Congruent Talk to us
Trust · How we use AI

How Congruent uses AI & protects your data.

Buyers in regulated and board-governed organisations ask us two questions before anything else: how do you use AI, and is our data safe? Here is the straight answer, in four parts.

The maths is deterministic, not AI; your raw data never leaves you; nothing you share trains a model; and an independent security review is on our roadmap.
Pillar 1

The engine computes the scores — not the AI.

Fuzzy-AHP produces every number, reproducible from the raw survey responses. The language model only drafts the written analysis around figures the engine has already fixed — it never computes, changes or invents a figure. A validation step rejects any sentence whose numbers don't reconcile to the engine, and the Diagnostic Lead signs off every figure.

Demonstrable: each report carries an independent figure-validation appendix and a recomputable build fingerprint, so anyone holding the engine outputs can confirm the report was built from them.

Pillar 2

Your raw data stays in your boundary.

Congruent is a measurement platform, not a data warehouse. With bring-your-own-warehouse, raw records never leave your environment — only pre-calculated KPI values and anonymous survey responses cross. No personally identifiable information, no transaction-level detail.

Anchored in the platform's data-boundary architecture.

Pillar 3

Nothing you share trains a model.

We use a hosted model for inference only, with the training opt-out set. Strategy text is held only for the duration of your engagement and purged afterwards, kept in your region.

Exact provider terms and retention periods are available during procurement; the sub-processor register is published in full.

Pillar 4

Built for independent review.Roadmap

Today: per-tenant Row-Level Security isolates every client's data. Planned: an independent security review, then Cyber Essentials Plus → SOC 2 → ISO 27001.

These certifications are on our roadmap and not yet held; we will update this page as each completes.

Document assurance · How we make the numbers trustworthy

And can you trust the document itself?

The four pillars above answer “is our data safe?” This answers the question every board asks next: how does a document earn the right to reach us? Every Congruent report passes a visible gate before it reaches you — every engine figure traced to its source, every narrative number reconciled or flagged, every check auditable.

Check 1

Engine-computed figures.

The AI never originates a number. Every figure is computed by the engine (Fuzzy-AHP, Buckley 1985) and reproducible from the raw responses.

Check 2

Narrative checked.

The analysis around the figures is checked for whether it applies to you and holds together; any prose number without an engine source is flagged.

Check 3

Human sign-off gate.

A named Diagnostic Lead (a defined role; currently the founder) signs off every figure before the document ships.

Check 4

Auditable run record.

The report ships with its own figure-validation record and build fingerprint, so any number can be traced back to its source.

Enforced vs. disciplined — said precisely: the reconciliation is system-enforced — a figure that doesn’t match engine output withholds the report automatically. The Diagnostic-Lead sign-off is a disciplined human step. We are productising the gate into a visible, shareable run record now; the traceability, the reconciliation and the sign-off are how we work today.

For procurement & security teams

The sub-processor register is published in full. The Data Processing Agreement — naming the inference provider, the training opt-out clause and retention periods, verbatim — is available on request during procurement. Email hello@congruentindex.com.

Plain-English glossary

Fuzzy-AHP
Fuzzy Analytic Hierarchy Process — a peer-reviewed decision-science method that turns pairwise judgements into weights and scores. It is arithmetic, fully reproducible, and runs with no AI involved.
BYOW — Bring-Your-Own-Warehouse
KPI calculations run inside your own data environment; only the resulting values leave it, so your raw records stay with you.
Inference only
The model is used to generate text from a prompt; it is not given your data to learn from, and the training opt-out is set.
Row-Level Security (RLS)
A database control that isolates each client's data so one tenant can never read another's.
Diagnostic Lead
The defined role accountable for signing off every figure in a diagnostic — currently the founder.