Trust & data handling

What ClaimEvidence is

ClaimEvidence is an evidence-preparation tool for R&D tax advisers. It reads a client's engineering records and produces a draft evidence pack in which each statement cites the record it comes from. It is used by, and reviewed by, a qualified adviser.

It is not a tax adviser. It doesn't decide whether work qualifies as R&D, calculate qualifying expenditure, complete or submit the Additional Information Form or Company Tax Return, or communicate with HMRC. Responsibility for the claim stays with the adviser and the claimant company.

Live, in pilot, not offered

We only call something live once it has run against real material in production. ClaimEvidence shares its engine with our SprintReady product, so some parts are proven and others are new. Here is the honest split.

Live in production today, in the shared engine

• Reading Word and Excel documentsExtraction and cleaning of specifications and spreadsheets, proven in SprintReady production runs.
• Claude on Amazon Bedrock in the EUAll AI processing runs through AWS's EU cross-region profile from our Stockholm-region infrastructure.
• Source-locked generationThe model is restricted to the supplied material and must flag missing information rather than fill it. In SprintReady this produces explicit data gaps.
• Deterministic checks after the modelRule-based checks that run on the model's output, such as SprintReady's check that every number in a ticket appears in the source.
• Traceability recordsA permanent record of what was produced, from which source, and when.
• Tenant isolation and account securitySeparation enforced on our server on every request, encrypted credentials, two-factor authentication, hardened sessions.

In pilot, being built with adviser partners

• Grouping records into candidate projectsThe clustering step for an accounting period's ticket history.
• The evidence-pack formatBaseline, uncertainty, attempts, outcome, exclusions and questions, shaped by adviser feedback from back-tests.
• Character-level quote verificationThe rule-based check that every quoted passage matches its source.
• Routine-work exclusion with reasonsFlagging likely routine work for the adviser to confirm.
• A UK reference setHMRC's published guidance on the meaning of R&D, used to structure the pack around the questions advisers must answer, never to decide eligibility.
• PDF and Jira/Azure DevOps exports at scaleBuilt in the shared engine; being verified on real claim-period exports.

During the pilot, back-tests are run by our team on your behalf, and we tell you which steps were checked by hand. We'll update this page as each item moves to live.

Not offered

• Tax advice or eligibility decisions
• Claim calculation, time apportionment or AIF submission
• US, Canadian or other non-UK schemesNot in scope for now.

Where data is processed

Application, storage and processing run on Amazon Web Services in the EU (Stockholm) region. AI processing uses Anthropic's Claude model through Amazon Bedrock, via AWS's EU cross-region inference profile, which keeps requests within AWS's EU regions. We don't run our own data centres; physical and network security of the underlying infrastructure is AWS's, under its own published certifications.

The UK recognises the EU as providing adequate protection for personal data, so processing in the EU doesn't require additional transfer safeguards for UK data.

This marketing website is hosted separately by Hostinger and holds no client records. It loads no third-party trackers, its fonts are self-hosted, and optional cookies only run with consent.

AI models and training

AWS. AWS's Bedrock documentation states that customer inputs and outputs are not shared with the model provider for its own purposes and are not used to train or improve the underlying models. Any temporary retention on AWS's side is limited and tied to abuse prevention, not training.

Anthropic. Anthropic's Commercial Terms state it does not train its models on inputs and outputs from commercial and API use by default. That is separate from Anthropic's consumer chat product, where different settings apply.

Us. We don't use your clients' records or your packs to train any model, and we don't use them for anything other than providing the service to you, unless you agree otherwise in writing.

Keeping clients separate

• Each adviser firm, and each end client within it, is a separate tenant. Access is resolved on our server from a signed session, never from a value a browser supplies.
• One client's records are never used as context for another client's pack.
• We don't need live access to your client's Jira, Azure DevOps or code repositories. Exports are enough. If you do connect a system later, its credentials are encrypted at rest and never shown again after entry.
• Records can be anonymised before they reach us: replacing client, product or staff names with placeholders doesn't stop the pack working.

Retention and deletion

• Back-test source material (exports and documents you send us) is deleted within 30 days of delivering the back-test report, unless you ask us in writing to keep it.
• Evidence packs and run records are kept for as long as you're a customer, because they are the audit trail behind your work, and deleted on request or within 90 days of the end of your agreement.
• Backups containing deleted material roll off on their normal cycle and aren't restored except to recover from an incident.

If your engagement or your client's needs a different retention period, we'll agree it in writing.

Data protection roles

Ticket histories and documents can contain personal data, typically engineers' names, comments and dates. For that data, the adviser firm (or its client) is the controller and Sprint Ready Ltd acts as processor, processing only on documented instructions. We offer a data processing agreement meeting UK GDPR Article 28 for every engagement, and will sign a mutual non-disclosure agreement before any material is shared. Our sub-processor for client data is Amazon Web Services; we'll tell you in advance of any change.

For data we hold about you as our own customer, see our privacy policy.

Your professional obligations

The PCRT bodies' January 2026 guidance on AI says advisers remain responsible for their work, should keep appropriate oversight of AI tools, should consider transparency with clients, and should be able to explain how a conclusion was reached. ClaimEvidence is designed to support that:

• Oversight: every output is a draft; nothing reaches HMRC unless you put it there.
• Explainability: every statement is tied to a dated source item, and each pack records its inputs.
• Competence and due care: unsupported points are flagged, not smoothed over, so review focuses where it's needed.
• Transparency: this page is written so you can share it with your client.

Using ClaimEvidence doesn't by itself satisfy PCRT or any professional standard; how you use it within your firm's policies does.

Sources

• AWS: Amazon Bedrock data retention
• AWS: Amazon Bedrock security, compliance and privacy
• Anthropic: Commercial Terms of Service
• Anthropic Privacy Center: Is my data used for model training?
• ICAEW: PCRT guidance on using AI in tax
• SprintReady: Trust & Compliance, including production testing evidence for the shared engine