The hardest part of buying legal AI is rarely the technology evaluation - it is the memo. Somewhere between the demo that impressed the team and the signature that funds the subscription sits a business case that has to survive a managing partner's scepticism, a GC's risk instincts, or a CFO's spreadsheet. Most stall there. Not because the economics are weak, but because the case is argued the wrong way: vendor benchmarks instead of the firm's own numbers, transformation promises instead of one measurable workflow, and no answer to the question every approver silently asks - "how do we stop paying if it doesn't work?" This guide walks the structure that gets legal AI approved, and pairs with our ROI measurement framework, which covers what to track after the approval.
Why legal AI business cases stall
Three recurring failure modes. First, borrowed numbers: a deck built on industry statistics invites the obvious rebuttal - "that's their firm, not ours". Second, the transformation pitch: a proposal that promises to change how the whole department works asks the approver to underwrite organisational change, which is a much bigger yes than funding one workflow. Third, unpriced risk: if confidentiality, accuracy, and professional-duty questions are not answered in writing before the meeting, the meeting becomes about them - and "let's revisit next quarter" is how risk-shaped doubts express themselves.
The fix for all three is the same: make the case small, local, and falsifiable. The sections below are that structure in order.
Step 1: Baseline what the status quo costs
Every credible case starts with the cost of doing nothing, measured in your own numbers. Pick the candidate workflow and run two weeks of honest time-tracking: how many hours went to first-pass NDA review, to research memos, to building the chronology in that running dispute. Multiply by loaded hourly cost - salary plus overheads for in-house teams, opportunity cost of billable time for firms - and, where relevant, add what the same work costs when sent outside.
This is the number nobody in the room can argue with, and it reframes the entire conversation: the subscription is no longer a new cost but a fraction of an existing, currently invisible one. Industry data then plays its proper role as corroboration - our legal AI statistics page compiles the published benchmarks (adoption rates, time-savings projections, revenue impact) with each figure cited to its original study, ready to footnote rather than to carry the argument.
Step 2: Pick one wedge workflow, not a transformation
The approvable ask funds one workflow with one owner and one metric. Good wedges share three properties: the work is frequent (so evidence accumulates fast), the output is checkable (so quality is measurable, not debatable), and the pain is acknowledged (so nobody disputes the baseline). The classic candidates:
| Wedge workflow | Baseline metric | Why it works as a pilot |
|---|---|---|
| First-pass contract review | Hours per contract; outside-counsel spend on routine paper | High volume, playbook-checkable output |
| Research memos | Hours per question answered with authorities | Citations are verifiable, quality is auditable |
| Chronologies / bundle review | Hours per matter to first usable chronology | Painful, unbillable, universally disliked |
| Document intake and organisation | Hours from "files received" to "matter navigable" | Measurable before/after on real matters |
Resist the temptation to list every feature the platform has. The business case can note that the same subscription covers review, research, drafting, and timelines - that is upside for the expansion decision later. The approval you are asking for is the wedge.
Step 3: Price the tool - and the adoption - honestly
Understating cost is the fastest way to lose credibility in month two. Price three things. The subscription: with credit-based platforms like Judicio the arithmetic is transparent - plans carry a monthly credit pool, every action shows its estimated cost before it runs, and admins can set monthly spending limits with usage alerts, so the "what if usage explodes" question has an early-warning system, not a shrug (current tiers are on the pricing page). The people: a named internal owner spending real hours in the pilot weeks - template setup, playbook configuration, and the short feedback loop that separates pilots that stick from pilots that fade. The change: an hour of training per user and the workflow documentation that makes the new path the easy path.
A business case that says "the tool costs X and adoption costs another 0.2X of our time" reads as written by someone who has run a rollout before - which is exactly the impression that gets budgets approved.
Step 4: Project savings the CFO can believe
Project three scenarios and lead with the middle one. The conservative case assumes the tool saves a fraction of the baselined hours on the wedge workflow only - if two weeks of tracking showed 60 hours a month on first-pass review, model 30-40% of that, priced at loaded cost, against the full subscription and adoption spend. The expected case adds moderate spillover (the same seats using research or drafting occasionally). The upside case - clearly labelled as such - is where the published numbers live: Thomson Reuters' Future of Professionals research projecting around five hours saved per professional per week, worth roughly $19,000 a year per professional; Clio's finding that positive revenue impact from AI nearly doubles - 36% of firms overall to 69% - among firms adopting it widely. Put those in the upside column with citations, not in the promise.
Then state the payback arithmetic plainly: at what utilisation does the subscription break even against loaded hours saved? When the conservative case clears payback on its own, say so - that sentence, more than any statistic, is what a sceptical approver is scanning for. The full metric set (time-to-completion, cost per matter, outside-spend displacement, quality spot-checks) is in our measurement framework; the business case only needs the three scenarios and the break-even line.
Step 5: Answer the risk questions in writing
Every legal AI approval meeting has the same four questions waiting. Write the answers into the case so the meeting is about economics, not fears.
- Confidentiality: where documents live, who can access them, and whether client data trains models. Point to the vendor's security page and encryption, access-control, and no-training commitments - and to your own engagement-letter and client-consent position.
- Accuracy: what the tool does to make verification easy - citation-grounded outputs, source links, review workflows - and your internal rule that AI output is checked like a junior's work. Cite the vendor's published methodology rather than accepting accuracy claims on faith.
- Professional duty: a one-paragraph note on competence and supervision duties in your jurisdiction, and the firm's AI-use policy status. (Our policy template covers this groundwork.)
- Vendor risk: data portability on exit, deletion commitments, and admin controls - the boring clauses that make the procurement team an ally instead of a blocker.
Step 6: Give the approver kill criteria
The most persuasive paragraph in a business case is the one that says how it ends. Define the pilot - real matters, 4 to 8 weeks, the named owner - and the decision rule at the end of it: "If time-per-task on the wedge workflow has not fallen by at least X% against baseline, or fewer than Y of the pilot users want to keep the tool, we do not renew." Pre-committed kill criteria transform the psychology of approval: the decision stops being "do we believe in AI?" and becomes "are we willing to spend a bounded amount to find out?" - a much easier yes.
Kill criteria also protect you, the sponsor. A pilot that fails against honest criteria and gets killed cleanly builds your credibility for the next proposal; a pilot that limps on ambiguously poisons the well for every tool that follows.
The one-page template
The whole case fits on a page, in this order:
- Problem: the wedge workflow and its baselined cost (your own two-week numbers).
- Proposal: the tool, the seats, the pilot scope, the named owner.
- Cost: subscription + adoption time, honestly stated.
- Return: three scenarios with the break-even line; published benchmarks cited in the upside column only.
- Risk: the four answers - confidentiality, accuracy, professional duty, vendor exit.
- Decision rule: pilot length, metrics, and pre-committed kill criteria.
Everything else - feature lists, screenshots, vendor comparisons (our platform evaluation guide covers that stage) - is appendix material.
Getting started with Judicio
If Judicio is the platform you are building the case around, the pieces map directly: transparent per-action credit pricing with pre-run estimates and admin spending limits with alerts for the cost section; the security and methodology pages for the risk section; and a free 7-day trial - 500 credits, no card required - that lets you run the baseline-versus-tool comparison on real work before any budget conversation happens at all. The strongest business case, after all, is the one whose pilot has quietly already succeeded.