Evidence review for compliance, quality and risk teams
Find the evidence.Show the source.Say when it isn’t there.
DataStruct works through supplier files, audit records, contracts and registers against the requirements you set. Every finding points to the page, section or spreadsheet row it came from. Where nothing supports a requirement, it says so, and a named reviewer makes every call.
£750 Founding Customer pilot · 30 days · AI processing included
Illustrative finding
RequirementQ-07
Show that measuring equipment is calibrated at defined intervals.
SourceSheet “Register” · rows 41–43
Gauge register 2026.xlsx
| Row | Asset | Last calibrated | Interval | Next due |
|---|---|---|---|---|
| 40 | Torque wrench TW-11 | 02/09/2025 | 12 months | 02/09/2026 |
| 41 | Micrometer MC-204 | 14/01/2026 | 12 months | 14/01/2027 |
| 42 | Height gauge HG-31 | 14/01/2026 | 12 months | 14/01/2027 |
| 43 | Bore gauge BG-09 | 20/01/2026 | 6 months | 20/07/2026 |
Finding
Located is not the same as met. The reviewer records whether the finding, as stated, is accurate.
How findings read
Evidence, related material and gaps are never merged
Key to findings
- …certificate valid until 2 March 2028…
- Evidence located
- A passage DataStruct could support against the requirement, shown with its file, page and section, or its sheet and rows.
- …annual review completed…
- Cited, not confirmed
- A source the answer relied on that DataStruct could not confirm as support. Listed first, so the reviewer opens it first.
- Related material
- Returned by the search and not used as evidence. Kept apart so it never reads as proof.
- [ no supporting passage ]
- No supporting evidence located
- Nothing in the set supports the requirement. The finding says so and names what is missing.
Only a named person can confirm, reject or mark a finding for follow-up. The decision is attributed and logged. Green is theirs alone: no system state is ever shown in green.
How a programme runs
From an evidence set to a reviewed finding, in five steps
- Step 1
Load the evidence set
Contracts, certificates, policies, registers and audit records. PDF (scans are read with OCR), Word, Excel, CSV, HTML and plain text, uploaded in bulk and deduplicated by content.
- Step 2
Write the requirements
Start from the Supplier Compliance Review template or write your own, in your own words. Requirements persist, so next quarter asks the same question the same way.
- Step 3
DataStruct searches and sorts
Keyword and semantic search across the whole set finds evidence even when a document uses different words. Each finding keeps located evidence apart from citations it could not confirm and from related material.
- Step 4
Your reviewer decides
Confirm, reject or mark for follow-up. Every decision is attributed to a named person and logged. DataStruct records no verdict of its own.
- Step 5
Report, then rerun
Export a run as DOCX, XLSX, CSV or JSON. Rerun next quarter: earlier findings and decisions stay on record under the run that produced them.
What it reads
Cited to the page, the section, or the row
- PDFp. 4 · §3.2 Scope of supply
- Scanned PDFread with OCR · p. 2
- Word§ Testing and exercising
- ExcelSheet “Register” · rows 41–43
Also read: CSV, HTML and plain text.
Where teams start
Start with one requirement you already answer by hand
- Read more
Supplier compliance review
“Provide evidence of ISO 9001 certification for the site that makes our parts.”
- Read more
Audit and quality evidence
“Show that nonconforming material is segregated and that its disposition is approved.”
- Read more
Due diligence
“Identify agreements that restrict assignment on a change of control.”
- Read more
Internal compliance evidence
“Show that access reviews for finance systems were completed each quarter.”
Also configured for 11 sectors, from corporate reporting to defence. See industries
Data handling
Straight answers before you ask
- Documents belong to a workspace. Every route that returns content checks the session, workspace membership and permission.
- OpenAI is the only AI provider that receives customer content: passages and extracts, not whole files.
- DataStruct trains no model on your documents, public or private.
- Deleting a document removes it from search, Q&A and every list at once. The stored file is purged on request, and we confirm when it is done.
Founding Customer pilot
£750for 30 days
One fixed price, AI processing included. No self-serve sign-up: every Founding Customer is onboarded personally, because the first evidence set matters more than the software.
- 30 days of access for your team
- Onboarding session and evidence-set setup
- Direct support from the people building it
- Evidence Programmes with reruns and history
- OCR for scanned documents; spreadsheet evidence cited to the row
- Programme reports as DOCX, XLSX, CSV or JSON
Questions
What teams ask first
Does DataStruct decide whether we are compliant?
No. DataStruct locates and prepares evidence and states what it could not confirm. A named reviewer confirms, rejects or marks each finding for follow-up, and that decision is attributed and logged. There is no path for the system to record a compliance verdict of its own.
What happens when the evidence is not there?
The finding says so. It reads No supporting evidence located, or Support not confirmed when only related material was found, and it names what was asked for and not located. Related material is listed separately and never presented as evidence.
Which files can it read?
PDF, including scanned PDFs read with OCR, Word (DOCX), Excel (XLSX), CSV, HTML and plain text. Spreadsheet evidence is cited to the sheet and to the row or row range it came from.
Do you train AI models on our documents?
No. DataStruct trains no model on your content. OpenAI is the only AI provider that receives customer content, as passages and extracts rather than whole files, under API terms that exclude training on API data. We do not hold a Zero Data Retention agreement; the detail is on our security page.
How much does it cost?
The Founding Customer pilot is £750 for 30 days, with onboarding, direct support and AI processing included. After the pilot, pricing is quoted for your team, volumes and industry. See pricing.
Where is our data stored?
Original files are held in Cloudflare R2 object storage under a path scoped to your organisation and workspace. Extracted text and search vectors are held in a managed PostgreSQL database on Railway. The service runs in a single US region; there is no EU or UK residency option today.