Twenty points, in the author’s words, grouped into the three phases named in his title. Numbering follows the original.
Before: Define the problem and the shortlist
1. Start with the problem, not the product.
Be able to articulate the pain point in one or two sentences before talking to vendors. “We need AI” is not a problem statement.
Don’t buy technology to solve a technology problem. Buy technology to solve a business problem.
2. Fix the process before you automate it.
Don’t spend money to automate a bad workflow. Map the current state, eliminate unnecessary steps, and then determine where technology adds value.
Don’t automate a bad process. You’ll just make the bad process faster.
3. Prioritize ruthlessly.
Buy the technology that will have the greatest impact on cost, risk, capacity, or the business, not necessarily the tool generating the most buzz.
4. Revisit build vs. buy.
Historically, legal departments often defaulted to buying specialized legal applications. Increasingly, enterprise AI, workflow platforms, and internal development can solve some problems without another point solution. Sophisticated legal organizations are reconsidering where proprietary infrastructure makes sense.
5. Look at what you already own before buying something new.
Ask whether Microsoft, ServiceNow, Salesforce, your ELM/CLM platform, an existing AI platform, or another enterprise tool can solve 70–80% of the problem.
6. Leverage strategic vendor relationships.
An existing vendor that understands your environment, security requirements and contracting process may be able to add functionality faster and more economically than bringing in another vendor.
7. Avoid creating a collection of disconnected point solutions.
8. Buy for your ecosystem, not just Legal.
A fantastic legal application can fail if business users hate interacting with it. Particularly for intake, CLM and workflow technology, understand how Finance, HR, Procurement, Technology and business users will interact with it.
9. Integration should be a selection criterion, not an implementation problem.
Ask early how the product integrates with your identity platform, Microsoft environment, data warehouse, APIs, document repositories, eBilling, CLM and other core systems.
Every new point solution creates organizational debt.
During: Evaluate, pressure-test, negotiate
10. Make vendors demonstrate your workflows.
Never let the vendor write your demo script. Give finalists the same scenarios and data, and have them demonstrate exactly how your users would accomplish the work.
11. Separate “available today” from “on the roadmap.”
Never buy software because of promised functionality.
12. For AI, understand exactly where your data goes.
Ask what models are being used, whether your information can train or improve models, what data is retained, where it is stored, what subprocessors receive it, and whether administrators can control those settings.
13. Be very cautious with token- or consumption-based pricing.
Ask the vendor to model the cost at low, expected and heavy adoption. Ironically, your economics shouldn’t deteriorate dramatically because the product is successful.
With AI, understand the economics at 10% adoption and at 90% adoption.
14. Negotiate for success, not just Year 1.
Look at the three- to five-year economics. Negotiate renewal caps, AI/usage pricing, additional licenses, implementation fees, storage, API calls, support, migration and exit costs before you sign.
15. Do diligence on the vendor, not merely the product.
Legal tech is volatile. Understand funding, ownership, profitability/cash runway where available, customer concentration, leadership stability and acquisition risk.
16. Talk to customers the vendor didn’t handpick.
References supplied by sales are useful but naturally curated. Your legal-ops network is often more valuable. Ask peers: Would you buy it again? What did implementation really cost? What did the salesperson not tell you?
17. Define success metrics.
Time saved, cycle time, adoption, accuracy, outside-spend reduction, requests deflected, user satisfaction, etc. Be willing to walk away.
After: Adoption is the real scoreboard
18. Don’t confuse licenses with adoption.
“We purchased 500 seats” tells you nothing about value. Measure active users, frequency, workflows completed, hours saved, cycle-time reduction, etc.
19. Treat change management as part of the purchase price.
A technically successful implementation with poor adoption is still a failed implementation.
Implementation isn’t the finish line. Adoption is.
20. Know how you’ll get out before you get in.
Switching costs can create far more leverage for the vendor at renewal than buyers anticipate.
About Checkbox
Checkbox is the AI Legal Front Door trusted by in-house legal teams across the globe. From structured intake and intelligent triage to automated workflows, self-service portals, and cross-tool orchestration, Checkbox provides the infrastructure layer that makes every other investment in your legal tech stack work harder.
