Most legaltech purchases fail for the same reason: the firm buys what the demo showed, not what the practice does on a Tuesday afternoon. The fix is procedural, not technological. A firm that writes down its current workflow, prices three years of ownership instead of one year of subscription, and tests migration before signing will avoid most of the damage that procurement teams later describe as surprise.
This guide walks through the three stages where buying decisions actually get made. It covers what to watch for in a vendor demonstration, how the common pricing models behave over time, and which migration problems show up consistently. Where a claim belongs to a vendor, this guide treats it as a vendor claim. Nothing here is a substitute for reading the contract itself. This connects to our earlier piece, EU AI Act Obligations for General-Purpose Models Took Effect August 2: What Legaltech Vendors Selling Into Europe Must Track.
The stakes are ordinary but real. A tool that does not fit gets quietly abandoned, and the firm keeps paying. A migration that goes badly can interrupt billing, document management, and client reporting at once. The sections below are ordered the way a purchase unfolds.
What should a firm do before contacting any vendor?
Map the current process first. Pick one recurring task — intake, contract review, e-discovery collection, whatever the firm actually does weekly — and write down every step, every handoff, and every place work stalls. The map matters more than the feature list, because vendors sell features and firms live in processes.
Two documents should come out of this stage. The first is a short requirements list, split into what the tool must do, what it should do, and what would be nice. The second is a set of test scenarios: three or four real matters, anonymized, that any candidate tool has to handle during evaluation. A vendor who cannot run the firm's own scenario on the firm's own sample data is telling the firm something.
It also helps to know the landscape before the first call. Coverage of what firms actually run, category by category, is available in The Legal Tech Stack: What Modern Firms Actually Run, Category by Category, and readers tracking the wider market can follow the publication's Legaltech News section for product releases and procurement developments. Readers following this should also see Legaltech News Law Com — The Legal Tech Stack: What Modern Firms Actually Run, Category by Category.
Which demo red flags should buyers watch for?
Demos are staged performances. The vendor has rehearsed the path, seeded the data, and chosen the example matter that flatters the product. None of that is dishonest; it is the format. The buyer's job is to break the staging.
Several signals deserve attention:
- The canned path. If every click lands on a pre-loaded example, ask the vendor to start from a blank environment and import the firm's sample data live. A tool that shines only on its own seed data will not shine on the firm's.
- The feature that exists in the roadmap. Distinguish what the product does today from what it will do next quarter. Ask for the release the feature shipped in, and treat any answer without a version number as a roadmap item.
- The integration claim. "It integrates with your document management system" can mean a native two-way sync, an export button, or nothing at all. Ask which fields move, in which direction, and how often.
- The compliance word. Vendors use regulatory language loosely. When a tool is described as compliant, ask: compliant with what instrument, in which jurisdiction, verified by whom? Precision matters here — the word "legal" itself, as Dictionary.com defines it, means "permitted by law; lawful," and a vendor saying a product is "legal-compliant" without naming a rule has said very little.
- The missing person. If the demo is run entirely by sales, request a session with the person who will implement the tool. Implementation staff answer questions sales staff deflect.
One more habit is worth adopting: record every answer and attach it to the contract negotiation as a side letter or statement of work. A claim made in a demo room has no weight later. A claim written into the agreement does.
How do legaltech pricing models differ in practice?
Three pricing models dominate, and each hides a different cost.
Per-seat subscription. The most common model, and the easiest to budget — until usage grows. The question to ask is who counts as a seat. Paralegals? Occasional users? A billing partner who logs in twice a year? Seat definitions that look generous in the sales deck become expensive in year two.
Usage-based pricing. Common in e-discovery, contract review, and AI-assisted tools, where cost tracks documents processed, gigabytes stored, or queries run. This model can be cheap for a quiet practice area and punishing during a large matter. Buyers should ask for historical usage patterns from comparable customers, or at minimum a cap or tier structure, before signing.
Platform or enterprise licensing. A flat fee for the firm, often bundled across modules. It removes per-seat anxiety but invites a different problem: paying for modules nobody uses. Firms should ask which modules are included, which are extra, and what happens to the price when the bundle is renewed.
Under every model, three costs sit outside the headline number: implementation and data migration fees, training, and annual escalators. A useful discipline is to price three years of total ownership — subscription, services, escalators, and the internal staff time the rollout will consume — and compare vendors on that figure. The cheapest annual subscription is frequently not the cheapest tool.
Where do migrations actually go wrong?
Migration is the stage buyers underestimate most, and the failure modes are consistent enough to name.
- Data does not move cleanly. Legacy systems hold matters, contacts, and documents in structures the new tool does not mirror. Field mapping — deciding which old field becomes which new one — takes longer than anyone plans for, and unmapped data is usually data the firm discovers it still needs.
- History gets left behind. Vendors often propose migrating only active matters. That is reasonable until a closed matter reopens, an old engagement letter is needed, or a discovery request reaches back years. The cost of archiving old data should be priced, not assumed away.
- Integrations break quietly. A new platform can sever the connection between billing and document management without anyone noticing for weeks. Test the integrations that touch money and deadlines first.
- Adoption stalls. If training is a single webinar, most users revert to old habits. Rollouts that pair each practice group with a named superuser tend to hold; rollouts that rely on a manual tend not to.
The practical safeguard is a pilot with real work. Run one practice group, or one matter type, on the new tool for a defined period before cutting over the firm. Ask the vendor to commit migration support in writing, including who fixes what when a field maps wrong at scale.
What belongs in the contract before signature?
The contract is where demo promises either survive or evaporate. Several clauses deserve particular attention in legaltech deals.
- Data ownership and exit. The agreement should state that the firm's data belongs to the firm, in a documented format, exportable on request and at the end of the relationship. Exit terms negotiated at signing are cheaper than exit terms negotiated during a breakup.
- Security and confidentiality commitments. Ask where data is stored, who can access it, and what happens on a breach. For tools touching client communications, the governance questions run deeper — see the publication's coverage of AI meeting note-takers and privilege for an example of what happens when this is left unexamined.
- Service levels. Uptime commitments mean little without a remedy. Ask what compensation applies when the platform is down during a filing deadline.
- AI-specific terms. For tools that use generative models, ask whether the firm's data trains anything, which model provider is involved, and what the vendor warrants about outputs. Firms selling or deploying into Europe face additional documentation duties under the EU AI Act, as covered in the publication's analysis of general-purpose model obligations, and state-level AI statutes are adding documentation duties of their own, as described in the January state AI law coverage.
- Supervision and responsibility. Where a tool acts inside a legal workflow — drafting, reviewing, filing — the professional-responsibility questions are live, and the publication's piece on Rule 5.3 supervision when the tool acts sets out the framework courts and bar authorities are working from.
Buyers who want a broader compliance frame for vendor relationships will find the general regulatory context in the publication's Regulation and Compliance sections.
What this means in practice
The evidence of how purchases succeed is procedural, and it points to a short checklist. Write the workflow down before the first vendor call, so the demo is judged against the firm's process rather than the vendor's script. Convert demo claims into contract language, because spoken promises do not survive implementation. Price three years, not one, and include services, escalators, and staff time. Pilot with one real practice group before any firm-wide cutover, and negotiate exit and data-export terms at signing, when the vendor still wants the deal.
What remains unknown is product-specific. No guide can tell a firm whether a given vendor's implementation team is competent or its roadmap is real; only the firm's own scripted test on its own sample data can. The tools change every year. The discipline of buying them does not.
