Contract data has always been a quiet superpower for in-house legal teams. The challenge most lawyers faced was figuring out how to actually use it. Generic playbooks and off-the-shelf forms only got teams so far. The real value lived in the contracts the team had already negotiated, and most of it sat untapped.
A recent How to Contract webinar dug into that gap. Laura Frederick hosted Danny Di Maria, Co-founder and Chief Revenue Officer at Spellbook, who has spent more than eight years building AI tools for contracting workflows. His perspective made the conversation useful because he had watched thousands of legal teams adopt these tools, and he had clear views on what worked, what stalled, and where the technology was actually heading.
The discussion covered five concrete ways legal teams put contract data to work with AI, from using your own knowledge base instead of generic data, to benchmarking against market terms, to building playbooks from model contracts, to drafting from precedent, to the future of preference learning. Laura and Danny also dug into the rollout mistake that stalled teams out and the skills lawyers needed to stay relevant as the technology kept advancing.
Here are our top ten takeaways from the speakers' comments during the webinar:
Your own contracts were your most valuable training data. Generic LLMs produced generic answers. The real unlock came when the tool had access to your team's contract history, your industry, and your accepted terms. We should think about contract data the same way we think about institutional knowledge. It compounded over time, and it walked out the door if you did not capture it.
Clause search, review standards, and bulk queries were the three high-value workflows. Danny described these as the main ways customers actually used their knowledge bases. We could pull clauses that fit the contract in front of us, flag terms in a third-party draft that fell outside what we had previously accepted, and run plain-language queries across thousands of contracts at once. These three workflows alone justified the setup for most teams.
Benchmarking mattered more for internal conversations than counterparty arguments. Throwing market data at the other side tended to invite a fight. The better use was telling your business stakeholder that the term they wanted was rare in your industry. The conversation became about the data instead of about trust in the lawyer, which was usually a more productive frame.
Pick fallbacks that were close enough to market to actually get accepted. Benchmarks gave you context. If your preferred term was rare and a different term was common, you could choose the common one as a fallback knowing it was more likely to land. That was a smarter use of market data than digging in on a position that nobody else used.
AI-built playbooks got teams to 95 percent of the way there, fast. The old approach of hand-building playbooks took years and rarely finished. Uploading a model document and letting the tool generate a playbook took hours. We still had to review and refine the rules, but the lift was reasonable. The teams that adopted this approach freed up real capacity for actual negotiation work.
Detail mattered when asking the tool for a precedent. Asking for "an MSA" pulled whatever the tool found. Asking for an MSA that was neutral on commercial terms, company-friendly on limitation of liability, and pointed at a specific folder of trusted contracts produced something usable. The discipline of being specific with prompts was one of the highest-leverage skills lawyers could build right now.
The "find your own precedent and upload it" workflow still worked. If you trusted your judgment about which contract to start from, you could do that first and let the AI handle the editing. That kept the human in the loop on the part of the work where judgment mattered most. It was a good model for any AI workflow where the stakes were high.
Preference learning was the direction, even if it was not fully here yet. Danny described a future where the tool understood the team's preferences well enough to pick a precedent, identify leverage dynamics, and red line terms automatically. We were not there yet, but the teams that got comfortable with today's tools would be positioned to use the smarter versions when they arrived.
Start small was the rule for AI rollouts. Boiling the ocean stalled them. The most common failure mode Danny saw was teams trying to configure everything before using the tool for real work. The project stalled, the team lost momentum, and six months in nothing was deployed. We should pick two or three workflows, get them running, build the habit, and expand from there.
Future-proofing came from using AI daily, not from waiting for the perfect tool. Danny was clear that in-house lawyers were not going to be replaced. What was changing was the volume of synthesis, drafting, and editing being handled by AI. Lawyers who built muscle memory around prompting and iterating across the available tools would be the ones who stayed relevant as the technology kept moving.
Subscribe to Stay in the Loop
Our weekly newsletter rounded up the upcoming How to Contract webinars and recaps like this one. Subscribe now to get the insights even if you missed the live session.
