AI consultant
AI for law firms, consulting and B2B services
I help consulting firms, law firms and B2B service businesses put AI to work on their files: sourced research, contract analysis, an automated back office. In production, under your supervision, with your data staying with you.
Where the time goes
In a firm, the raw material is the file, and there are years of them stacked up that no one has time to reopen. The knowledge exists, but it is buried: in PDFs, in email threads, in internal notes, in case law a partner keeps in their head and no one else can reach.
Meanwhile the back office sorts by hand: qualifying incoming requests, keying in data, chasing, updating a CRM that is never quite current. Precise, repetitive, time-hungry work that has nothing to do with the value you actually bill.
- Years of files and case law no one has time to reread.
- A back office that qualifies, keys in and chases by hand.
- A CRM and a set of deadlines that are never really up to date.
What AI can take off your plate
The point is not to replace your judgment, it is to hand back the hours that research and the back office take from you. Here is what works well in consulting and legal.
Under the hood I assemble my existing building blocks for your practice: the sourced copilot is LLM & RAG, triage and follow-ups are Automation & workflows, the guardrails are Reliability & security.
- Sourced research copilot: you query your document base and internal case law in plain language, and every answer cites the material it relies on.
- Contract analysis: sensitive clauses flagged, gaps against your templates, points to watch surfaced, all reviewed by a human.
- Assisted drafting: first drafts of notes, letters and case summaries from your own material and templates, never from a blank page.
- Triage and qualification of incoming requests, routed to the right person instead of piling up in a shared inbox.
- Follow-ups and tracking: deadlines, missing documents, CRM updates, run by supervised workflows.
Governance: sourced, partitioned, confidential
In a regulated profession, an AI is only worth anything if you can trust it. Three principles I set from the design stage.
First, every answer is sourced: the copilot cites the file, clause or ruling it relies on, so you can check before you act on it. Second, permissions are respected: everyone queries only the files they are already allowed to see, partner and intern alike, and the RAG does not route around your walls. Third, confidentiality: your data stays inside your perimeter, hosted in the EU or on your own servers, never used to train a public model. Professional privilege is not a box ticked at the end, it is an architecture criterion (see Reliability & security and Cloud & on-prem).
The legal caveat: AI does not state the law
Let me be clear about one thing: I do not deliver a machine that issues opinions. A language model can summarize, retrieve and connect, but it can also be wrong with total confidence, and on the law a confidently phrased error is dangerous. So I build these systems as assistants, not oracles: the answer is always tied to its sources, and the decision stays with a professional who verifies. On questions of law, human validation is not optional, it is in the loop. And when a use case cannot be made reliable, I tell you instead of shipping it.
A concrete example
At a consulting firm, on an internal document base, the sourced copilot took finding a piece of information from a multi-minute hunt down to an answer in two seconds, sources cited: around 70% less time spent searching, at a quality you can check. The number matters less than the mechanism: it is not magic, it is a well-prepared corpus, a search that cross-checks, and every answer verifiable by the person reading it.
Where to start
You do not begin with months of build. I start with a short audit: which use cases actually pay off for you, what state your data and files are in, and what needs fixing before a single line of code (that is Audit & optimization). Then a focused POC on a narrow scope (one type of file, one team), to prove value on your real material before scaling it up.
You keep the wheel at every step: documented architecture, knowledge handover, no dependency on me alone. I am an independent consultant, a single point of contact, and it is built so you never get locked in.
Do our files stay confidential?
Yes, that is the starting point. Depending on your constraints, the system runs on EU-hosted cloud or entirely inside your own infrastructure with private models. Your material is never used to train a public model, and professional privilege is treated as an architecture requirement, not a footnote.
Can the AI make up an answer on a point of law?
A model can be confidently wrong, that is real and I will not hide it. That is why every answer is sourced, why questions outside the corpus are refused rather than guessed, and why the final call stays with a professional who checks. I do not ship an AI that gives legal opinions, I ship an assistant that saves you time under your control.
Where exactly do the answers come from?
From your documents: files, contracts, internal notes, the case law you point me to. The copilot does not browse the web or draw on the model's general knowledge, it answers from your base and cites the specific source. If the information is not there, it says so.
Can we keep everything in-house (on-prem)?
Yes. I can deploy open-source models directly inside your infrastructure: neither your files nor your queries leave your perimeter. For firms without that constraint, EU hosting with a data processing agreement is often enough. We weigh cost, performance and sovereignty together (see Cloud & on-prem).
How does it fit our existing tools?
I connect your tools (document management, CRM, email, practice software) through their APIs, and through MCP where possible: the connection is written once and stays reusable if you change model. You do not replace your stack, the AI sits on top of it.
How long before a first result?
An audit gives you a first read in a few days. A POC on a narrow scope, on your real material, takes a few weeks depending on volume and the state of your documents. We prove value before industrializing, not the other way around.
Contact
Ready to go from demo to production?
Reply within 24 hours · first conversation free, no strings attached.