AI agent for law firms that names the document behind every answer
In a law firm time is billed to the minute, and a fair share of it goes on work no client will ever pay for: hunting for an attachment in a mailbox, retyping party details into the next filing, rebuilding a timesheet at the end of the month. An AI agent gathers that material and shows which document it came from. The legal judgment and the wording of the filing stay with the lawyer.
- 14 daysto the first working automation, counting from the end of the workshop
- EUR 3,500net per week of workshop, published up front and not negotiated
- 5 to 10 secondshow long the agent takes to find an answer in your own filings and name the document it came from
The problem
Why does half a day in a law firm never reach an invoice?
A law firm sells a lawyer's time, so every minute carries a price. The catch is that a large share of those minutes goes on work nobody writes on an invoice: finding an attachment in an email from six months ago, retyping client details into another filing, briefing the partner who is taking the hearing on where the matter stands.
One matter runs to dozens of files: pleadings, correspondence with the court, powers of attorney, contracts, scans. They sit in the mailbox, on the drive and in the practice management system at the same time, so the question which version is the current one has three answers. With fifty matters running in parallel that stops being a question of tidiness and becomes arithmetic.
Then there is the layer a law firm cannot skip. Professional secrecy covers everything a lawyer learns while handling a matter, and Article 33 of the GDPR gives you 72 hours to report a personal data breach to the supervisory authority. That is why we ask who has access to what, and for what purpose, before the first line of configuration rather than after the rollout.
- An hour spent assembling a case file that there is no way to bill
- A filing drafted from scratch, because that beats searching the archive for a similar one
- A court deadline held in one person's calendar and in that person's memory
- A timesheet rebuilt at the end of the month out of three weeks of recollection
Why now
What changed that makes this pay off today
Microsoft 365, Google Workspace, practice management systems and time recording tools now expose interfaces you can read data through without exporting anything to a spreadsheet. Three years ago wiring the mailbox, the drive and the matter register together was a project for a quarter. Today it is counted in working days.
The second change matters more. A rule will move data from a form into a register, but it will not answer the question have we used a liquidated damages clause in a construction contract before. A language model searches your own base of filings and contracts and answers in a sentence, naming the file and the date the answer came from. Not from the internet, from the firm's own documents.
The third thing is a boundary we set ourselves and state out loud. The agent does not produce legal advice and does not interpret the law on a lawyer's behalf. It prepares material, points to the source in the firm's own documents and leaves the decision to a person. The wording of a filing is approved by a lawyer, and the responsibility for it stays exactly where it was.
What changes
What is visible after the first month?
Case files assemble themselves
An email with an attachment, a scan from reception and a file from the drive land under one matter the moment they appear. The lawyer opens a matter instead of three windows at once.
The attachment lands in the matter on arrival, not when somebody tidies the file
Answers from your own documents, with the file name
The agent searches the firm's base of filings and contracts and answers in a plain sentence. Every answer carries the document and the date, so it can be checked before anybody relies on it.
Every answer names a document, checking it takes 30 seconds
A new matter registered on the spot
An enquiry from the form opens the matter, the file and the task list, and writes out the parties to run a conflict check on. A person still decides whether to take the matter, but the list is already in front of them.
The conflict check list before the matter is accepted, not after
The deadline reminds you by itself
A court deadline entered once is visible to the whole team and reminds twice, not on the day before. One person's calendar stops being the only place where that date exists.
Reminders at two thresholds and visibility for the team, not for one person
Billable hours ready before the invoice
Time from the calendar, the mailbox and work on documents adds up per matter and per activity as it happens. The lawyer reviews and corrects it instead of reconstructing three weeks from memory.
An invoice basis split by matter and by activity, ready for approval
Proof
Do you use this yourselves?
We do not have a law firm willing to publish its numbers, so we show nobody else's and describe no case that never happened. What we show is our own rollout in a company billed on time exactly the way a law firm is: every hour tied to a client and to a task, profitability counted per project. Below is six months of measurement, together with what these numbers do NOT prove.
Built for ourselves. These are not client deployments but implementations inside the founders' own companies. We say so plainly, because proof you cannot verify is not proof.

JustBottheir agentJustIdea Agency, a performance marketing agency of around 30 people
- 23
- people from the company work with the agent
- 38
- systems wired in permanently
- 112
- processes run without a human
- 475
- thousand lines of the agent’s code
A services company billed on time, with hours recorded per client and per task. The agent is wired into the task system, the CRM, the data warehouse and the analytics tools. First commit 1 March 2026, today 23 people use it.
8 of 38 systems it works in
Google Ads
Meta Ads
- HubSpot
- ActiveCollab
- Ahrefs
GA4
Search Console
BigQuery
- Before
- In the twelve months before launch the agency ran at a margin of minus 4.42 percent. Across a twenty-month window there was not a single run of three profitable months.
- After
- May, June and July 2026 are three profitable months in a row, the first such run in the whole window. Revenue barely moved, so the change sits on the cost of delivery, not on sales.
Margin up 5.55 percentage points year on year, with revenue up 1.5 percent
And in your firm
What if our firm works differently?
Four sentences come up in law firms more often than any others, and all four are fair. The lawful basis for processing, the data processing agreement and what to ask a model provider are written up separately in our guide to AI and GDPR.
Professional secrecy. Client files cannot leave the firm.
Matter data stays in the sources you name: your mail, your drive, your server or a cloud you choose. Access is set per person and per matter, so a trainee sees what a trainee is meant to see rather than the whole archive. We sign a data processing agreement with the model provider, and every query stays in a log: who asked, what about, and which document the answer came from. The scope is written down during the workshop, before the rollout.
I will not have software writing filings and legal opinions for a lawyer.
It will not. The agent assembles material, pulls dates and party details out of the file, finds a similar filing in your archive and shows which document each item came from. The legal judgment, the choice of argument and the wording of the filing stay with a person. The signature under the filing is still your responsibility and nothing here changes that.
We have a practice management system and we are not replacing it.
You do not need to. We read data out of it through its interface and leave it where it is. Replacing a practice management system is a project for a quarter, and the first working automation is due 14 days after the workshop ends.
We are a three-lawyer firm, this is not for us.
At three people a large agent genuinely makes no sense, and that is what we will tell you. The gain sits somewhere else: in registering a new matter and in the timesheet behind the invoice, two things you do every month whatever your size. If the workshop shows there is nothing worth automating, we say so and stop at the workshop.
- Mail and documents
- Microsoft 365, Google Workspace, shared mailboxes, network drives, scans from reception
- Matter management
- practice management systems, time recording tools, repositories of filings and contracts
- Billing
- Fakturownia, sevDesk, Xero, Comarch Optima
- Team tools
- Slack, Microsoft Teams, Notion, ClickUp, task systems
- Data and reporting
- Google Sheets, SQL databases, BigQuery
- Automation layer
- Make, n8n, custom API integrations
What exactly we automate in a law firm
A typical split of administrative work in a law firm before the rollout and after it.
| Area | Today | After the rollout |
|---|---|---|
| Registering a new matter | file and entries opened by hand | matter opened from the form, with a list of parties for the conflict check |
| The case file | filings in the mailbox, on the drive and in the system at once | one address per matter, the document lands there on arrival |
| Finding a precedent filing | searching the archive by file names | the agent answers from your own filings and names the source document |
| Court deadlines | one person's calendar and that person's memory | entered once, reminders at two thresholds, visible to the team |
| Timesheet | reconstructed from memory at the end of the month | built as it happens from the calendar, the mailbox and work on documents |
| Invoice basis | retyped from a spreadsheet into the system | ready, split by matter and by activity, for approval |
| Status update for the client | the lawyer writes a summary from zero | material pulled from the file, the wording approved by a lawyer |
The scope of the first rollout is agreed in the workshop. We take the three items with the highest return, not all of them at once.
How we work
Five steps from the mailbox to the agent
Every step ends with something you can judge before you decide on the next one.
Workshop inside the firm
One week. We walk a matter from the first client enquiry to the last invoice and look for the points where the same information is created twice. Along the way we write down the access scope: who, to which files, and for what purpose.
Picking three automations
From the map we take the three with the highest return and measure them before the start, so that after 30 days it is possible to say whether they worked. In a law firm it is usually assembling case files, registering a matter and the timesheet.
Building the first one
14 days to a working solution. We connect to the mailbox, the drive and the matter system through their interfaces, without replacing anything that already works for you.
Pilot on real case files
The agent answers and a lawyer checks the source it points to. Until the answers land on the right document, we do not move on. This is the step where it comes out whether the base of filings is ordered enough to be usable at all.
Measurement and the next area
After 30 and 90 days we check the indicators agreed in the workshop: how much time goes on assembling case files and how many hours reach an invoice. Only then do we decide on the next area.
Before and after
How a lawyer's week changes
Before
- Half an hour spent finding the current version of a contract in the mailbox
- A filing drafted from scratch, because that beats searching the archive for a similar one
- A deadline held by one person, who is in a hearing right now
- Friday spent reconstructing hours from memory, before the invoices go out
After
- The document found by asking a plain question, with the file name in the answer
- The material for the filing assembled, the lawyer writes the text and owns it
- The deadline visible to the team, with reminders at two thresholds
- The timesheet ready to be reviewed rather than reconstructed
What does not change is who answers for the wording of a filing and for the legal judgment. What changes is how much time goes on gathering the material before anybody starts writing.
For the partner running the firm
What the person signing this off gets out of it
The first thing shows up after a month: the profitability of a matter stops being one person's private knowledge. You can see how many hours went into a matter, on which activities, and how much of that reached an invoice. That is the basis for a conversation about rates and about which matters are no longer worth taking.
The second is quieter, and more useful during an audit or when somebody leaves. The access scope is written down per person and per matter, queries stay in a log, and the history of a matter stays inside the firm rather than in somebody's mailbox. Risk stops depending on who happens to remember.
- Profitability per matter and per lawyer, counted from recorded time rather than memory
- Access to case files written down per person and per matter, with a query log
- Court deadlines held in the system, not in one person's calendar
- A departing employee no longer takes the history of a matter with them
What it costs
The workshop price is published, the rollout is quoted after it
The workshop costs EUR 3,500 net per week and we give that number before the first call, because a firm has a right to know what it is agreeing to. In a law firm one to two weeks is usually enough. The rollout is quoted after the workshop, once we know how many systems we are connecting to and what state the document base is in.
The cost of the model and the tools you pay directly to the provider, with no margin from us. Expect roughly EUR 0.005 to 0.07 per query on the provider's side. Where data is covered by professional secrecy that arrangement also matters formally: the contract is between you and the provider, so it is clear who is entrusting data to whom.
Process workshop
EUR 3,500 net
per week, one to two weeks in a typical law firm
A self-contained piece of work. It ends with a map of the route a matter takes and a 30 day plan you can execute without us.
- A week of work inside the firm and inside the systems
- A map of the route a matter takes, from enquiry to invoice
- The access scope written down, per person and per matter
- Three automations with the highest return, counted before the start
- Most popular
Workshop and first rollout
Quoted after the workshop
the scope follows the number of systems and the state of the document base
For a firm that wants case files assembling themselves within 14 days of the workshop ending.
- Everything in the workshop
- The first automation live 14 days after the workshop
- Connection to the mailbox, the drive and the matter system
- Training for the person who runs it on your side
- Measurement after 30 and 90 days
AI agent for the firm
Quoted after the workshop
for firms where questions about case files come up a dozen times a day
BOND, the company AI agent, answering from your own filings and case files, always naming the document.
- Everything in the rollout
- Search across your base of filings and contracts, with the source named
- Matter status summary: stage, next deadline, documents, hours
- Access per person and per matter, with a query log
The workshop ends with a document that stands on its own without us: a process map and a 30 day plan your own team can run. If after the workshop we conclude that there is nothing worth automating in your firm, we say so plainly and propose no rollout.
Frequently asked questions
What does automation mean in a law firm?
It means a system taking over the administrative work a lawyer or the office does today: assembling case files from mail and the drive, registering a new matter, watching court deadlines, building the timesheet and the basis for an invoice. The data those tasks need is usually already inside the firm, it just sits in three places at once. Automation is about pulling it into one flow, not about buying another piece of software.
Will the AI agent write our legal opinions or court filings?
No, and that is not what we build it for. The agent assembles material, pulls dates and party details out of the file, finds a similar filing in your archive and names the document behind every piece of information. Interpreting a provision, choosing the argument and settling the final wording stay with the lawyer. We also promise nothing about the outcome of a matter, because that is decided by a court and not by software.
What about professional secrecy and GDPR?
Matter data stays in the sources you name: your mail, your drive, your server or a cloud you choose. Access is set per person and per matter, you sign a data processing agreement with the model provider, and every query stays in a log together with the document the answer came from. The scope is written down during the workshop and it is part of the document the workshop produces. The lawful basis is covered in detail in our guide to AI and GDPR.
Where does the agent get its answers and how do I check it did not invent one?
It answers from the firm's documents that you point it at: filings, contracts, case files, templates. Every answer carries the file name and the date, so checking it takes a matter of seconds. If the base holds no grounds for an answer, it is built to say it does not know rather than guess. We test that condition in the pilot on real case files, and until it holds we do not move to production.
Do we have to change our practice management system?
No. We read data out of it through its interface and leave it where it is. Replacing a system like that is a project for a quarter and a separate decision, while the first working automation is due 14 days after the workshop ends. If the system exposes no interface, we say so in the workshop and show what can be done without it, on the mailbox and the drive.
How does watching court deadlines actually work?
A person enters the deadline, because that is their responsibility, but the moment it is entered it stops being a private note: it is visible to the team and it reminds at two thresholds, an early one and a final one. The system watches the date, it does not judge whether the date is right. Checking that the deadline was calculated correctly stays with the lawyer.
How long does it take and what does it cost?
A week of workshop at EUR 3,500 net, then 14 days to the first working automation. A full agent answering from the document base usually takes around 2 months, because it needs a pilot on real case files. The rollout is quoted after the workshop, once we know the number of systems and the state of the document base. Results are measured after 30 and 90 days against indicators agreed before the start.
We run matters for clients abroad, does that change the scope?
It changes the formal layer more than the technical one. We establish where the data physically sits and whether it may be processed outside the area it was created in, and the access scope is written down the same way as for domestic matters. The agent answers in the language of the document, so a filing in English gets an answer in English, pointing at the same source.
Let us start with one matter
The free consultation takes 30 minutes. We walk one of your matters from the client enquiry to the invoice and tell you whether we see anything there worth automating. If we do not see it, we say so on the call. You write directly to Pawel, with no reception and no ticket in a system.
We run at most two rollouts at a time, so that somebody is genuinely sitting on each of them.
Book a free consultationPS. If this week somebody in your firm went through their mail looking for one attachment in order to send one filing, that is exactly the time nobody can put on an invoice. The agent will not write the filing for you, but it will gather the material and show which document it came from.