Before you build a legal AI platform: seven questions every founder should answer
A working prototype is not proof of a workable business. Establish the boundaries, responsibilities and commercial assumptions before the next expensive commitment.
It is easy to understand why founders begin with the technology. The first demonstration is exciting: a document appears in seconds, a research question produces an answer, or a user reaches a resolution without speaking to a professional.
But the difficult question is rarely whether the software can perform the task. It is whether the proposed service, delivered to the intended customers in the intended jurisdictions, can operate responsibly and on commercially sensible terms.
These are seven questions we believe every founder should be able to answer before committing substantial development or launch expenditure.
1. What, precisely, are you selling?
There is a difference between providing general legal information, helping users assemble their own documents, triaging requests, supporting an authorised professional and delivering a service that users will understand as individual legal advice. Marketing language, product behaviour and customer expectations all matter.
Write down the actual user journey: the question asked, the answer produced, what is checked, who checks it and what the customer is invited to do next. If you cannot describe that plainly, the business model is not yet defined.
2. Which jurisdiction applies at every stage?
“UK law” is not one undifferentiated body of law. England and Wales, Scotland and Northern Ireland have distinct legal systems in important respects. Serving customers internationally can introduce additional rules about professional activity, consumer rights, data and marketing.
Start with a realistic launch market, then list any further jurisdictions as separate expansion decisions. A narrow initial scope may be commercially more valuable than promising universal coverage and later discovering exceptions everywhere.
3. Which activities are regulated, and who is accountable?
A platform may use artificial intelligence without being authorised to undertake activities reserved to appropriately entitled providers. In England and Wales, the Legal Services Act 2007 sets out specific reserved legal activities. That is one starting point, not a substitute for assessing the full service and every intended market.
Ask who takes responsibility for outputs, what human review takes place, which complaints process applies, how errors are corrected and where a user is redirected when the matter exceeds the product's proper scope. A disclaimer cannot do the work of a coherent operating model.
4. What happens to documents, confidential information and personal data?
Founders should map the data pathway, not merely purchase an off-the-shelf privacy notice. What information does the user enter? Where is it hosted? Is it supplied to an external model provider? Is it retained, accessed, logged or reused? What happens if a user uploads material belonging to someone else?
Data protection, client confidentiality and security need operational controls alongside documentation. The detail depends on the platform, the people whose information it processes and the processing arrangements.
5. Who actually owns the technology and the right to commercialise it?
Before speaking to investors, co-founders or development agencies, examine existing employment or consultancy obligations, contractor assignments, licences, access permissions, confidential material and any third-party code or data. A founder's personal conviction that an idea is “theirs” is not a substitute for a documented ownership position.
This can be an early-stage question that changes the order of operations: resolve access and ownership first, then consider further collaboration, disclosures and fundraising.
6. Can the model sustain the real cost of delivering the service?
Software usage charges may be the easiest number to forecast. Human supervision, quality control, insurance, referrals, support, complaints, specialist review and ongoing changes can be less visible but more consequential.
Build a working unit-economic model that includes the real cost of a difficult case. Consider what happens when a high-volume user or an unsuccessful output requires more human involvement than the headline price can support.
7. What would a controlled first launch look like?
Before attempting a full launch, define a narrower pilot: the intended user group, supported jurisdiction, use cases deliberately excluded, review procedure, escalation triggers, customer communications and measurable criteria for expansion.
A tightly controlled pilot is not a retreat from ambition. It can be the most sensible way to gather evidence, protect customers and learn whether the commercial model works.
The next investment should be in clarity
At Aubrey International, we have reviewed emerging legal-technology concepts where the important conversation was not about adding another feature. It was about timing, contractual rights, jurisdictional coverage, supervision or revising the intended operational model. We describe this work at a deliberately general level to protect confidentiality.
Our fixed-fee Legal AI Venture Advisory programme begins with a £395 strategy consultation and offers a £995 feasibility diagnostic, a £2,495 written launch blueprint and optional implementation support from £1,495. You can commission stages separately.
Explore Legal AI Venture Advisory
This article is general information, not a legal opinion on any particular platform. Whether activity is regulated or reserved depends on its facts and the jurisdictions involved. Specialist authorisation or advice may be necessary.