Frequently asked questions

Interiu builds analytical systems for financial due diligence and deploys them inside the client's own infrastructure. Clients are advisory and consulting firms and private capital. The founders come from restructuring at KPMG, corporate debt recovery at a bank with over €500bn in assets, and production machine learning in banking. Interiu Report Review is the last read before a draft reaches the Partner. The Interiu Financial Due Diligence Toolkit carries a whole engagement.

Three founders: an ex-Partner at KPMG Deal Advisory who led international corporate debt restructurings totalling €50bn, the head of corporate debt recovery at a top-tier bank with over €500bn in assets, who helped build its in-house forensic unit and ran a distressed corporate-asset portfolio of over €12bn, and an AI engineer who built systems at the largest banks and marketplaces. Full names, records and photos are on the About page. A founder leads every engagement.

No. It removes the part of the work that never needed judgement: rebuilding the databook, tracing a number back to the page it came from, running the same checks a fourth time. What is left is the part your analysts are actually paid for, and there is more time to do it. A firm that uses this to cut its junior bench loses the people who were going to become its Partners.

A software subscription is the same tool for every customer. We build the system around your matter and run it inside your environment. Compared with a Big4 engagement, the difference is economics: AI does the routine data work, so the work is priced without a pyramid of junior hours. Our founders ran this work from inside a Big4 firm and a top-tier bank. Interiu is the unit we wished our clients had.

You would not switch. Those tools stay where they are, and where a firm already runs models under its own enterprise agreement we deploy against that infrastructure rather than introducing another. What a general assistant does not have is your firm's method: the scoping rules, the materiality thresholds, what your review bar rejects and what a Partner will not sign. It also has no memory of the engagement and no obligation to cite what it asserts, which is the part that decides whether a finding can be signed.

Yes, and that is the default rather than an option. The system runs inside your environment. Where the work needs a language model it runs on your own infrastructure, or on the AI platform your firm already contracts for, in your own jurisdiction and under the terms you already negotiated. Only where a firm has neither do we provide the endpoint. In the first two cases the model provider is your subprocessor rather than ours, so there is no new vendor for your risk function to assess.

Yes. Your data is read where it already sits, under your access controls and your retention rules, and we do not ask you to copy anything into a new repository. Where the system is delivered as an agent inside your own Microsoft 365 tenancy, it is built to sign in as the person using it and to act on delegated permissions only, so it cannot open a document, site or mailbox they could not already open themselves. Before any such agent is released we verify from its application manifest that it requests delegated permissions only and holds no application permission.

Client data lives inside the client's own environment and never trains a model. The systems we build are deployed on your infrastructure, pilots included, so client data stays inside your perimeter and does not pass through ours. Where a piece of work cannot be done without data leaving, we say so in writing before it starts and proceed only on your written instruction setting out what may move, where to, and for how long. We keep no copy of it and hold no standing access, and anything that reaches us incidentally, in a support request or a log, is confidential and deleted within ninety days. Where a build requires access, it is named, scoped to the work and revoked on delivery. The no-training rule binds us and everyone under our control, and where we procure the model endpoint we contract that provider to the same prohibition. Where the model runs on your own infrastructure or under your own enterprise agreement, the terms governing it are the ones you already hold. Each of these statements can go into a vendor questionnaire word for word.

By building the controls into the architecture and into the contract, and by publishing both. The control register on the security page sets out every control and what it rests on, and it is published rather than supplied on request. The controls are designed around the areas covered by established standards and regulations. The agreement is published in full, which makes those controls enforceable rather than merely inspectable.

In writing. Interactive AI systems we provide carry this notice: "You are interacting with an AI system provided by Interiu AI. Its outputs are decision-support and require review by a qualified professional before use." Written deliverables carry an equivalent notice. We name AI openly for professional audiences, because questions about reliability and admissibility deserve direct answers, and a hidden method would fail the first adversarial reader.

Every finding is traced to its source, and who reviews it depends on who runs the work. Traced means the finding cites the specific documents, transactions or messages it rests on, so checking it takes minutes, and an adversarial reader, opposing counsel or an auditor, can follow the same trail. Where we produce the report under an engagement letter, a professional with domain experience reviews the findings before it reaches you, and findings that do not survive that review are not delivered. The engagement letter states what that review covers, either every finding or an expressly defined risk-weighted sample. Where your own team runs the system under a subscription, the findings are produced inside your environment and reviewed by your own people, because they never reach us.

By architecture first, and by making every claim checkable. Where the work can run as purpose-written software, it does: executed code does not hallucinate and does not burn token budgets. Language models are used only where genuine deep reasoning is unavoidable. And every finding cites the documents, transactions or messages it rests on, so a wrong claim has to survive a citation check by whoever reviews it, our reviewer in a report we deliver and your own team where you run the system.

We run seeded tests: we plant known errors in real reports and measure what comes back. That shows whether the checks fire and whether they raise anything against text we did not touch. It cannot show full recall on a live report, because nobody holds the complete list of what there is to find. A seeded test is the only kind anyone on this task can actually score.

It starts with a conversation about the matter, then a scoped pilot before any larger commitment. A founder leads every engagement. On investigation work, the target is first findings in days. For custom builds, the promise we make is a first win in weeks. If a short conversation shows we cannot help, we say so.

A fair question, and the answer has two halves. On the work: an engagement lead can be replaced by someone of comparable seniority and relevant experience, and that is written into the agreement rather than promised in a meeting. On the system: it runs on your infrastructure and your data never leaves it, so nothing has to be recovered from us. If we ceased to carry on business, the licence key already sits inside the delivered system and is checked without any connection to us, so it keeps running for as long as the key you hold has left to run, and nothing we do or fail to do can shorten that. When it expires it stops, and we will not call it escrow, because it is not. How long that gives you is the licence period in your order form, which is a commercial term: a pilot bought against a multi-year licence carries that whole period with it. Whatever it is, it is time to decide in rather than time to recover in, because there is nothing to retrieve: your data never left, your own methodology was never ours, and the work already delivered stays yours without limit of time.

Your methodology stays yours and our product stays ours. The agreement keeps them apart on purpose, and it is the part a firm should read first. Your own methods, templates, checklists and thresholds are yours: we take no right in them, we will not reproduce what you give us or pass it to another client, and nothing in the agreement stops you using your own methodology anywhere, with anyone, at any time. What we build to run it is ours, including the part built for you, and it is licensed to you for the term of the subscription. When the subscription ends the system stops. What you keep without limit of time is your data, which never left your environment, and the work already delivered to you.

Pricing is on request, scoped to the matter, and we do not publish prices. Interiu Report Review is a subscription for Partner teams, with pricing on request. For bespoke systems the model is that you pay for the result: the engagement is scoped around a defined first win rather than open-ended hours. A conversation is enough to get a concrete number for a concrete matter.

Two named products alongside bespoke engagements. Interiu Report Review is the last read before a draft reaches the Partner, run inside Microsoft 365 Copilot: figures that disagree between pages, adjustments the underlying accounts do not support, references that lead nowhere, conclusions the evidence in the deck does not carry. The Interiu Financial Due Diligence Toolkit lives inside Word, Excel and PowerPoint, from the information request list to a Partner-reviewed, always-up-to-date report. The Toolkit is shown on a full synthetic engagement. We configure it per firm, it is not a self-serve product.

Both demos are open on the site: no form, no email, no login. The DD Toolkit demo runs a full financial due diligence engagement end to end, and the Report Review demo works a draft report the way the agent does before a Partner sees it. Every company, number and document in them is synthetic or planted for the test, and no client data appears in either. What the demos cannot show is your own data, which is the point of a scoped pilot.