What It Takes to Get Real AI ROI: Lessons From the HIKE2 Lodge at Dreamforce

Every fall, HIKE2 opens the doors of the HIKE2 Lodge at Dreamforce — now in its third year — to bring together general counsel, legal operations leaders, law firm pricing directors, CRM owners, and technologists for candid, off-the-keynote-stage conversations. Dreamforce itself spans nearly every industry and function; the HIKE2 Lodge is our chance to go deep on one slice of it: how organizations operating under real governance, risk, and compliance pressure are actually putting AI to work.

Across seven presentations this year, a clear and consistent story emerged — not about which tool is winning, but about what separates organizations that are turning AI into real operational change from those still stuck in the pilot phase.

Here’s what we discussed.

1. Data governance is the new competitive edge

The adoption numbers coming out of legal alone are remarkable — legal departments have gone from 24% AI adoption to 98% in just two years, with a majority now experimenting with agentic AI specifically. But governance hasn’t kept pace, and that gap is quickly becoming the most consequential risk in the room.

Catherine Krow, HIKE2’s Principal in Legal Advisory Services, opened her session on corporate legal AI by grounding the excitement in a hard truth: “So many things can be done, but it really starts with clean, connected data.” Quoting Whitney Staffko of Ford, she pushed the point further: “You have to slow down before you can speed up.” Her own closing takeaway from that panel cut just as sharply: the right question isn’t what an organization can do with AI, it’s what it should do.

• Validate data at intake and prioritize cleanup around your highest-value use cases, rather than trying to govern everything at once.
• Treat governance as a discipline with owners, not a one-time project. Several organizations described creating dedicated roles just to monitor and validate agent output.
• Ask “should we?” before “can we?” — the gating question that determines whether an AI investment pays off or just adds noise.

The organizations getting real value aren’t the ones moving fastest, they’re the ones who slowed down long enough to build a foundation worth building on.

2. The winning AI is narrow, not all-knowing

A pattern showed up again and again across the contracting, legal ops, and CRM conversations: broad “do-everything” AI assistants demo beautifully and struggle to get adopted, while narrow, task-specific agents embedded directly inside an existing workflow get used constantly because they don’t ask anyone to change how they already work. Panelists pointed to contract intake, obligation tracking, and negotiation-support agents as examples of tools succeeding specifically because they’re deeply embedded rather than broadly capable, with some vendors now shifting to consumption-based pricing to match.

• The more ambiguous a tool’s job, the harder it is to drive adoption. Narrow scope beats broad promise.
• Embed AI in the workflow where the task already happens, rather than asking users to visit a separate tool.
• Expect this to reshape vendor pricing, too, as more tools move from per-seat to consumption- or outcome-based models.

Adoption isn’t a training problem. It’s a design problem, and the fix is usually to make the AI’s job smaller, not bigger.

3. Pricing is shifting from hours to outcomes faster than expected

This theme ran through nearly every legal-focused session. Law firm leverage models built on a four-associate-to-one-partner ratio are compressing as AI absorbs more of the work junior associates used to do. Meanwhile, corporate legal departments are getting sharper about holding firms accountable: bill rejection and write-down rates have climbed roughly 64% year over year, driven in part by clients running AI-powered bill review before an invoice ever reaches an attorney.

Deborah Haile, who leads outside counsel, vendor, and contract operations at Uber, described the shift bluntly: her team is introducing task codes specifically to track where AI is being used on a matter and closing the gap when a firm’s bill doesn’t reflect the efficiency it should. On the pricing side, Fenwick & West’s Kevin Vaarsi and Sheppard’s Paco Rivilla both pointed to the same emerging reality: clients want to see the savings AI generates, not just hear that a firm is “using it.”

• Expect more alternative fee arrangements, portfolio-based fixed fees, and consumption-based pricing tied to outcomes rather than hours.
• If you manage outside counsel spend, focus your AI-tracking effort on your highest-spend relationships first. That’s where the leverage is.
• If you’re a services provider, the firms proactively bringing pricing innovation to clients are building trust faster than the ones waiting to be asked.

Nobody has fully solved AI-era pricing yet, but the organizations treating it as an open conversation with clients are pulling ahead of the ones waiting for the market to settle it for them.

4. Adoption lives or dies on culture, not code

Across CRM, law firm operations, and data leadership discussions, the same insight kept resurfacing from different angles: the technology was rarely the hard part. Getting people to actually use it was.

Amy Gradnik, HIKE2’s Chief Strategy Officer, framed it plainly while moderating a panel of data leaders: “Technology gives us access to data, but it can’t make anyone use it.” Laura Long, COO and CFO of Hanson Bridget, shared a similar lesson from her firm’s AI rollout: mandatory training backfired, while peer-led sessions where colleagues shared exactly what problem they’d solved and how drove real engagement. And in the “You’ve Got the CRM. Now What?” session, panelists described deliberately showing colleagues an unfinished, imperfect system early, rather than waiting to unveil something polished, because early involvement built more trust than a flawless launch ever could.

• Peer-led, use-case-specific training consistently outperforms mandatory, generic training.
• Bring end users in early and let the system be visibly imperfect. Ownership drives adoption more than polish does.
• Measure success by whether someone’s day got easier, not by login counts or feature usage.

People don’t resist AI. They resist being handed a finished tool they had no hand in shaping.

5. The delivery model itself is being rewritten

The final shift is less about AI directly and more about how organizations get help building with it. Amanda Wodzenski, HIKE2 Founding Partner, and Branden Bellanca, HIKE2 Principal overseeing engineering and delivery, walked through the rise of forward-deployed engineering — small, embedded teams that sit inside a client’s business and ship working solutions in days or weeks instead of months.

“We’ve got to think far less about what tasks we’re executing and more about what problems we’re solving,” Wodzenski told the room. Bellanca was careful to note that speed can’t come at the cost of judgment: “Just because I can build something quickly doesn’t mean I can’t do it safely.” Trust, he added, is still foundational. It doesn’t go away just because the tools got faster.

• Small, embedded teams working side-by-side with a client build trust faster than traditional big-proposal engagements.
• Speed still requires domain expertise. Someone on the team has to understand the regulatory and security constraints the client actually operates under.
• The goal of a forward-deployed engagement isn’t a proof of concept that gets shelved, it’s a production-ready foundation the client’s own team can build on.

Fast and thoughtful aren’t opposites. The teams getting this right are proving you can have both.

The common thread

Strip away the specific topics (contracting, CRM, outside counsel spend, pricing, delivery models), and the same ideas surfaced in nearly every room at the HIKE2 Lodge this year: govern the data before you trust the output, keep the AI’s job narrow enough that people actually adopt it, price and deliver based on outcomes rather than hours, and never underestimate how much of “AI transformation” is really change management wearing a technology costume.