AI for professional firms and SMEs, a 90-day plan

In short

AI doesn’t replace whole jobs: it automates tasks. The first step is knowing which tasks fill your team’s time.
The biggest gains in small organisations are in email, documents, research and reports, not futuristic projects.
Measure before and after: hours saved per week is the number that counts.
Without rules on data and checks, AI enters the company anyway: through the back door, on employees’ personal accounts.

Tasks, not jobs

Andrew Ng has said it for years: AI automates tasks, not entire jobs. An accountant isn’t replaced, but the task “log 300 supplier invoices” largely is. A lawyer isn’t replaced, but “find the termination clauses in 40 contracts” is.

So the starting point isn’t “which AI to buy” but “where does the team’s time go”. A week of notes (activity, duration, frequency) is worth more than any demo.

Ten uses that really pay off

Ordered from easiest to introduce to those needing more preparation.

Email: draft replies, reminders, difficult messages, translations. Typical saving: several minutes per email, tens for delicate ones.
Document summaries: contracts, minutes, circulars, rulings, with page references.
Research with sources: regulation, practice, market, competitors. Always checked against the primary source.
Meeting minutes and notes: transcription, summary, action list with owner and deadline.
Client communications: newsletters, circulars, posts, in your own style with human review.
Data extraction: invoices, receipts, bank statements into tables ready to check and log.
Data analysis: sales, collections, costs by centre, with charts and commentary.
Consistency checks: comparing financial statements and tax returns, checking totals, spotting anomalies.
Internal training: procedures turned into guides, quizzes for new colleagues, answers to the team’s FAQs.
Automations: email triage, invoices from inbox to logging sheet, weekly reports.

The method I recommend

AI drafts, a person checks and signs. Emails sorted and summarised, invoices read and proposed for logging, regulatory research with sources, first drafts of client memos and opinions. Every step is reviewed before it reaches the client, and the client knows: for professionals it’s in the engagement notice, as Art. 13 of Italian Law 132/2025 requires.

Calculating the return

A realistic example for a firm or small business with 5 people.

Item	Value
Hours saved per person per week (email, documents, research)	3
Working weeks per year	46
Hours saved per year (5 people)	690
Average fully loaded hourly cost	€35
Value of time freed	about €24,000 a year
Business subscriptions (5 × about €25-30/month)	about €1,500-1,800 a year
Training and initial setup	€1,500-3,000 one-off

Three hours a week is a cautious estimate for people who work a lot with text and documents. The point isn’t the exact number but the order of magnitude: tool costs are small compared to time. The real cost is adoption: learning, changing habits, setting rules.

Freed time must be used

If saved time just dissipates, the return is zero. Decide beforehand where it goes: more clients, more checks, more advisory work, less overtime.

The 90-day plan

Days 1-30: rules and first wins

Inventory: who already uses AI, with which accounts, for what. (Spoiler: more people than you think, often on personal accounts.)
Short policy: allowed tools, forbidden data, duty to verify, who decides.
Business plan for those using it with client data.
Half-day basic training on the team’s real documents.
Three simple use cases (email, summaries, research) with a shared prompt for each.

Days 31-60: processes

Pick two repetitive processes using the automation guide’s test: frequent, boring, verifiable.
Write down the current procedure, step by step.
Build a project (or a Skill) for each, with instructions, templates and examples.
Measure time before and after on 10-20 real cases.

Days 61-90: automation and measurement

Turn the most stable process into an automation with human approval.
Update notices and documentation (AI Act, Law 132/2025, GDPR).
Review the numbers: hours saved, errors found, team satisfaction. Decide what to extend and what to drop.

The mistakes I see most often

Buying before understanding. Licences for everyone, nobody knows what to do with them. Better a few trained users, then widen.
No data rules. The colleague pasting a client’s bank statement into a free personal account isn’t malicious: nobody told them how to do it properly.
Expecting perfection. AI makes mistakes: design the check, don’t hope it won’t be needed.
Delegating judgement. AI drafts; the decision and responsibility stay with whoever signs. For professionals it’s also a legal duty.
Projects too big. “An AI that runs the whole firm” never starts. “Invoices from inbox to sheet” starts on Monday.

A prompt to start the analysis

Map of the team's activities

I am [role] at [type of organisation, number of people]. Here are the activities we do every week with estimated time: [activity 1 — who — hours/week] [activity 2 — …] For each, score 1-5: frequency, repetitiveness, ease of checking, data risk. Then name the 3 activities where AI would save the most time with the least risk, the right tool, and a first experiment to run within a week.

If you want to go through this with a method and someone who has already done it in a firm, see the AI Check-up and other services.

Sources

Informational content, current as of the date shown. Prices, models and rules change often: always check official sources. Written with the support of AI tools and reviewed by the author.