the evidence baseline the draft report is checked against — before anyone looks at what the report itself claims.

2

Report Verification

The draft report is checked against the evidence from Stage 1, testing whether it maintains a coherent analytical thread from the Terms of Reference through to its recommendations. Claims are classified by the strength of their support, and evidence present in the source material but missing from the report is identified.

In practice

The draft states: “The programme delivered cost savings of $2M for the implementing partner.” Checked against the financial data, only $650K is directly attributable to programme activities — the remainder includes cost avoidance from an unrelated organisational restructure. Flagged for the reviewer to separate what the programme actually caused from what happened alongside it.

3

AI Integrity

A detailed content analysis to identify where closer human review may be required — including generic, unsupported, uncited, or stylistically inconsistent sections that may carry higher integrity risk.

In practice

A findings section makes four separate claims about programme reach, each written in a noticeably different style and tone from the rest of the report, with none of them tied to a cited source. Flagged not for any single claim being wrong, but because the cluster of uncited, stylistically inconsistent statements sitting together warrants a closer look before submission.

4

Recommendations

QualCheck synthesises findings from across all stages of analysis into prioritised, actionable improvement suggestions. Recommendations are grounded in specific evidence gaps, alignment issues, and integrity signals identified during review.

In practice

The draft recommends: “Strengthen monitoring systems going forward.” Flagged as generic — no mechanism, owner, or timeframe attached. QualCheck suggests sharpening it to name which monitoring gap, and who’s responsible for closing it.

Overview and Quality Scorecard

A high-level view is provided of where the report stands, across four dimensions — Coherence, Rigour, Utility, and Credibility — derived from the detailed analysis above.

In practice

A draft evaluation report scores Moderate on Coherence and Utility, but Limited on Rigour and Credibility — flagging early that the report’s structure and usefulness are fine, but its evidentiary grounding needs work before submission.

Built for how your team works

Secure and access-controlled.

Your organisation’s data stays siloed — nothing you upload trains other clients or any AI model.

A fully managed workflow.

Teams work together to submit the draft, review findings, revise, and finalise the review — all in one place.

Detail where you need it.

Full findings live in the platform; a summary exports as a PDF.

Ready to see what QualCheck finds in your own reports?

Book a call →