Your attorneys are doing intake. They should be practicing law.
Mid-size law firms lose 40% of billable capacity to client intake, lead qualification, and administrative overhead. AI fixes this, but only if you implement it the right way. Not with pilots. Not with ChatGPT. With governed automation that your team actually uses.
Yes, if you follow Florida Bar Ethics Opinion 24-1. The opinion requires four things: confidentiality (Rule 4-1.6), competence (Rule 4-1.1), supervision (Rule 4-5.3), and reasonable fees (Rule 4-1.5). Use tools that do not train on inputs, document client consent, and publish an AI use policy.
Real Automation•Governance Built In•Attorney Adoption Included•No Adoption Barriers
Your law firm's real bottleneck is not client quality. It's client intake.
An attorney bills 2.3 hours per day. The other 5.7 hours are email, WhatsApp, client screening calls, database entry, document routing, and managing administrative chaos. That's not an AI problem yet. That's a process problem that becomes an AI problem once you realize how much billable time is buried there.
The founding partner or managing partner ends up handling lead qualification because nobody else trusts themselves to do it. That means you're paying partner rate for intake work. You've been quoted two paths out of this:
Hire a paralegal or intake coordinator.
Salary plus overhead, every year. Takes 4 weeks to hire and train. Turns over every 2 years. You're back to recruiting.
Buy an AI tool. Copilot, ChatGPT, or a legal tech vendor.
Most law firms try this first. The tool doesn't know your specific practice areas. It doesn't understand your intake criteria. Your team uses it once, then goes back to WhatsApp. Three months later, you've paid for a license you stopped using.
This is where most law firms get stuck. You know there's a better way. You've tried the obvious solutions. And nothing actually changes.
I've worked with three law firms. The first two tried generic AI tools before reaching out. Both reached the same conclusion: "We spent money. Nobody used it."
Accuracy concerns.
Lawyers think in terms of liability. When an anonymous legal professional told me "If I need to worry about 10% inaccuracy, I don't want to use it," that's not paranoia. That's professional responsibility. A tool that gets 90% of intake criteria right means 10% of leads are misqualified. That costs relationships and reputation. Generic AI tools don't promise 100% accuracy. They promise "helpful suggestions."
No integration with your workflow.
Your intake happens on WhatsApp, email, Calendly, and your CRM (or spreadsheet). A new "AI solution" means another tool. Another login. Another place to check. Your team won't use it. They'll use their existing tools.
Adoption resistance.
This is the biggest one. Law firms have a 200-year track record of not adopting new technology quickly. You're selling your team on a tool that's supposed to replace what they're already doing. Even if it works, change is friction. Your attorneys will revert to what's familiar.
That's why big legal tech vendors struggle in mid-market. They build a platform and expect adoption. Platforms are hard to adopt.
Every legal engagement starts with these protections.
Your data stays in your infrastructure.
I don't copy client files to my systems. I build on your environment, your servers, your cloud, your network.
Zero-retention API calls.
The AI processes your query and deletes it immediately, no data stored, no logs kept, nothing used for training. The query runs, the answer returns, and nothing persists on the provider's side.
NDA before access.
I sign a mutual NDA before touching any client data or case files. Standard practice, non-negotiable.
Privilege protection.
AI systems I build never store attorney-client privileged communications in searchable indexes without explicit firm approval of the scope. Privilege boundaries are defined before the first document is processed.
This is how you get the efficiency of AI without the malpractice exposure. If your firm has specific compliance requirements (HIPAA overlay, state bar data rules, or internal security policies), I build to those specs.
What the Florida Bar actually says about AI
Florida Bar Ethics Opinion 24-1 was issued in January 2024. It is the clearest guidance any state bar has published on generative AI, and it is the one your firm will be measured against if a complaint ever surfaces. Read it once. Then build your policy around it.
The opinion does not prohibit generative AI. It does not name any tool. It restates four existing obligations and applies them to AI use. Confidentiality under Rule 4-1.6: you cannot feed client confidential information into a tool that retains, trains on, or discloses it. Competence under Rule 4-1.1: you must understand the benefits and risks of the technology well enough to supervise its output. Supervision of non-lawyer assistance under Rule 4-5.3: AI is treated like a non-lawyer assistant, and you are responsible for its work product. Fee reasonableness under Rule 4-1.5: you cannot bill a client for hours the AI did, and you cannot pass vendor costs through without disclosure.
What this means in practice: a written AI use policy, tools configured so they do not train on your inputs, and documented client consent when confidential information is involved. Business-tier ChatGPT, Claude for Work, and Microsoft Copilot for Microsoft 365 all offer the no-training setting. Public consumer tools do not. That is the line the opinion draws, even though it never uses the word "ChatGPT."
Read the full opinion at floridabar.org/etopinions/etopinion-24-1.
This is not legal advice. Consult your firm's ethics counsel.
How to evaluate an AI vendor for a law firm
Ten questions to ask before you sign. If a vendor cannot answer any of these in writing, walk away.
- Does the vendor train on your inputs by default? Must be no, or the opt-out must be on by default for the business tier you are buying.
- Where is data stored and processed? US, EU, elsewhere, and which subprocessors touch the data. Get the list in writing.
- Is a data processing addendum available? A signed DPA (the BAA-equivalent for non-HIPAA work) should be standard, not an upsell.
- Can you get a SOC 2 Type II or equivalent audit report? Type II covers operation over time. Type I is a point-in-time snapshot and is weaker evidence.
- Is there a zero-retention option for sensitive matters? Some providers let you flag queries so nothing is stored server-side, even for abuse review.
- Does the tool maintain an audit log you can export? Who ran what prompt, when, on which matter. This is how you prove supervision under Rule 4-5.3.
- Can you restrict access by user, matter, or client? Conflict walls inside the firm need to extend into the AI tool, not just the document system.
- Is the vendor name-check clean? Search for known data leaks, hallucination lawsuits, and sanctions against firms that cited the tool.
- Does the contract allow you to exit and take your data with you? Export format, retention period after termination, and deletion certification.
- Is the pricing per-seat or per-query? Per-seat is predictable. Per-query scales with use and can surprise you in a discovery-heavy matter.
If you want a starting point before a full build, the Shadow AI Playbook covers the policy side, and the AI Consultant ROI Framework covers how to price the economics. For authority positioning once the compliance work is done, see AI Visibility.
Grupo Lyown, a Miami-based law firm with operations in Colombia, had the intake problem on steroids.
The Diagnosis
The firm spent $1M per year on advertising. They had excellent lead volume. But their founding attorney, the only attorney, was manually qualifying every lead via WhatsApp. Most leads fell through the cracks because he couldn't respond fast enough. Qualified leads were being lost to response delay. Conversion rate was 0% from cold leads.
The founding attorney was answering client messages at 11pm while trying to run the firm. It was unsustainable.
What I Built
Victoria, a WhatsApp AI agent that handles the qualification conversation. The bot asks the right questions based on the firm's specific case criteria. It determines if a potential client is a real fit. If they are, it books the meeting directly into Calendly. If they're not, it routes them to a fallback. Every new lead automatically syncs to the CRM.
The AI agent followed explicit qualification rules built from the firm's real intake criteria. Accuracy was 100% on qualification logic. The same logic the founding attorney would have applied, just automated.
The Results
- •60 days from discovery to live system
- •Meeting booking rate: 0% → 42% on same $1M ad spend
- •Founding attorney stopped answering intake at 11pm
- •CRM went from chaotic to clean. Every lead through the same process
- •Qualification logic auditable and adjustable by the team
Why It Worked
It lived in WhatsApp.
The AI agent didn't move the buyer to a new platform. It met clients where they already were. Adoption was automatic because there was no new process to learn.
It was governed, not just "smart."
The qualification logic was explicit. A decision tree built from the firm's actual criteria, not a black-box model. The team understood exactly why the AI said yes or no. They could audit it and adjust the rules.
It shipped in 60 days, not 6 months.
The firm saw working automation within two months. That's fast enough to maintain momentum and keep leadership attention.
It produced real metrics.
Not "improved efficiency." A specific, measurable result: 0% to 42% meeting booking rate on existing spend.
Half the law firms I talk to have already bought an "AI solution" and stopped using it.
A partner at a South Florida firm told me about it directly: "Bad experience two years ago, spent money, nobody used it."
This is not a referendum on AI. This is a referendum on how bad implementations get purchased.
Here's what's different about how I work:
You don't pilot. You go production.
A pilot is a thing that sits in a corner and doesn't touch the real business. Work-Smart doesn't do pilots. I build something production-ready that solves a real intake problem in 60 days. The system ships into your live WhatsApp, your live Calendly, your live CRM. Real leads come through it. Real conversions happen. People use it because it works, not because someone mandated it.
Your team is trained, not abandoned.
Most tech implementations hand off a tool and say "let me know if you need help." Work-Smart includes structured training, how the system works, how to handle edge cases, when to escalate. This is why adoption rates improve. Your team isn't guessing.
The system is governed, not magic.
You know exactly why the AI made a decision. No black-box confusion. No "the algorithm decided" hand-waving. This builds trust. Lawyers understand rules. Governed automation speaks their language.
You get a builder, not an advisor.
The last two law firms that failed had hired a consultant who said "here's what you should do" and then left them to figure out implementation. I build the system. I'm in the code. I'm testing with real leads. I'm the one responsible if it doesn't work. That's a very different incentive structure.
This applies if:
20-250 attorneys across one or multiple locations.
Small solos have different economics. Mega-firms (BigLaw) need enterprise implementation.
Losing partner billable time to administrative overhead.
If your managing partner spends 5+ hours per week on intake or case routing, that's a real problem worth a focused build to solve.
Intake or case management still mostly email and spreadsheet.
Some firms have invested in case management systems. That helps. But they are often not connected to your CRM or communication tools.
Bought AI tools before and adoption failed.
You still believe automation could work if it was built for your specific workflow. You're not wrong.
Care about AI accuracy and governance.
Enough to invest in clarity over "magic."
If none of these apply, Work-Smart is not the right fit.
Most law firms don't know what to build until an AI Ops Audit shows them what's broken.
The audit is two to three weeks where I:
Map your entire intake process: how leads come in, where they get qualified, how they move to cases, where the bottlenecks are.
Interview your managing partner and intake staff to understand your specific qualification criteria.
Audit your current tools: CRM, case management, email, WhatsApp, Calendly, document storage.
Calculate the hidden cost: how many billable hours are lost to administrative work per week.
Recommend what to build first and what to phase later.
The cost is a fixed fee scoped to firm size. You know exactly what you're paying before we start. At the end, you get a written diagnostic and a clear recommendation.
For example: "You need Layer 4 automation first (intake + booking). That's a 60-day build." Or: "Your bigger problem is Layer 1 data. Case files are in five places. You need consolidation first. That's an 8 to 10 week engagement." The diagnostic is specific. It leads directly to the build scope.
Clear pricing. No surprises.
AI Ops Audit
Fixed-fee diagnostic
Diagnostic only. Two to three weeks. You know the build scope (and price) before committing.
AI Foundation Build
Fixed-fee build
Depends on scope. A single WhatsApp AI agent like Victoria is a smaller engagement. A full data consolidation + dashboard + automation layer is larger. You pay as deliverables ship.
Timeline
4 to 16 weeks
First working system ships in 4 to 8 weeks. A 60-day AI agent timeline is typical for a single high-impact automation.
What's included:
- •The system itself (code, AI agents, automations, dashboards)
- •Structured training for your team (included half-day session)
- •Milestone-based billing (you pay when systems ship)
- •Ownership: you own the code, the data, everything I build
You've probably thought about this before.
"AI makes mistakes. If I need to worry about 10% inaccuracy, I don't want to use it."
Valid concern. That's why I build private AI systems trained exclusively on your documents, not public models that hallucinate. When the AI agent qualifies a lead, it asks questions you defined. The system doesn't guess. It follows rules you set.
"Our attorneys are too busy to learn new tools."
They don't have to. The AI handles intake, qualification, and scheduling. Your attorneys do what they're trained for: billable work. One firm's founding attorney was answering WhatsApp leads during billable hours. After deploying an AI agent, response time dropped to 47 seconds and the attorney got those hours back.
The AI Operating System. Applied to Your Industry
Case File Consolidation
Your case files, client records, billing data, and documents consolidated into a single source of truth. A partner can search "all briefs mentioning Section 1983 across the last five years" and get results immediately, not after emailing three associates.
Partnership Dashboard
A live dashboard showing open cases, billable hours per attorney, realization rates, pipeline health, and matters approaching statute-of-limitations deadlines. Everything a managing partner needs without asking staff for a weekly update.
Firm Knowledge AI
An AI assistant trained on your specific practice. Your firm's playbooks, past briefs (redacted for client confidentiality), filing approaches. An attorney asks a question and gets an answer grounded in your firm's actual approach, not generic ChatGPT.
Governed Intake Automation
Lead qualification. Calendly booking. Case routing. Intake document generation. Document review workflows. This is where Victoria, the WhatsApp AI agent we built for Grupo Lyown, lives. Every automated task has a clear rule set. Every decision is auditable. Your team understands what the automation did and why.
AI Governance
An AI use policy. Approved tools list. Data handling rules. Shadow ChatGPT monitoring, which associates are using it without firm approval. This is how you prevent an associate from pasting client confidential information into a public AI tool.
AI Visibility
Getting found by LLMs when prospects search for your practice area. When someone asks ChatGPT "Who handles IP disputes in construction contracts?" your firm's content shows up in the response. This is a 9-month play, but critical for authority positioning.
Victoria, WhatsApp AI Agent for Grupo Lyown
Grupo Lyown, a Miami-based law firm with operations in Colombia, spent $1M per year on advertising. Excellent lead volume. But the founding attorney was manually qualifying every lead via WhatsApp, most fell through the cracks because he couldn't respond fast enough. Conversion rate was effectively 0% from cold leads. He was answering client messages at 11pm.
Victoria, a WhatsApp AI agent that handles qualification, books meetings directly into Calendly, and syncs every lead to the CRM. 60 days from discovery to live system. On the same $1M ad spend, the meeting booking rate went from 0% to 42%. The founding attorney got his nights back.
See the full case study →Frequently Asked Questions
No. Florida Bar Ethics Opinion 24-1 does not prohibit generative AI. It requires informed use. You must protect client confidentiality under Rule 4-1.6, stay competent in the technology under Rule 4-1.1, supervise AI output under Rule 4-5.3, and charge reasonable fees under Rule 4-1.5. Public ChatGPT with client confidential information is the risk. Business-tier tools with zero training on inputs plus documented client consent are compliant.
If confidential client information goes into the tool, yes. Florida Bar Opinion 24-1 ties this to Rule 4-1.6. The practical version: a short written disclosure in the engagement letter that names the categories of AI tools you use, confirms the vendor does not train on inputs, and gets the client signature. No confidential data in the prompt, no consent required.
Business-tier ChatGPT or Claude with training off, plus a one-page AI use policy, plus a client consent clause in the engagement letter. That is roughly $30 per seat per month plus a few hours of policy drafting. It satisfies the four obligations in Florida Bar Opinion 24-1 for general drafting and research. It does not cover intake automation, private case file search, or governed agents. Those are separate builds.
A firm AI use policy covers five things: approved tools list, prohibited inputs (no client confidential data in public tools), required settings (training off, zero retention where available), human review requirements, and client consent language. Two pages is enough for most mid-size firms. Review it with your ethics counsel before publishing.
For drafting with client confidential information, use business-tier tools with training disabled, ideally with zero retention. For legal research, any tool that cites primary sources you can verify is acceptable. Never cite an AI output without reading the underlying case. Rule 4-1.1 competence means you are responsible for every citation, hallucinated or not.
Yes, if it's governed. Victoria, the WhatsApp AI agent we built for Grupo Lyown, qualifies leads at 100% accuracy on the firm's specific criteria because the decision logic is explicit. It is a decision tree, not a black-box model. Where AI is riskier is in legal interpretation or writing. I don't use AI to generate legal advice. I use it to automate routine tasks (intake, routing, document management) where the rules are clear. Those can be 100% accurate.
No. The agent asks qualification questions: practice area, matter size, timeline, location. The same questions a paralegal would ask. It doesn't interpret law or give advice. It books meetings. The attorney handles legal substance. Same separation of duties you'd have with a human intake coordinator.
Your data stays in your infrastructure. The agent runs on your Airtable or CRM instance, not on a third-party SaaS platform. Client information is never sent to public APIs. If you're worried about data exposure, private AI is non-negotiable, which is why it's built in.
Depends on the system. If it has an API (Caseload, Everlaw, LawLab, most modern systems), yes. If it's on-premise legacy software or doesn't expose data via API, we'll need to consolidate first. The diagnostic will tell you if your current system can be integrated or needs to be replaced.
Grupo Lyown saw results in 60 days. Conversion improved immediately because the system went live with real leads. For a firm losing partner billable time, ROI is usually 2-3 months. For firms doing full data consolidation, it might be 4-6 months as the system matures. The diagnostic will give you a clear timeline.
Yes, as long as the system has an API or data export. Modern case management systems are compatible. Legacy systems (on-premise, no API) require more work, usually data consolidation to a modern platform first.
Yes. Layer 5 is exactly that. I'll build an AI use policy, create an approved tools list, and set up monitoring for unauthorized AI usage. This is critical for mid-size firms managing malpractice risk and professional responsibility.
AI visibility is a medium-term play. 3-6 months for measurable citation improvement. But the lead qualification AI agent delivers immediately. One firm saw 11 confirmed meetings from 57 leads within the first month. The visibility strategy and the operational AI work in parallel.
That's normal. New sites take 2-4 months to fully index on Google. But AI tools index differently. They look for structured, authoritative content. I build pages optimized for LLM extraction from day one. One firm ranked for target queries within weeks because the content was structured correctly, even though overall traffic was still building.
Most law firms get stuck because they tried AI before, and the tool didn't stick.
The difference isn't AI. It's implementation. Production systems beat pilots. Governed automation beats magic. Adoption-first beats tool-first. If your firm is losing billable time to intake and you want a clear diagnosis of what to build, start with the AI Ops Audit. Two to three weeks, fixed fee. You'll know exactly what to build and what it costs.