Every AI proposal assumes your data is fine. We check before you sign.

DataTruth is the independent second opinion on AI investments. Before a company signs an AI proposal or commits an AI budget, we compare the systems it runs on, record by record, and say in writing what the data can actually support: sign, fix first, or walk away. We sell no AI, no software and no fixes, and our fee is fixed before we start.

We are new, and we would rather say so than dress it up. That is why the method is published, including the cap that holds a score down when a critical pillar is weak. Every finding points to records your own team can open and check, and the scope line says what we didn’t check. You can stop at the memo. Our first clients get a founding rate in exchange for a private reference call once they’re happy with the work, and only with their consent. We never name a client, fund or target publicly without written consent.

Because that data phase is run by the firm that is paid to build what comes after it, and it starts after you have signed. Ours happens before, when a finding can still change the scope, the price or the decision, and our fee is the same whatever we find. If the two agree, you sign with evidence from both sides of the table.

The facts in your exports on the date we read them: which records match across systems, which don’t, and what that means for the AI in question. Every red flag is checked by hand against sample records before sign-off, and each one points to the records behind it. The memo doesn’t predict how well a model will perform, value the AI plays or audit your accounts; it is not assurance or an audit opinion. The scope line says exactly what was and wasn’t checked.

Any system that exports to CSV or Excel, and the spreadsheets between them: CRM, billing, ERP and finance, HR and payroll, projects, support, and warehouse, transport or production systems where they export records. If two systems record the same customer, person, order or invoice, we can match them. If one can’t export the fields we need, we’ll say so on the call, before you send anything.

The Truth Score is rule-based: the same data, scored on the same date, gives the same score, and no AI model decides it. We do use a language model (Anthropic’s Claude, through its commercial API) to read the plan or proposal you send and to help draft the memo’s wording from our findings. A person checks every line before it reaches you. If you’d rather we didn’t use it on your engagement, say so and we do that part by hand.

Our engine matches the customers, people, deals and invoices across the exports you send. Every red flag is checked by hand against sample records before sign-off. We take a limited number of reads at a time to keep it that way. For a portfolio we agree a sequence, one or two companies in parallel, each memo within 5 working days of complete exports.

No. We never fix what we find and never run the remediation. You get a prioritised fix list, each item with an owner, an effort estimate and a test for done, written to hand to your own team or any outside firm you choose. An AI Investment Review adds the full remediation roadmap.

No access: one person on your team runs standard exports and sends them under NDA and our data processing agreement. We send the exact field list first, and nothing is installed. The full list of sub-processors comes with our data processing agreement, before any data moves, and any transfer outside the EU is covered by Standard Contractual Clauses. Raw exports are deleted after the engagement. We keep quality metrics and the memo, which cites a handful of sample records as evidence. How we handle data

A fixed fee, agreed before we start, that never moves with the verdict. It is a small fraction of the proposal it tests; if it wouldn’t be, we’ll tell you on the call.