Choosing Legal Tech
Why it Matters
A software demo shows a tool on its best day, with clean sample data and a presenter who knows every click. What a firm needs to know shows up later: whether its own work fits the tool, what the vendor may do with client information, and how the firm’s records get back out. Florida puts those questions to the lawyers, and the comment to its competence rule expects them to understand the benefits and risks of the technology they use, generative AI included. A paralegal who can weigh two products against the firm’s needs, read what each vendor has actually promised, and price the whole first year gives the lawyers what no demo will. You will do that for one function at Temko Eden, then test your choice against the best case for the other tool.
Current Context
In April 2026, a California personal-injury firm sued the vendor of its AI phone system (opens in a new tab), alleging dropped calls, Spanish transcription that never worked, and no help moving its phone numbers to another carrier. The vendor, which rejected those allegations in a statement to The Register (opens in a new tab), had already sued in New York for invoices under a contract it says renewed automatically. The contract sends disputes to New York courts under New York law, and the California case is now stayed because the New York case came first. On September 1, 2026, a federal magistrate judge in Manhattan let discovery go forward (opens in a new tab) while the firm’s motion to dismiss, which invokes a New York statute on automatic renewals, is pending. Whatever the sales pitch promised, the renewal and forum clauses in the signed contract now decide how long, and where, the firm fights to leave.
Key Concepts
Needs Assessment
The written account of what the firm needs a tool to do, gathered from the people who will use it before anyone looks at products. Each requirement names a task and who needs it, such as reconciling the trust account each month, never a product or a feature label from a vendor’s page. Must-haves are the requirements a tool cannot fail; nice-to-haves only add to its score.
Evaluation Rubric
A fixed set of criteria, each with a weight, against which every candidate is scored on the evidence. The firm sets the weights before anyone watches a demo, so a smooth presentation cannot change what matters, and a tool that fails a must-have is out whatever its total. A score is only as good as its source: a feature seen working in a trial counts for more than a claim on a marketing page.
Vendor Due Diligence
The research a lawyer must do before trusting a provider with client information. Florida Bar Ethics Opinion 12-3 asks whether the provider is bound to keep the data confidential and to report a subpoena, and how it protects and recovers the data. It also asks what the agreement says about the forum for disputes, the limits on the provider’s liability, and the firm’s access to its data once the relationship ends.
Technology Competence
Rule 4-1.1 requires competent representation, and its comment says that staying competent includes understanding the benefits and risks of technology, including generative AI. The comment adds that competent representation may involve retaining a nonlawyer advisor of established technological competence. Rule 6-10.3(b) gives the duty a number: 3 of the 30 continuing legal education hours a Florida lawyer must complete every three years must be in approved technology courses.
Data Portability
Whether the firm can get its own records out of a tool, complete and in a form another program can read, both while it subscribes and after it leaves. Some records must outlast any vendor: Rule 5-1.2(f) requires trust records to be kept for six years after each representation ends, and Opinion 06-1 advises that files kept electronically must stay readily reproducible. An exit plan says what comes out, in what format, who pulls it, and what the vendor does with the rest.
Total Cost of Ownership
Everything a tool costs the firm in its first year, not only the price per user. It adds the plan that actually covers the must-haves, the add-ons, setup and migration fees, any payment processing, and the hours people spend moving records and learning the tool. A price page often quotes the annual rate, so the budget states the plan, the billing term, and the number of seats with every figure.
Implementation and Training
The work of putting a chosen tool into use: moving the records, setting who may see and change what, training each person on the tasks that person does, and deciding when the old way stops. A plan names a go-live date, a period when the old and new records run side by side, and who answers questions afterward. Rule 4-5.3(b) has the partners make reasonable efforts to put measures in place that keep the staff’s conduct within the lawyers’ duties, which is hard to do with a tool the partners never learned.
Generative AI Features
Features that draft, summarize, or answer questions from what a user puts in, now built into many legal products. Florida Bar Ethics Opinion 24-1 lets lawyers use them while protecting client confidences, which starts with researching the program’s policies on data retention, data sharing, and self-learning. It recommends the client’s informed consent before confidential information goes into a third-party program, and it has lawyers develop policies and practices so the use meets their ethical duties.
Resources
- Evaluation workbook (The Firm Profile, with each person’s notes and the partners’ budgets, and the Requirements, Scorecard, Security Review, Cost, and Exit Plan sheets you fill in. Its Start Here sheet says what goes where.)
- Rules Regulating The Florida Bar (opens in a new tab) (Chapter 4 holds Rule 4-1.1, whose comment on maintaining competence names technology and generative AI; Rule 4-1.5(f)(5), on keeping contingency fee contracts and closing statements; Rule 4-1.6(e); and the comment to Rule 4-5.3 on services outside the firm. Chapter 5 holds Rule 5-1.2(f), and Chapter 6 holds Rule 6-10.3(b). Read the current chapter PDFs.)
- Florida Bar Ethics Opinion 12-3 (opens in a new tab) (Cloud computing, and the research a lawyer owes a provider before trusting it with client files. Sections A to C of the Security Review come from its questions. Advisory, not binding.)
- Florida Bar Ethics Opinion 24-1 (opens in a new tab) (Generative AI: what a program does with what users enter, client consent before confidential information goes into a third-party program, and the firm’s own policies. Section D of the Security Review comes from it. Advisory, not binding.)
- Florida Bar Ethics Opinion 06-1 (opens in a new tab) (Keeping files electronically: they must stay readily reproducible and protected from change or loss, which an exit plan has to keep true after the firm leaves a vendor. It cites the retention rules by their older subdivision letters. Advisory, not binding.)
- ABA Formal Opinion 512 (opens in a new tab) (The ABA on generative AI tools, including reading a tool’s terms of use and privacy policy before relying on it. It guides Florida lawyers without binding them.)
- Data Security Breach Reports, California Attorney General (opens in a new tab) (Search by organization name for the breach notices a company has filed. It holds only notices sent to more than 500 California residents, so an empty result is not a clean record.)
- Member Benefits: Practice Resources and Software, The Florida Bar (opens in a new tab) (Discounts the Bar arranges for its members, including on practice management and e-signature software. The firm’s lawyers are members, so a discount on a product you compare belongs in the budget.)
- Clio Manage free trial (opens in a new tab) (Billing and trust accounting: seven days with no credit card. The form asks for a phone number.)
- MyCase pricing and free trial (opens in a new tab) (Billing and trust accounting: ten days with full access and no credit card. The plan list shows what each plan and each add-on includes.)
- PracticePanther pricing and free trial (opens in a new tab) (Billing and trust accounting: seven days on an email address alone, with no credit card.)
- BoldSign pricing (opens in a new tab) (Electronic signatures: a free plan with an audit trail and no credit card, and 30-day trials of the paid plans. A signup with a personal address such as Gmail gets fewer free documents than one with a school or business address.)
- Zoho Sign pricing (opens in a new tab) (Electronic signatures: a 14-day trial of the top plan with no credit card, after which the account moves to the free plan with its data intact.)
- AI Attachment for Docusign Services (opens in a new tab) (One vendor’s binding terms for its AI features, in the version dated July 8, 2026. Section 4 lets it train AI on customer data, anonymized and aggregated, unless the customer opts out, and an opt-out works only going forward.)
What to Do
The partners will pay for one new tool next year, and they want a recommendation they can check. You take one function, write down what the firm needs from it, and compare two real products against weights you set before seeing either one. Then you price the first year, plan the firm’s way out, and write the memo that recommends one. What you post is the evaluation workbook and the memo.
Part 1: Read the Firm’s Needs
Temko & Eden, P.A. A three-lawyer firm on Beach Street in Daytona Beach. It still carries the names of its late founders. Dana Whitfield, the managing partner, handles civil litigation and real estate closings. Luis Ortega, a partner, handles family law and probate, and Kim Nguyen, an associate, handles personal injury. Marcy Cole is the legal assistant, Ron Pike is the bookkeeper and office manager, and you are the firm’s paralegal.
An email from Dana Whitfield, Monday, October 5, 2026. The partners have agreed to pay for one technology change next year, and we’ll pick it at our November meeting. Two parts of this office run on workarounds. Ron keeps the trust records in a spreadsheet because our practice management system can’t, and that system’s contract ends December 31. Marcy has been sending engagement letters from a free e-signature account in her own name. Take one of the two. Compare two real products, recommend one, and tell me what the first year costs, what could happen to our clients’ information, and how we’d get our records back if we ever left. Everyone’s notes and the budgets are in the workbook. Don’t let a demo make this decision for us.
Download the evaluation workbook and read its Start Here and Firm Profile sheets. Choose billing and trust accounting or electronic signatures, and enter your choice on the Requirements sheet. Then turn the notes for that function into requirements. Some notes name a product or a wish instead of a need; write down the task underneath it. Add anything the rules in Resources require that no one at the firm mentioned. Mark each requirement Must-have or Nice-to-have, and run the Requirements Interviewer on the ones you are unsure about.
Part 2: Weigh Before You Look
Pick two real products for your function. Resources lists some with a free trial or a free plan, and you may choose others that offer a trial, a free plan, or enough public documentation to judge them. Before you open either one, set the weights on the Scorecard so they total 100, and record the date. A weight set after a demo can be bent to fit the tool you liked. Then run the Trial Script Builder to turn your must-haves into tasks you will try in each trial, each with what counts as passing.
Part 3: Read What the Vendor Promises
Work through the Security Review for both tools. Start from each vendor’s security or trust page, then read its terms of service, privacy policy, and help articles. For every answer, record the page and whether it binds the vendor; a promise on a security page is marketing until the agreement repeats it. Opinions 12-3 and 24-1 explain why each question is on the sheet. For the breach record, search the California Attorney General’s list under the vendor’s name and its parent company’s, then search the news. The Security Page Reader can check your reading of a long page, but find each quotation it gives you on the page itself before you rely on it.
Part 4: Try Both Tools
Run your trial script in each tool, with made-up names and figures only, never a real person’s information. If a trial asks for a credit card, skip it, score that tool from its documentation and the vendor’s own demo videos, and say so in your evidence. Fill in the Met columns on the Requirements sheet as you go. Try the export too: pull out whatever the trial allows and open it in Excel or a PDF reader. Then score both tools on the Scorecard, with the evidence for each score.
Part 5: Price the Year and Plan the Exit
Fill in the Cost sheet for the plan each tool needs to meet your must-haves, which may not be the cheapest plan on the page. Count seats by who needs to sign in, check whether the price assumes annual billing, and look for any add-on a must-have depends on. The Bar’s member benefits page shows discounts the firm’s lawyers can claim. Then answer the Exit Plan for both tools from their help articles, their terms, and your test export. The rules in Resources say which records must outlast the tool, and for how long.
Part 6: Write the Recommendation
Write Dana a memo of 800 to 1,200 words that recommends one tool. Open with the recommendation and the criterion that decided it. Then give the first-year budget, what the security review found, and the questions still open for the vendor. Add the exit plan, and a training plan that names who learns what and by when, lawyers included, with a go-live date. Write one paragraph of policy on the tool’s AI features: which ones staff may use, with what information, who checks the output, and when a client must agree first. Then paste the memo and your Scorecard into the Counter-Recommendation. End the memo with a short section that states the strongest point it made and answers it, or change your recommendation if you cannot.
Part 7: Post Your Deliverables
Create one new discussion thread with a headline-style subject line that names the tool you recommend. Attach the workbook with every sheet filled in, and the memo as a PDF or Word file. In the post itself, write two or three sentences on the requirement that did the most to separate the two tools.
Part 8: Share Chat Link
Include one AI chat link with a 1–2 sentence explanation of what the conversation shows and why you chose to share it.
Suggested AI Prompts
Use these prompts as a starting point, then adjust them to fit your goal. Strong prompting develops through trial, revision, and testing. It’s a foundational skill that grows into more advanced AI work such as context engineering and agent-based workflows.
Requirements Interviewer
I am a paralegal at a three-lawyer Florida law firm, helping choose software for [billing and trust accounting / electronic signatures]. I will paste notes from the people at the firm and my draft list of requirements. Act as an experienced law office manager and interview me about the list, one question at a time. Push me to rewrite any requirement that names a product or a feature label as the task the firm actually needs done, to say who needs each one, and to defend every Must-have: would the firm really reject a tool that cannot do it? Ask whether something the Rules Regulating The Florida Bar require is missing, but do not tell me what the rules say; I will look them up. After ten questions, give me the list back with your suggested changes marked, and leave the final decisions to me.
Socratic elicitation makes you defend each requirement instead of accepting a list the model wrote for you. The rule that the model may ask about the Bar’s rules but not state them keeps the legal answers in the rule text, where you can check them.
Trial Script Builder
Here is my list of Must-have and Nice-to-have requirements for a [billing and trust accounting / electronic signature] tool at a small Florida law firm. Turn each Must-have into a short task I can perform myself during a free trial, using only made-up names and figures. For each task, give the steps in the order I would try them, what I should see if the tool passes, and what would count as a failure. Put it in a table with columns for the requirement, the task, the pass condition, and a blank column for my result. If a requirement cannot be tested in a trial, say so and suggest which vendor document would answer it instead. Do not guess how any particular product works.
Breaking each requirement into a task with a pass condition is decomposition, and it turns a trial from a tour into a test. The model does not know how these products work, so try each step in the tool before you trust the script.
Security Page Reader
I will paste a question from my security review, then the full text of one page a software vendor publishes (a security page, terms of service, privacy policy, or help article), and the page’s web address. Using only the text I paste, answer the question in one or two sentences and quote the exact words that support your answer. Then tell me whether the page reads as a commitment the customer agrees to or as a description or marketing claim, and why. If the page does not answer the question, say “Not found on this page,” and do not fill the gap from anything you know about the company. Wait for me to paste.
Grounding the answer in pasted text, with a quotation, makes every claim checkable, and the instruction against filling gaps keeps the model from supplying a promise the vendor never made. Find each quotation on the page yourself before you record it, because a model can change wording while it quotes.
Counter-Recommendation
I am recommending [Tool A] over [Tool B] for [billing and trust accounting / electronic signatures] at a three-lawyer Florida law firm. I will paste my memo and my scorecard. Act as a skeptical partner who prefers [Tool B]. Using only the facts in what I paste, make the strongest case for [Tool B]: where my weights favor my choice, which scores rest on marketing claims rather than evidence, what my budget leaves out, and what my exit plan assumes. Then name the single point I most need to answer. Do not bring in facts about either product that are not in my materials.
Red-teaming asks the model to attack your conclusion, which brings out weaknesses a request for feedback tends to smooth over. Holding it to your own materials means every objection is one you can test against your evidence.