The demo produces an answer in seconds. Your lawyers still need to find the source, check it, and rewrite the result. A license price won't tell you what that workflow costs.
In our AM100 law firm engagement, roughly eight weeks of engineering delivered private AI that let the firm eliminate about $1.7M a year in cloud-AI spending. The work also addressed what happened after an answer appeared: attorneys needed sources they could check and output they could use.
A useful build-or-buy decision starts there. Define one legal task, test each delivery option against the same work, count review and operating effort, and choose the smallest route that meets the practice's requirements.
Write the job before choosing the product
Take one recurring task in a corporate practice. For example: find a relevant precedent for an issue in a current matter, show where it came from, and prepare a short note for the supervising lawyer.
Specify the workflow in terms the practice can judge:
- Input: The matter, issue, permitted document collection, and the lawyer's access rights.
- Output: A proposed answer with resolvable sources, exact supporting passages, and unanswered questions.
- Next action: A lawyer checks applicability and decides what goes into the matter's work.
- Failure handling: The workflow asks for clarification or declines when the available sources don't support an answer.
Choose the permitted test documents with the firm's security and knowledge teams. Include routine matters and difficult cases. Write down what a good answer needs before a vendor or builder sees the test.
Compare three delivery routes
Start with the capability already available. Your existing tool may support better retrieval, permissions, or review routing once someone configures it around the task.
Then compare a new product and a focused custom workflow against the same requirements.
| Requirement | What the test must show |
|---|---|
| Relevant answer | The source addresses the issue and the answer stays within what it supports. |
| Checkable evidence | Citations open the right document and passage; quoted text matches. |
| Correct access | A user can't retrieve a document they aren't allowed to see. |
| Useful handoff | The lawyer can review the result in the place they do the work. |
| Sustainable ownership | A named team can maintain sources, access, tests, and support. |
A product may handle the core task while a small integration handles routing. A custom build may be justified when a necessary capability remains unavailable. Keep those requirements visible so a compelling demonstration doesn't quietly change the decision.
If you’re a law firm managing partner and want to grow your firm without making partners carry more work, we’ll identify the biggest bottleneck standing in your way, build the plan to remove it, and implement it in eight weeks* or less - guaranteed or we continue to work for free and you don’t pay.
Find my bottleneck* Eight-week delivery, or twelve weeks when substantial data foundations are required, as agreed before work begins. If a covered stage fails its agreed benchmarks, we provide up to four additional weeks of remedy work at no charge, then waive or refund that stage’s fee if it still fails. Successfully delivered stages remain payable. Agreed client-caused delays pause the clock. Delivery guarantee terms.
Use one evaluation pack for every option
Give each option the same permitted source collection and task instructions. Have the firm's experts record the expected evidence and unacceptable failures. Keep a set of unseen tasks for the final comparison.
For each task, record whether the answer was useful, which claims the sources supported, how much the lawyer corrected, and the total time to an accepted result. Capture failures as examples someone can inspect.
For example, ask the system to find a relevant precedent for a transaction issue. A persuasive answer with a citation to an inaccessible document fails the access test. An accurate quotation from an unrelated matter still needs the lawyer's applicability judgment. A system that says it lacks enough evidence may be handling the task correctly.
If a custom option hasn't been built, use a bounded prototype to test the uncertain requirement. Keep the prototype's purpose explicit: establish whether that requirement can be met and estimate the remaining delivery work.
Before accepting the evaluation, check the reference answers themselves. In our biotech engagement, a rented service and an internal build had inherited conflicting expert labels. We established a shared reference and delivered production AI in under six weeks, unblocking a platform with $58M in first-year potential. A tool comparison built on inconsistent answers would've missed the problem again.
Count the cost of an accepted result
Compare the options over the same period and expected workload. Include:
- Setup, integration, migration, and evaluation.
- Licenses or infrastructure and model usage.
- Lawyer review and correction time.
- Source updates, access administration, support, and retesting.
- Training, documentation, and a workable exit or handover.
For example, suppose two options process 100 review items a month. One takes 20 minutes of checking per item; the other takes eight. That's 20 hours of reviewer capacity a month to put alongside the price difference. Decide what that capacity would enable before assigning it a financial value.
For a custom system, identify who pays when an integration changes or a model update breaks an important behavior. For a product, test export, configuration ownership, and the support route. Put those commitments in the comparison before selecting either.
Design the workflow around the chosen capability
If a custom retrieval step is needed, scope the whole path: check access, retrieve permitted sources, prepare a structured answer, verify references and quotations, and send the result to the lawyer for review.
Keep source-existence and quote-match checks in code where practical. Let AI help interpret the request and prepare the answer. The firm's lawyers own legal judgment and the decision to use it.
Name an operational owner in the practice and a technical owner for maintenance. Agree how new sources enter the collection, who approves changes to the instructions, and which reference tasks must pass before an update goes live.
In the AM100 engagement, the firm's security team connected authorization, and knowledge-management staff and named champions became the workflow builders. That allocation matters: a tool needs people who can keep it useful alongside the practice's billable work.
Choose the route with a workable operating plan
Summarize the decision in a short table: requirement, observed result, remaining gap, delivery cost, recurring cost, and owner. Record why the winning option meets the job and what would make you reconsider it.
A configuration change, product purchase, focused integration, or custom system can each be the right answer. The evaluation should show which one fits this practice's work.
Want to see how a private system delivered in roughly eight weeks of engineering removed about $1.7M in annual cloud-AI expense? The AM100 case shows the diagnosis, build, and handover behind that decision.
If you’re a law firm managing partner and want to grow your firm without making partners carry more work, we’ll identify the biggest bottleneck standing in your way, build the plan to remove it, and implement it in eight weeks* or less - guaranteed or we continue to work for free and you don’t pay.
Find my bottleneck* Eight-week delivery, or twelve weeks when substantial data foundations are required, as agreed before work begins. If a covered stage fails its agreed benchmarks, we provide up to four additional weeks of remedy work at no charge, then waive or refund that stage’s fee if it still fails. Successfully delivered stages remain payable. Agreed client-caused delays pause the clock. Delivery guarantee terms.