Most legal AI pitches a law firm sits through this year will lead with a number. For example: A platform scored 92% on M&A contract extraction; or, It beat human lawyers on the bar exam; or, It outperformed a competing model on some legal evaluation benchmark. These numbers are usually true, and they’re usually the wrong thing to be evaluating.
What actually predicts whether a tool gets used inside a law firm has very little to do with how it scores on a benchmark, and almost everything to do with how much friction stands between a lawyer opening their laptop and getting value out of it.
What the market has already shown
Spellbook is the clearest evidence of this. It scaled across thousands of mid-market firms and corporate legal teams while competing against platforms with far larger valuations and, by most benchmark measures, more sophisticated underlying models (Harvey at an $11 billion valuation, Legora at $5.55 billion). Spellbook’s founder, Scott Stevenson, describes the design principle behind it as an “electric bicycle” for lawyers: the technology doesn’t ask you to change where you work or how you think, it just pushes you a little further along the path you were already on. In practice, that meant building inside Microsoft Word and Outlook rather than building a new destination for lawyers to log into.
The difference between those two approaches, laid out step by step, is stark:
| Standalone web platform | Tool embedded in existing workflow |
| Open a browser, navigate to the portal | Open Word or Outlook as usual |
| Authenticate, multi-factor login | Click the sidebar add-in already installed |
| Export the draft from Word, upload as PDF or docx | Run the review directly on the document you’re already in |
| Wait for processing, view a separate dashboard | Accept or reject inline, tracked-changes edits with one click |
| Copy findings, paste them back into Word | Done |
| Result: high context-switching cost, abandoned by busy fee-earners | Result: near-zero friction, high daily active use |
A lawyer will not open a new browser tab, log into an unfamiliar portal, and re-upload a document for a routine drafting task, no matter how good the tool behind that portal is. The same lawyer will click a sidebar button inside Word they already have open twenty times a day, almost without noticing they’re doing it. Adoption follows the second pattern, not the first, essentially every time. This operational knowledge helps before your law firm commits budget and training time to a platform built the other way.
Friction is only half the test
The biggest threat to actual adoption in 2026 isn’t a clunky interface but what’s increasingly called “AI slop”: plausible-sounding legal text generated without any grounding in verified market data, that a lawyer has to fact-check as carefully as if they’d written it from nothing themselves. A tool that’s frictionless to open but generates largle unreliable output just gets abandoned for a different reason.
The tools worth taking seriously combine both properties. Rather than generating generic contract language, they surface verified, firm-approved clause libraries or live market benchmarks directly where the lawyer is already looking. As an illustration, the result here would be something closer to “this indemnification cap sits at 15%, where standard market terms on precedent deals over $50 million sit at 10%,” rather than a bare suggestion with no source attached. And every suggestion should trace back to something a lawyer can actually check, for example, a statute, an underlying deal document, a precedent already in the firm’s own document management system visible in the same screen, not buried in a separate citation panel three clicks away.
What this should actually change about how you evaluate a pilot
Treat benchmark scores and model-parameter claims as close to irrelevant to your own decision. They answer a question about the underlying model’s competence. That isn’t the question that predicts whether any particular law firm’s associates will still be using the tool in six months.
Ask, instead, how long it takes a fee-earner to get real value from the first time they open it. A tool that requires a multi-week onboarding programme before anyone sees a benefit is telling you something about how much friction is baked into its design, regardless of what its marketing claims about underlying capability.
“Where a tool lives” isn’t just a convenience question but a security question that a firm’s IT department will care about as much as its lawyers do. A new standalone portal is a new place where the firm’s data has to be evaluated, secured, and monitored, which is exactly why enterprise IT teams often take months reviewing genuinely new platforms. A tool that operates as an add-in inside infrastructure the firm has already secured, for example its existing Microsoft 365 or Azure environment is a fundamentally smaller review. The difference in review timeline is a meaningful signal about how disruptive the tool actually is to the workflow already built.
The instinct to be impressed by a high benchmark score in a demo is natural. The more useful instinct is to ask where the tool actually lives, and whether your own people will still be opening it a month after the novelty wears off.
Lawfinity Solutions advises international law firms on cross-border legal market positioning. If the India corridor is a live question for your firm, we would be interested in a conversation. Lawfinity works with one firm per jurisdiction. Engagements begin with a single conversation about your firm’s current position and where the corridor question is live for you. Write to Prachi Shrivastava
