The main pros are faster drafting, summarising and replies, more capacity without hiring, and consistent first drafts. The main cons are errors that need checking, data-privacy exposure, subscription creep, learning time and uneven quality. For a small team the tools cost about $20 to $30 per person a month; the larger cost is usually staff time spent checking.
A list of pros and cons helps nobody decide unless each item has a price. So every pro and con here comes with what it costs or saves, in dollars or hours, using an illustrative eight-person law firm. The short version: the subscriptions are rarely the deciding cost. Review time, learning time and what you do with the freed hours decide whether AI pays.
The ledger at a glance
| Item | Pro or con | How it shows up in an eight-person law firm | Size (illustrative) |
|---|---|---|---|
| Faster first drafts | Pro | Routine client letters, attendance notes, engagement letters | 10 to 20 minutes saved per letter |
| Summaries of long documents | Pro | Key terms from a 40-page contract, chronology from a bundle | 30 to 60 minutes saved per document |
| Capacity without hiring | Pro | More matters per fee earner | Depends on how freed time is used |
| Faster first response | Pro | New enquiries acknowledged with next steps the same hour | Hard to price; shows up in conversion |
| Review time | Con | Every AI draft read and corrected | Around 5 minutes per letter |
| Errors that escape | Con | An invented citation or wrong date reaches a client or court | One incident can cost more than years of seats |
| Data exposure | Con | Client material in a consumer tool | Avoided by paying for business plans |
| Subscription creep | Con | Individual tools bought on personal cards and expensed | $20 to $100 a month nobody tracks |
| Learning time | Con | Everyone learning prompts, rules and checks | About 5 hours per person in month one |
| Skill fade | Con | Juniors who never draft from a blank page | Shows up years later |
The upside, with prices attached
Faster first drafts. Many small law firms send dozens of routine letters a week: updates on progress, requests for documents, covering letters. Say the firm sends 30 a week across all fee earners and AI saves 15 minutes on each before review. That's seven and a half hours a week, close to a full working day, from one use case.
Here's what one of those drafts looks like in practice. A fee earner types rough notes straight after a call and asks for an attendance note in the firm's format:
Notes: call w client 20 mins re lease renewal. wants 10yr, not 5.
worried about rent review clause - how often, any cap? told her
we'd check. landlord's agent wants answer by fri. she'll send
current lease pdf today.
Prompt: Turn these notes into an attendance note under our headings:
Date/duration, Attendees, Summary, Advice given, Actions (who, by
when). Don't add advice that isn't in the notes.An illustrative result sorts everything under the right headings in seconds, but under "Advice given" it writes: "Advised client that the rent review clause is standard and acceptable." The notes record the opposite: the fee earner promised to check. Left in, that line is a file record of advice nobody gave. The five minutes of review below exist to catch exactly this.
Summaries of long documents. Pulling the key dates, obligations and termination terms from a 40-page contract, or building a chronology from a bundle of correspondence, takes a fee earner an hour or more. A first pass from AI takes minutes, and the lawyer's time goes into reading the clauses that matter rather than finding them. The saving is real, but only if the lawyer still reads the operative parts; a summary is a map, not the territory.
The way a summary misleads is usually by omission. An illustrative lease summary might say: "Break clause: the tenant may end the lease on the fifth anniversary by giving six months' written notice." Accurate as far as it goes. The clause itself makes the break conditional on the tenant having paid all rent due and giving vacant possession, and those conditions are what decide whether a break actually works. A client told only the summary version could serve notice, fail a condition and stay bound for another five years. The summary saved 45 minutes of finding; reading the clause it pointed to took four.
Capacity without hiring. The hours saved above only become money in one of three ways: you take on more work with the same team, you stop paying overtime or agency cover, or you deliver fixed-fee work at a better margin. For firms billing by the hour there's a tension: faster work means fewer billable hours unless the pricing changes. Whether to charge clients less when AI speeds up the work deals with that directly.
Faster first response. A prospective client who emails three firms may well instruct the first one that replies sensibly. An AI-drafted acknowledgement, reviewed and sent within the hour, with the next steps and what documents to have ready, is cheap to produce. It's hard to price in advance, but you can measure it: track how many enquiries become instructions before and after.
The sum is simple once you have the counts. Suppose the firm had 40 new enquiries in the quarter before and turned 9 into instructions, about 22%. In the quarter after, with every enquiry acknowledged within the hour, 42 enquiries became 13 instructions, about 31%. At an illustrative average fee of $1,800, those four extra matters are worth $7,200 in a quarter. One quarter is a small sample, and seasonal swings can move the numbers on their own, so compare the same quarter a year apart if you can before crediting the whole change to AI.
Consistency. The same checklist applied to every new matter, the same structure in every attendance note, the same key points in every engagement letter. Consistency is rarely the reason firms start with AI, but it's often what they value most after six months.
The downside, with prices attached
Review time. Every AI draft needs reading and correcting, and at the start that takes longer than people expect. Allow five minutes of review per letter and a third of the 15-minute saving has already gone. Budget for it; a business case that ignores review time is fiction.
Review also fails gradually. In month one everyone reads every draft closely. By month four the drafts are usually right, so reading turns into skimming, then into glancing at the first paragraph. That's when a letter goes out confirming completion on the 14th when the file says the 21st, because the AI picked the date up from an older email further down the thread. The guard is a fixed habit: dates, amounts and names in every AI draft get checked against the file, however good the last fifty drafts were.
Errors that escape. The expensive con isn't the error you catch, it's the one you don't. Courts have sanctioned lawyers for filing submissions that cited cases an AI tool invented. The fix is a rule, not a tool: every reference is opened at its source before anything leaves the firm. What AI still gets wrong in a small business covers the other error types and their checks.
Data exposure. Client material doesn't belong in consumer tools. Business plans don't train on your content by default, and they cost money: ChatGPT Business Standard and Claude Team Standard are each $25 per user a month, or $20 billed annually; Microsoft 365 Copilot Business is $21 per user a month on annual billing ($25.20 monthly), with a promotional $18 through 31 December 2026. Treat the business-plan premium as the price of keeping client confidentiality, not an optional extra.
Subscription creep. One associate expenses a personal ChatGPT Plus plan at $20 a month, a paralegal signs up for a transcription tool, a partner tries a PDF assistant. Within a year the firm pays for four overlapping tools, two on personal accounts outside its data rules. How to audit your AI subscriptions shows how to find and cut them.
A first pass through card statements and expense claims might turn up something like this (illustrative):
| Tool | Who | Monthly cost | Account | Decision |
|---|---|---|---|---|
| ChatGPT Plus | Associate | $20 | Personal | Cancel; the firm's Copilot licence covers the drafting |
| Transcription app | Paralegal | $17 | Personal | Cancel, and delete the client calls stored in it |
| PDF assistant | Partner | $15 | Personal; a free trial that rolled over | Cancel |
| Slide-design tool | Office manager | $12 | Firm card | Keep; no client material goes into it |
The money is small, $64 a month. The finding that matters is the transcription app: recordings of client calls sitting in a personal account the firm can't see, secure or delete.
Learning time. Plan on about five hours per person in the first month to learn the approved tool, the rules and the checks. For eight people that's 40 hours: a real cost that most budgets leave out.
Skill fade. A trainee who has never drafted a letter from scratch struggles to judge whether an AI draft is right. It won't show up in year one. It shows up when that trainee is supervising others. A cheap guard: trainees draft one routine letter a week from a blank page, then compare it with the AI's version and the supervisor's corrections. Twenty minutes a week keeps the skill that makes their later reviewing worth anything.
Lock-in. Prompts, saved projects and workflows built inside one vendor's product are awkward to move. Keep your best prompts and instructions in your own documents too, so switching later costs days, not months.
This con already has a live example. OpenAI is retiring custom GPTs, which stop running on 11 December 2026. A firm that built its engagement-letter assistant as a GPT now has to rebuild it, as a shared ChatGPT Project or through OpenAI's migration to a plugin. If the instructions, the clause library and three example letters also sit in a document the firm controls, that's an afternoon's work. If they exist only inside the GPT, someone first has to reconstruct what it was told, from memory and old outputs.
An eight-person law firm's first-year ledger (illustrative)
Assume the firm is already on a Microsoft 365 business plan. It buys Microsoft 365 Copilot Business for its five fee earners, while the three support staff use Copilot Chat, which comes with the business plan at no extra cost. Internal time is costed at a blended $60 an hour.
| Cost line | First year | Notes |
|---|---|---|
| Copilot Business, 5 seats at $21 a month | $1,260 | Annual billing, list price |
| Copilot Chat, 3 staff | $0 | Included with the business plan |
| Learning time, 40 hours | $2,400 | Mostly in month one |
| Policy, prompts and setup, 10 hours | $600 | One person, spread over the first month |
| Total first-year cost | $4,260 |
Now the saving. After a ramp-up month, assume the tool runs for 44 working weeks. What matters is the net time saved per fee earner per week, after review.
| Net hours saved per fee earner per week | Hours saved in the year (5 people, 44 weeks) | Value at $60 an hour | Value minus first-year cost |
|---|---|---|---|
| 0.5 | 110 | $6,600 | $2,340 |
| 1 | 220 | $13,200 | $8,940 |
| 2 | 440 | $26,400 | $22,140 |
On paper, even the cautious row pays. The catch is the word "value". Those hours only become money if the firm uses them for more instructions, faster fixed-fee work or less overtime. If they disappear into a slightly more relaxed week, the firm has spent $4,260 and saved nothing on its bank statement. Why AI saves time but not money explains how to capture the gain.
Three spending levels and what each buys
Minimal, about $0 extra. Copilot Chat on Microsoft 365 business plans, or Gemini on Google Workspace plans, plus a one-page usage rule. This buys safe drafting, rewriting and summarising of material you paste or upload. It doesn't bring AI into your own emails, files and meetings the way a paid licence does; Copilot Chat compared with Microsoft 365 Copilot sets out the difference.
Middle, about $20 to $35 per person a month. A business chat plan or a Copilot licence for the people who draft most, and perhaps one automation platform (Zapier starts at $19.99 a month billed annually) for tasks like routing enquiries. For most small firms this is where the ledger is strongest.
Outside law the middle level looks much the same. A five-person recruitment agency, for instance, might take ChatGPT Business for the two consultants who write most (the two-seat minimum, $50 a month on monthly billing), keep everyone else on Copilot Chat through its Microsoft 365 plan, and add Zapier at $19.99 a month billed annually to send each new application to the right consultant with a two-line AI summary attached. That's about $70 a month, and every line of it maps to a job someone was doing by hand.
High, $100 or more per person a month, or a project budget. Premium seats for heavy users (ChatGPT Business Premium and Claude Team Premium are $125 per seat a month, or $100 billed annually), specialist sector tools priced by quote, or a custom build. Worth it only when a specific high-volume task justifies it and the middle level has proved the use case first.
Costs that never appear on an invoice
- Client trust. Clients who discover AI use by accident feel misled, even when the work was checked. Decide in advance what you'll say. Whether to tell customers you use AI helps with the wording.
- Professional obligations. Confidentiality, supervision of juniors' work and any guidance from your professional body still apply to AI-assisted work. Check what yours says before you roll anything out.
- Management attention. Someone has to own the tool list, the rules and the prompts. Budget about two hours a week of one person's time, or the tools drift.
- The billing conversation. For hourly-billing firms, the biggest con may be commercial rather than technical: efficiency the client benefits from and the firm doesn't.
How to tip the balance your way
- Start with tasks where review is quick, such as routine letters and internal notes, so the net saving is high from week one.
- Pay for one business tool, not four individual ones, and cancel personal subscriptions used for work.
- Measure net hours for a month before you add seats; the sensitivity table tells you what the number has to be.
- Put every reference and figure through a check before it leaves the firm, without exception.
- Decide in advance where the freed hours go, and review the ledger at three months.
Further reads
- Is AI Worth It for a Small Business? How to Work Out Your Answer — Turn the ledger into a yes or no for your own firm.
- How to Calculate AI ROI for Your Business (Worked Example) — A fuller return-on-investment calculation with a worked example.
- How to Handle Staff Who Over-Rely on AI — The con that creeps up slowly: skills fading behind the tool.
- Is It Safe to Put Customer Data Into ChatGPT? — The data-exposure con in detail.
- How Much Should a Small Business Budget for AI? — Set a budget line once you've weighed the trade-offs.
- What an AI Consultant Does for a Law Firm, and What It Costs — For law firms weighing outside help with the setup.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Microsoft 365 business plans and Microsoft 365 Copilot Business pricing, ChatGPT Business and Claude Team pricing pages, Zapier pricing page.