Why “intuition” should be a tech category everyone wants to solve
When we started building our Smart Drafter in 2018, we thought we were solving "the drafting problem" for transactional lawyers.
We started where everyone starts: data. Break contracts into consistent chunks. Arrange them into a searchable database. Build a taxonomy system where provisions can be labeled, searched, and retrieved. The idea was to give junior lawyers the same information as senior lawyers, and watch their performance level up.
But, as it turns out, contract drafting isn't a single problem. Even for transactional lawyers, it's several distinct problems that just happen to sound similar.
Problem one: "What did we do the last time when we encountered this issue?"
This sounds like a search problem, but it’s actually a context problem.
Even if you break down every provision in every contract, the data won't tell you what "last time" means. Was it the Series B for that medtech startup? The enterprise SaaS deal where regulatory concerns drove decisions? The one where opposing counsel played hardball on reps but folded on indemnities?
The information lawyers need often isn't in the document. It's in the matter file. The deal memo. The negotiation dynamics. The heads of the people working on the deal.
This is a data capture problem. Law firms don't capture the right data. They try, but lawyers do not like spending time labeling and documenting context.
So, how do we capture it? Can we infer it from documents alone?
Problem two: "What's market?"
This one actually is a data problem, but one that would never be fully solved.
When lawyers negotiating and someone asks "what's market", they are really asking: what's the common practice? If someone deviates from this, it is an anomaly. That gives you leverage to pull the other side back to your position.
For this, you need statistics. You can't answer it by reading contracts one at a time. You need to aggregate across deals and say: "In 10 comparable contracts, eight of them include X provision with Y terms, therefore this is market."
But, generally speaking, obody actually has complete market information. The legal industry is structured so nobody can. Most deals are confidential. The whole profession is built on duties of confidentiality that prevent information sharing. The market data lawyers need for negotiation and drafting doesn't exist in any complete form because confidentiality is a feature, not a bug.
The Real Problem Underneath Both Questions
These information and market intelligence problems are really just proxies for the ulterior goal of deal lawyers: how do we negotiate better terms for our clients?
This goes beyond an information problem.
Given limited information — knowing that it will always be incomplete — what does someone do with it? How do you leverage it? How do you know what's right and what's wrong? How do you shape the ingredients in front of you into something that works?
This is art.
The right answer requires intuition. It requires “taste”. That's what a good lawyer offers.
Where Intuition Meets Information
Structured data gets us partway there. You can retrieve a lot of information and put it in front of lawyers. But then come the artful questions:
What is the right information to surface?
How much information do you include and exclude?
What information do they need that they haven't asked for?
Can the software learn from experience and improve? How does it realize over time: "I should be surfacing this new type of information" or "I should be surfacing these two unconnected categories together so the lawyer can make better judgments"?
Information alone is not enough.
We have to deliver information in a way that allows lawyers to leverage their intuition to gain insights.
A lot of GenAI tools today are trying to replace intuition. They're building agents that make decisions and then present insights to lawyers: "Here's what I found, here's what matters."
For commoditized work, this is fine. An AI can take the expertise in a senior lawyer's mind, build it into a workflow, and use that to emulate their intuition. It's a facsimile of expertise, but that is probably sufficient for low-level routine problems.
The complex deals, that requires “art” and “taste”. It is difficult to fully capture and emulate the expertise of a seasoned lawyer.
What Should We Build?
Category one: Use AI to build workflows that automate the commoditized work.
One type of software would be replacing “intuition”.
Take routine expertise, encode it into systems, let AI execute it. Junior lawyers check the output.
Many technology vendors are doing this already.
Category two: Build systems that help lawyers leverage their own intuition.
Another type of software would be enhancing “intuition”,
Surface the right information at the right time. Connect unrelated data points that trigger pattern recognition. Learn what information helps lawyers reach insights faster. Design interfaces that reduce cognitive load so lawyers can focus on judgment, not searching.
Train lawyers to develop intuition faster by exposing them to the patterns senior lawyers recognize instinctively.
Category one and category two are fundamentally different products for different types of work.
Why Choosing a Category Matters
Structured data is step one. You need the database. You need the taxonomy. You need retrieval. But structured data alone doesn't create value; it creates the possibility of value.
The value is recognized when a lawyer (or an AI) looks at that information and recognizes a pattern. Sees a risk nobody asked about. Connects this deal to that precedent in a way that changes negotiation strategy.
Those are insights that win deals.
In 2025, too many vendors were focused on category one — building agents and workflows to automate lawyers away.
In our opinion. lawyers are not going away (not completely). So, the lawyers who remain will need good software that belong in category two — which isn’t trying to automate the answers; instead, it is about how we can help lawyers see and answer better questions.
Can we design technology that makes intuition — not automation — the category we're solving for?
