Most firms adopt a legal research platform the way they adopt any software: someone buys a subscription, a few people try it, and usage drifts. Six months later the tool is either an unexamined habit or shelfware, and nobody can say which because no one defined what good use looks like. An operating model fixes that by turning a vague intention into specific plays with clear owners.
This playbook lays out that operating model as a sequence. Each play has a trigger that tells you when to run it, the person accountable for it, and the output it produces. You can adopt the sequence wholesale or lift the plays that fit your practice.
The plays assume you have already cleared the conceptual ground covered in What Lawyers Get Wrong About Machine-Assisted Case Research. If you are still weighing whether the category fits at all, start there. What follows is for firms that have decided to take the category seriously and want adoption to produce a capability rather than a habit.
A note on sequence before the plays themselves: the order matters. Each play produces an output the next one depends on, so resist the temptation to jump to the pilot before you have settled the data terms, or to roll out before you have codified the workflow. Skipping a step does not save time; it just moves the cost to a worse moment.
Play One: Scope the Decision
Trigger: A partner or office manager raises buying or expanding a platform.
What happens
Before any demo, define what you need the tool to do: which practice areas, which jurisdictions, how many users, and what your current research-hour spend looks like. This becomes the scorecard you measure vendors against. Without it, every demo looks impressive, because demos are built to look impressive. The scorecard is what lets you compare tools on the dimensions that matter to your practice rather than the ones the vendor chose to highlight.
Owner
A designated evaluation lead, usually a senior associate who does substantial research, not the most senior partner who does the least. The owner needs daily familiarity with the work the tool is meant to accelerate, because that familiarity is what makes the scorecard realistic rather than aspirational.
Play Two: Test the Corpus
Trigger: A vendor reaches the shortlist.
What happens
Run queries you already know the answers to, in your real practice areas. Compare retrieval quality, citation accuracy, and how directly each platform takes you to the source document. The questions worth probing during this play appear in Real Questions Practitioners Bring to Legal Research Tools.
The known-answer approach is the heart of this play. Because you already know what good looks like for these queries, you can judge the tool honestly rather than being dazzled by fluent prose. A platform that confidently misstates a holding you know cold has just told you everything you need to know about how it will perform on the questions you cannot check.
Owner
The evaluation lead, with input from each practice group affected. Each group should run its own queries, because a tool strong in one practice area can be thin in another, and the lead cannot test areas they do not practice.
Play Three: Settle the Data Terms
Trigger: A platform passes the corpus test.
What happens
Get written answers on training, retention, and storage before any privileged data touches the system. If the vendor cannot answer clearly, that is itself an answer.
The three questions to nail down are whether your queries train future models, how long your data is retained, and where it is stored. Each has implications for confidentiality, and each is a contractual fact rather than a feature of being a lawyer. A vendor that treats these questions as routine and answers them in writing is signaling maturity; one that deflects is signaling the opposite.
Owner
Whoever handles the firm's vendor contracts, in consultation with the evaluation lead. This is the play firms most often rush, because it is unglamorous and stands between them and the exciting pilot. Rushing it is how privileged material ends up in a system whose terms nobody read.
Play Four: Pilot With Real Matters
Trigger: Contract terms are acceptable.
What happens
Run the tool on a handful of live matters across a few weeks, with the verification discipline in place from day one. Measure time saved and, more importantly, whether anything slipped past review.
The pilot is where projections meet reality. The scorecard told you what you hoped to gain; the pilot tells you what you actually gained. Pay particular attention to the verification step, because a pilot that shows large time savings while quietly skipping verification is measuring the wrong thing. The number that matters is time saved with verification intact, since that is the only number that survives into real practice.
Owner
A small pilot group, each tracking their experience against the scorecard from Play One. Keep the group small enough to compare notes directly and diverse enough to cover the practice areas that matter.
Play Five: Codify the Workflow
Trigger: The pilot shows real value.
What happens
Write down the steps: how to query, how to verify every citation, when to anonymize, and what to document. A repeatable process is what turns a personal habit into a firm capability, and Building a Repeatable Workflow for Machine-Assisted Legal Research walks through constructing it.
Owner
The evaluation lead, producing a short written standard the whole firm follows.
Play Six: Train and Roll Out
Trigger: The workflow standard exists.
What happens
Onboard each user against the written standard, not against their intuition. Emphasize the verification step, since that is where consistency breaks down across a team.
Training against intuition is how you end up with as many workflows as you have users. Training against a written standard is how you end up with one workflow that produces defensible results regardless of who is at the keyboard. The difference shows up the first time a supervising attorney has to review work and finds it was done the way the standard prescribes rather than however the associate felt like that day.
Owner
Practice group leads, each accountable for their group's adherence. Accountability that lives at the group level rather than the firm level is what keeps adoption from drifting once the initial enthusiasm fades.
Play Seven: Review the Signal
Trigger: Quarterly, after rollout.
What happens
Revisit the scorecard. Are research hours down where expected? Have any verification failures occurred? Is corpus coverage still adequate as your practice mix shifts? Adjust or re-evaluate vendors based on what you see.
This play exists because adoption is not a one-time event. Your practice mix changes, vendors improve or stagnate, and the corpus that covered you last year may have gaps this year as you take on new work. A quarterly review keeps the decision alive rather than letting it calcify into an unexamined subscription. The review is also where you catch verification drift, the slow erosion of discipline that creeps in once a tool feels familiar and trustworthy.
Owner
The evaluation lead, reporting to firm leadership. Reporting upward keeps the tool on leadership's radar as a deliberate choice rather than a line item nobody questions.
Frequently Asked Questions
Who should own the evaluation?
Someone who does heavy research, not the most senior partner. The owner needs daily familiarity with the work the tool is meant to accelerate.
How long should the pilot run?
A few weeks across several real matters is usually enough to see both the time savings and any verification gaps. Longer pilots rarely reveal new information.
What is the most overlooked play?
Settling the data terms. Firms rush past it because it is unglamorous, then discover constraints after privileged material is already in the system.
Can a solo practitioner use this playbook?
Yes, with the roles collapsed into one person. The sequence still protects you from buying on a demo and skipping verification.
How often should we re-run the corpus test?
Annually, or whenever your practice mix shifts meaningfully, since coverage that was adequate for one area may be thin for a new one.
Key Takeaways
- Define a scorecard of practice areas, jurisdictions, and research-hour spend before watching a single demo.
- Test the corpus with known-answer queries in your real areas rather than trusting marketing claims.
- Settle training, retention, and storage terms in writing before privileged data enters any platform.
- Codify a written verification workflow so value survives across a team instead of living in one person's habit.
- Review the scorecard quarterly and re-test coverage as your practice mix changes.