Build vs buy for legal AI: a decision framework
Should your firm build its own legal AI or buy off the shelf? A practical framework for general counsel and managing partners, from someone who has done both.
“Should we build this ourselves?” is the question I get asked most often, and it’s almost always asked too early. Build versus buy is a real decision, but it’s the second decision. The first is whether the problem is worth solving at all.
Assuming it is, here is the framework I use.
Start with the source of advantage
Buy when the capability is valuable but not differentiating. Build when the capability is the differentiation itself, or when it depends on data or workflow that is genuinely yours.
Document review, contract extraction, research - these are largely commodities now. Several vendors do them well, and your version would not be meaningfully better. Buy. But the thin layer of judgement that is specific to how your firm works, on your matters, with your precedents - that is where building can pay off, because no vendor can ship it.
Three questions before you write a line of code
- Is the data ours, and is it good? Building on top of messy, ungoverned data is how internal AI projects die slowly and expensively. If the data isn’t ready, fix that first - it helps you whether you build or buy.
- Can we maintain it? A model you build is not a project, it’s a liability with upkeep. Models drift, vendors change APIs, regulations move. If you don’t have someone to own it for years, buy.
- What’s the cost of being wrong? Buying lets you change your mind cheaply. Building locks you in. Early on, when you understand the problem least, optionality is worth paying for.
The honest default
For most firms, most of the time, the answer is: buy the commodity, build the thin differentiating layer on top, and only when you’ve proven the workflow with bought tools first.
The firms that get burned are the ones that build because building feels like control, or buy because buying feels safe - rather than because the source of advantage pointed them there.
A simple test
If a capable competitor could buy the same thing tomorrow and be where you are, you should probably buy it too and compete elsewhere. If they couldn’t - because the value lives in your data, your workflow, or your judgement - that’s a candidate to build.
Everything else is detail.
Written by Dom Conte
Legal-tech founder, builder and speaker. More about me →