Legal and legaltech
Everywhere else a wrong answer is a bug. In law it is a sanction.
We build legal products a professional can put their name to. Every answer grounded in real sources, every citation verified before a human sees it, client confidentiality treated as an architectural constraint rather than a policy document.
The state of play
The prize is large. So is the professional risk.
There is more recoverable time in a legal practice than almost any other professional service. There is also a rapidly growing list of lawyers who have been sanctioned for trusting a tool that made something up.
2.9 of 8
hours in an average working day are billable. Roughly five hours a day go to work nobody pays for, which is the largest recoverable asset in most practices.
Clio Legal Trends Report, 2024
~240 hrs
a year is what legal professionals expect to free up through AI, worth roughly nineteen thousand dollars each. That is the prize, and it is why everyone is trying.
Thomson Reuters Future of Professionals, 2025
17–33%
of the time, leading legal research AI tools produce answers that are wrong or cite sources that do not support them. Specialised legal products, not general chatbots.
Stanford RegLab, peer-reviewed, 2025
1,600+
court cases worldwide now involve AI-fabricated citations, hundreds of them filed by practising lawyers. Sanctions, referrals to regulators, and public judgments have followed.
AI Hallucination Cases database, 2026
61% → 17%
of UK lawyers use generative AI at work, but only seventeen percent work somewhere it is actually embedded in how the firm operates. Usage is not adoption.
LexisNexis UK survey, October 2025
$4.08bn
went into legal technology in 2025, up seventy-seven percent in a year. The category is being rebuilt right now, and the tools being chosen today will be in place for a decade.
Crunchbase, 2026
Legal products built
to be relied on.
Case and matter management
Files, deadlines, hearings, and client data in one system, with matter-level permissions so the right people see the right things and the audit trail shows who saw what.
Document automation and drafting
Templating and first-draft generation that produces something a lawyer edits rather than something they have to rewrite, working inside the tools they already draft in.
Research with verified citations
Answers grounded in real sources through retrieval, with every citation checked against the primary source before it reaches a human. This is the part most tools skip.
Contract review and lifecycle
Clause extraction, risk flagging, and playbook-driven redlines for the review work that consumes the most hours and produces the least differentiated value.
Client portals and intake
Secure collaboration, document exchange, and intake that qualifies and routes properly, because responsiveness at the front door decides who wins the instruction.
Confidentiality and integrations
Data residency, no training on client material, full audit logging, and integration with the document management and filing systems your practice already runs on.
The current landscape
Where legal work leaks, and where the engineering is.
Five problems every legal practice and legaltech founder is living with, and what building for them properly actually looks like.
The friction
Five hours a day nobody pays for
Only about three of eight hours are billable. The rest goes to admin, chasing, filing, and repetitive drafting that a well-built system could absorb.
What we build
Automation aimed squarely at the non-billable hour, which is the safest place to start because it improves margin without touching the quality of legal work itself.
The friction
AI that invents case law
Independent research found leading legal AI tools wrong or misgrounded between seventeen and thirty-three percent of the time. Lawyers have been sanctioned, named in judgments, and referred to regulators for filing what those tools produced.
What we build
Retrieval grounding against authoritative sources plus automated citation verification against the primary record, so a fabricated authority cannot reach a filing in the first place.
The friction
Client data that cannot leave the building
Confidentiality and privilege are professional obligations, not preferences. Sending client material to a third-party model without knowing where it goes or whether it trains on it is a genuine risk to a practising certificate.
What we build
Architecture that keeps client data where it is allowed to be, contractually excluded from training, with access controlled per matter and every access logged.
The friction
Everyone is using it, almost nobody has deployed it
A large majority of lawyers now use generative AI, but only a small minority work somewhere it is genuinely part of how the firm operates. Individual experimentation is not a system.
What we build
Tools built into the actual workflow, with the supervision, review, and audit that a professional obligation demands, so it becomes firm infrastructure rather than a personal habit.
The friction
Nothing talks to anything else
Practice management, document management, e-signature, and filing systems are where legal work actually lives. A product that ignores them creates another silo and quietly gets abandoned.
What we build
Integration treated as core scope rather than a later phase, so the product joins the stack your team already uses instead of asking them to leave it.
Case study · Bayslope
A legal workflow product built to be checked, not trusted blindly.
We worked with Bayslope to take Techreport99 from working concept to a production platform: automated document workflows, source-grounded output, and a verification step built into the process rather than bolted on at the end. The result was faster turnaround without asking anyone to take the output on faith.
- −68%First-draft turnaround time
- ZeroUnverified citations reaching a client
- 9 wksFrom kickoff to production
How we help you win
Four places a product firm earns its keep.
Not a feature list. The four moves that, on the evidence above, decide whether a legal product gets adopted and stays adopted.
Attack the non-billable hour
The clearest return in legal is not replacing legal judgement, it is recovering the five hours a day that never make it onto an invoice. We build for that first because it pays for everything after it.
Ground it, then verify it
Retrieval alone is not enough, as the research shows plainly. We add citation verification against the primary source, so what reaches a lawyer has already been checked rather than merely generated confidently.
Confidentiality as architecture
Privilege and confidentiality are professional duties, so we design data residency, exclusion from model training, matter-level permissions, and audit logging into the foundation rather than the policy page.
Ship the last thirty percent
Verification, permissions, audit trails, and integration with the systems a practice already runs on are what turn a convincing demo into something a professional will put their name to. That stretch is what we do.
Fit
Whether this is right for you.
A good fit if
- You are building a legaltech product and need it engineered to a standard lawyers will accept
- You run a practice losing hours to admin and repetitive drafting
- You want AI in the workflow but not at the cost of anyone’s practising certificate
- Client confidentiality and data residency are genuine constraints for you
- The product has to work inside the practice and document systems you already run
Probably not a fit if
- You want a general chatbot pointed at case law and shipped quickly
- The plan is to give legal advice directly to consumers without regulatory thought
- Verification is considered a nice-to-have rather than the core of the product
- Nobody on your side can decide what the tool should and should not do
- You need legal or regulatory counsel rather than a product team, which is a different profession
Insights
- ArticleBuilding Client Trust Through Legal TechnologyHow secure collaboration portals are reshaping client expectations and enhancing transparency. Nov 29, 2025
- ArticleThe Future of Digital-First Law FirmsWhy cloud-native tools and automation are becoming essential for competitive legal practices. Nov 29, 2025
- ArticleHow AI is Transforming Legal Case ManagementA deep dive into how AI reduces administrative effort and improves decision-making for modern legal teams. Nov 29, 2025
Common questions
Answered before you have to ask.
How do you stop the AI inventing cases?
By not letting it answer from memory. Every response is grounded in retrieval against authoritative sources, and every citation is then checked programmatically against the primary record before a human ever sees it. Retrieval on its own is not sufficient, which independent research has demonstrated clearly, so the verification step is the part that actually matters. Anything that cannot be verified is surfaced as unverified rather than presented as fact.
Where does our client data go?
Wherever you require it to stay. We design for data residency from the start, use arrangements where client material is contractually excluded from model training, and keep access controlled at matter level with full logging of who accessed what. Confidentiality and privilege are professional obligations rather than preferences, so we treat them as architecture rather than as a clause in a policy document.
Can it work with our existing practice and document systems?
Yes, and for most firms this is what determines whether a tool survives its first month. Legal work lives inside practice management, document management, e-signature, and filing systems, so we treat those integrations as core scope. A product that asks a fee earner to work somewhere else quietly stops being used.
Will it work across jurisdictions?
It depends entirely on the sources it is grounded in, and that is a scoping decision rather than a technical afterthought. Citation formats, court structures, and authority hierarchies differ by jurisdiction, so we establish upfront which jurisdictions the product must serve and build the retrieval and verification accordingly. Claiming multi-jurisdictional coverage without doing that work is how tools produce plausible nonsense.
Do we own everything at the end?
Completely. The repository, the hosting, the data, the document pipeline, and every credential are in your name throughout. If you take the work elsewhere later, nothing is held hostage.
What does it cost?
We price after understanding the scope rather than before, because a document automation tool and a research platform with verified citations are very different builds. A thirty minute call at no charge, then a written scope with the work broken down and a fixed number against it, so nothing is a surprise later.
