If you run a law, accounting, or government affairs firm near the Capitol, your staff already spends hours a week on work that AI can do. Intake forms. Scheduling. Reminder calls. Rebuilding last Tuesday's time entries from memory. First drafts of the same letter you sent last month.
You also hold things your clients trust you to keep quiet. Social Security numbers. Tax returns. Privileged emails. A client's position on a bill before it is public. One careless paste into a free chatbot puts that at risk, and so does your license or your reputation.
So the real question for a managing partner is not "should we use AI?" It is three narrower questions. Which tasks are safe to automate now? Which data must never go into a public tool, and what do you use instead? And what does it cost and save for a firm of 5 to 20 people?
This post is not legal advice. Talk to your own counsel or your bar's ethics resources about your obligations.
What the time savings look like for a 5 to 20 person firm
Start with the money, because that is what decides whether any of this is worth doing.
Here is an example for a 12-person firm. Your numbers will be different. The point is the arithmetic.
| Task | Hours per week now | Whose time |
|---|---|---|
| New client intake and conflict-check prep | 4 | Staff |
| Scheduling, reminders, chasing documents | 5 | Staff |
| Reconstructing time entries at the end of the week | 7.5 | Professionals (6 people, 15 minutes a day) |
| First drafts of routine letters and engagement letters | 4 | Professionals |
Say automation takes half of that off the table. That is 4.5 staff hours and 5.75 professional hours a week.
- Staff time at $35 an hour, once you count payroll taxes and benefits: 4.5 x $35 x 50 weeks = $7,875 a year.
- Professional time at a loaded cost of $90 an hour (not the billing rate): 5.75 x $90 x 50 weeks = $25,875 a year.
That is about $33,750 a year in recovered time, using conservative rates and only half the hours. It does not count the bigger number for most firms: time entries that never get recorded at all. Fifteen minutes a day of unbilled work, at a $250 billing rate, is about $15,600 a year per professional.
On the cost side, plan for two things. A business-grade AI tool with a written no-training commitment, usually billed per seat per month. And a one-time build to connect it to your practice management, document, and calendar systems. If you assume $30 per seat per month, 12 seats is $4,320 a year. Check current vendor pricing before you budget.
The test is simple. A fix that costs less than one year of the problem, and keeps working after that, deserves a serious look.
Low-risk tasks law and accounting firms can automate now
These are the places to start. Each one saves real hours, and each one keeps a person in charge of anything that matters.
Intake. A web form or phone agent collects the basics: who the client is, what they need, who the other parties are. It drafts the intake record and the conflict-check list. A person runs the conflict check and decides whether to take the matter.
Scheduling and reminders. Booking consults, confirming appointments, and reminding clients about deadlines and missing documents. For an accounting firm in tax season, chasing the last W-2 or 1099 is a large share of staff time. A system that follows up on a schedule, and stops when the document arrives, takes that off the desk.
Time entry. The system drafts time entries from what already happened that day: calendar events, sent emails, documents edited, calls logged. The professional reviews and approves them in a few minutes instead of rebuilding the week on Friday afternoon. Nothing is billed until a person approves it.
First drafts with human review. Engagement letters, routine correspondence, meeting summaries, and status updates to clients. The system produces a draft from your own templates and the facts in the file. A professional edits and signs off. The draft is a starting point, never the finished work.
Internal search. Finding the memo you wrote two years ago, or the last time the firm handled a similar issue, across your own documents. This stays inside your systems.
What these have in common: the input is routine, the output is checked by a person, and a mistake gets caught before it reaches a client, a court, or an agency.
What the Pennsylvania ethics opinion says
Pennsylvania lawyers have specific guidance. The Pennsylvania Bar Association's Committee on Legal Ethics and Professional Responsibility and the Philadelphia Bar Association's Professional Guidance Committee issued Joint Formal Opinion 2024-200, "Ethical Issues Regarding the Use of Artificial Intelligence" in 2024.
The short version, from the opinion itself:
- A lawyer who uses AI has to understand the technology, its benefits, and its risks. That duty comes from the competence rule.
- A lawyer "must not input any confidential information of a client into AI that lacks adequate confidentiality and security protections."
- AI systems that handle confidential data must keep it confidential and must not share it with other clients or anyone outside the privilege.
- Every citation and every AI-generated statement has to be checked by the lawyer.
- Lawyers must explain their use of AI to clients and, where necessary, get the client's consent.
- AI-related expenses must be reasonable and appropriately disclosed to clients.
The opinion's conclusion is that, with appropriate safeguards, lawyers can use AI under the Pennsylvania Rules of Professional Conduct. It is advisory and does not bind the Disciplinary Board. But it tells you what a careful firm is expected to do.
Accounting firms have their own version. Under the federal Safeguards Rule, tax and accounting professionals count as financial institutions regardless of size and must keep a written information security plan. The IRS explains what that plan needs in Publication 5708. Any AI tool that touches client data belongs inside that plan.
Government affairs firms have no single rule like this. They do have clients whose strategy and positions are worth a great deal to the other side. Treat that information the way a law firm treats privileged material.
Data that never goes into a public chatbot
A public chatbot is the free or personal version of any AI tool, used through a personal account, with no contract with your firm. You do not control where the data goes, how long it is kept, or who can see it.
Keep these out of it, always:
- Social Security numbers, bank account numbers, and driver's license or state ID numbers
- Tax returns, financial statements, and payroll records
- Privileged communications and attorney work product
- Client names together with the facts of their matter
- Health information
- A client's unannounced plans, negotiating positions, or legislative strategy
The first item on that list has a price tag in Pennsylvania. Act 33 of 2024 (Senate Bill 824) took effect on September 26, 2024. A breach that exposes a person's name together with a Social Security number, bank account number, or driver's license or state ID number can leave the business paying for 12 months of credit monitoring for each affected person, plus a credit report for anyone who cannot get one free. If notice goes to more than 500 Pennsylvania residents, the Attorney General must be notified at the same time. That is before the cost of lost clients.
What a private AI setup looks like instead
You do not have to choose between the time savings and the risk. A private setup gives your staff the same kind of tool with controls you can show a client or an auditor.
- A business contract, not a personal account. The vendor agrees in writing not to train on your data and states how long it keeps it. Your firm owns the accounts, not individual employees.
- Access that matches your ethical walls. The system only sees the matters a user is allowed to see. A conflict wall in your practice management system stays a wall in the AI tool.
- Redaction before anything leaves the firm. Account numbers and identifiers are stripped or masked automatically before a request is sent.
- Logs. Every request is recorded: who asked, what data was used, what came back. If a client or regulator asks, you can answer.
- A person approves the output. Drafts are drafts. Nothing is filed, sent, or billed without review.
- Hosting in your own cloud account where it matters. For the most sensitive work, the model can run in a cloud environment your firm controls, so client data never goes to a shared service.
None of this requires replacing your practice management system, document management, or accounting software. A good setup connects to what you already use.
How to start without a leak or an ethics problem
- Write a one-page AI policy. Which tools are approved, which data is off limits, and who reviews output. Your staff are probably already using public tools. A policy turns that into something you control.
- Pick one process. The one that eats the most staff hours, with the least sensitive data. Intake or scheduling is usually the right first choice.
- Measure before and after. Hours per week, before and after. If the number does not move, the project did not work.
- Tell clients. Add a plain sentence to your engagement letter about how the firm uses AI and how it protects their information.
Help for firms in Harrisburg and central PA
StencilWash builds private AI and automation for professional firms in Harrisburg, Hershey, and the West Shore. Our office is in Elizabethtown, about 25 minutes from downtown Harrisburg. Security is where we started: we design the access controls, logging, and data handling first, then the automation.
If you want to know what this would save your firm before you spend anything on tools, start with the Technology ROI Assessment. A fixed-price review: $2,500, two weeks. We review what you spend on software, talk to the people who do the work, and map up to five of your core processes. You get a written report with the hours and dollars each one costs, a ranked list of what to automate first, and a 90-day plan, including which tasks should not touch AI at all. The report is yours whether or not you hire us. If you start build work with us within 60 days, the fee is credited against it. See what the assessment covers.
Free checklist
Where Is Your Tech Budget Leaking?
A 10-Point Checklist for Business Owners
Ten checks you can run in an afternoon: unused seats, duplicate tools, manual copying, idle cloud spend and more. Each one tells you what to record so you can put a dollar figure on it.

Decades of experience building and securing real-world systems.
