Know Your Line Before AI Joins the SaaS Negotiation

By Katie Harris

November 19, 2026

Know Your Line Before AI Joins the SaaS Negotiation

Katie Harris is Lead Solutions Attorney and Commercial Counsel at LegalOn. She previously spent 12 years as Associate General Counsel at PerfectVision. She can be reached at katie.harris@legalontech.com.

It’s 4 p.m. in the last week of the quarter, and a commercial counsel has just gotten updated terms from a long-time Software as a Service (SaaS) vendor ahead of renewal. The business wants signatures by Friday. So she drops the agreement into an AI tool and asks for a review. A few seconds later, it tells her the six-month liability cap is fairly standard and there’s nothing to worry about on AI training.

Great. Ship it?

Not quite. Two years ago, her company held the same vendor to a 12-month cap, and her own data policy says vendors don’t train models on customer data. The tool didn’t know either of those things, because nobody told it. She sends the redlines without knowing.

Meanwhile, across the table, the vendor’s lawyer ran her own review against the redlines. The difference is that her prompt included an attachment of the old deal and laid out their fallbacks, detailing exactly what her company would accept and what it wouldn’t. Both of them had AI. Only one of them had told it where the non-negotiable line was.

Same technology. Very different afternoons. One lawyer brought a tool. The other brought a strategy and let the tool run it.

Almost everyone has access to AI, and plenty of people are getting good at prompting. The harder part, in my experience, is finding time to work out your positions before the redlines come in.

Figuring it out mid-redline is not a negotiation strategy.

The fights don’t really change

If you negotiate SaaS for a living, you know where the fights are: limitation of liability, indemnification, data and IP ownership, AI training and outputs, privacy and security, term and renewal, payment terms, and service levels and warranties.

The liability cap is a good example. A reasonable customer position may be a cap of at least 12 months of fees, a higher super-cap for data breach and confidentiality, and no cap on willful misconduct and fraud. A reasonable vendor position may be one 12-month cap for every claim, with only willful misconduct and fraud left uncapped. A super-cap is where you’re likely to find the compromise. If you walk in knowing your strategy, you close the gap in a couple of rounds. If you don’t, it can drag out, which slows the deal cycle and keeps you reactive and bogged down in redlines.

AI training is newer, and I think it’s where teams are least prepared. A few years ago, it barely showed up in SaaS templates. Customers want to own what they put in and what comes out, and they don’t want their data ending up in someone else’s model. Vendors want to keep improving the product, usually with de-identified usage data, or more if they can get it. There’s a reasonable middle. A lot of customers are fine with de-identified usage data making the software better, as long as their actual data never trains a foundation model. But you can only get to that middle if you’ve figured out ahead of time what components matter and which part of the clause you can give on.

Read the latest thought leadership and analysis from legal experts

Put your positions in the prompt

I spent 12 years in-house reviewing every contract my company signed. Now I build contract review playbooks and train in-house teams to use them. The thing I keep coming back to is that AI will always give you an answer, even when it doesn’t have what it needs to give you the best one. If you don’t give it your positions, it fills the gap with its own idea of what’s normal.

The prompts that work best usually start with good context, like saying who you are and what the deal is, then explain the exact task and how you want the answer laid out. Here’s what our counsel could have written:

“I’m customer counsel reviewing this SaaS renewal. My lines are a cap of at least 12 months of fees, a data super-cap, uncapped willful misconduct, ownership of our inputs and outputs, no model training, and a renewal opt-out no more than 60 days out. I’ve also attached our previous agreement with this vendor. Check the new terms against each line and against the previous agreement, and tell me where they hold, fall short, or are missing. Cite the clause and propose a redline for anything short.”

Now the tool is checking her positions and reviewing her context instead of inventing its own. Vendor counsel can point the same approach at the other side’s markup and ask which requests look firm, which look like openers, where there might be deviations from market standard, and where there’s room to trade.

Then turn it into a playbook

Once you know your positions well enough to put them in a prompt, you’re most of the way to a playbook. A playbook is a collection of prompts that share one strategy: contract review. It gives every lawyer on the team, and every AI review, the same starting point.

To build a SaaS playbook, start with a favorable template . For each recurring term, write down (or bonus points, use an AI notetaker to capture your thoughts aloud) the primary position, the fallback, and the point you won’t go past. Your past redlines are a good place to find fallbacks, since the positions you’ve conceded under deadline pressure are your real fallbacks. Keep the playbook lean and update it whenever a review finds a gap.

Our counsel’s real blind spot was her own history. What cap did we accept from this vendor last time? AI can answer that kind of question across hundreds of agreements at once, as long as you ask specifically. “What liability cap did we accept?” becomes a number you can sort. “Does the contract auto-renew?” becomes a yes-or-no field. Check the source text on anything that will drive a decision.

Sign up for our weekly legal ops newsletter or our new monthly newsletter, The Compendium, which covers a different legal topic each month.

The responsibility is still yours

AI speeds up review, but it doesn’t shift the professional responsibility off your desk. The American Bar Association’ Formal Opinion 512 makes clear that generative AI use is governed by duties lawyers already carry. You need to understand what a tool can and can’t do, protect confidential information, and be able to stand behind any AI output.

Questions worth asking your team

Every SaaS negotiation comes down to a lawyer deciding what the company can live with. If your team can answer these questions, AI will help you hold that decision on every deal in the queue.

  • Have we written down our preferred position, fallback and walk-away point for each term we negotiate most?
  • Does our prompt include those positions, or are we asking AI to guess what’s acceptable?
  • What did we accept from this counterparty last time, and can we find it in minutes?
  • Which of our current vendors can train models on our data?
  • When do the opt-out windows close on renewals in the next two quarters?
  • Who sets the standards AI checks against, and who signs off on its work?
  • Is the tool we’re using approved for privileged and confidential deal information?

AI can help you get to a signature faster. But at the end of the day, the duties are still yours, and so is the line.

To learn more about negotiating SaaS agreements with AI, register to watch this webinar hosted by LegalOn and Today’s General Counsel.

Must read intelligence for general counsel

Subscribe to the Daily Updates newsletter to be at the forefront of best practices and the latest legal news.

Daily Updates

Sign up for our free daily newsletter for the latest news and business legal developments.

Test your knowledge with our monthly legal crossword puzzle.

Scroll to Top