Chapter 20: AI for Business and Professionals
Where the value actually is
Most business coverage of AI focuses on the wrong tasks. Drafting emails faster is real but small. The value that compounds sits in four places.
Volume reading. AI can help you work through long documents, research, competitor material, regulatory updates, and customer feedback that you would otherwise skim or ignore. Much of the gain comes from finally reading material that previously received no attention.
First drafts of structured work. Policies, plans, proposals, specifications, reports. Getting to a reviewable draft in ten minutes rather than two hours changes what gets attempted at all.
Preparation. Difficult conversations, negotiations, board meetings, client calls. Twenty minutes of preparation you would not otherwise have done, in the time you actually have.
Adversarial review. Having your own work attacked before someone else does it. The maker–critic loop, applied to anything with stakes.
By function
Sales. Research before calls, tailored follow-ups, objection preparation, turning notes into pipeline updates, drafting proposals from a template plus specifics.
Operations. Process documentation: the thing every organization needs and nobody does. Describe how something currently works and get back a usable procedure. Also: checklists, incident write-ups, vendor comparisons.
Finance and analysis. Formula construction, data cleaning, variance explanation, turning a model’s output into narrative for people who do not read spreadsheets. Verify every number, and know whether it computed or estimated.
HR and people management. Job descriptions, interview questions, performance feedback framing, policy drafts. Never for decisions about specific individuals, and never with employee data in a consumer account.
Marketing. Variant generation, audience-specific rewrites, campaign structure, competitor analysis, turning one piece of work into six formats.
Legal-adjacent work. Understanding contracts, identifying what to ask counsel, first-pass review against a checklist, plain-language summaries of complex documents. Not legal advice, and Chapter 8’s cautionary case is compulsory reading first.
Technical work. Code, documentation, error diagnosis, test writing, migration planning. The area where capability is furthest ahead.
The professional risk layer
This is what distinguishes business use from personal use, and it is where careers get damaged.
Confidentiality is the first rule
Keep client information, employee records, unreleased financials, trade secrets, and anything under NDA out of consumer AI accounts. Consumer and business terms differ materially, especially in how providers handle data. Check the terms that apply to the account you are using.
If your organization has not made a business tool available, that is a conversation to have before you improvise. Improvising is how most confidentiality incidents begin.
Your professional duties do not transfer
If you are licensed (law, medicine, accounting, engineering, financial advice, architecture) your obligations of competence, supervision, and confidentiality apply to work you produce with AI exactly as they apply to work you produce with a junior colleague or a template.
The Avianca case from Chapter 8 generalizes cleanly, and its actual lesson bears repeating: the court did not sanction the attorneys for using AI. It sanctioned them for failing to verify, and then for defending the output once doubt arose. Every profession has an equivalent standard, and no regulator has yet accepted the tool as an excuse.
Many professional bodies have now issued specific guidance. Find yours. It is short, it is current, and it is written for exactly your situation.
Disclosure
Whether to tell clients you used AI is a real question with no universal answer yet, and norms are forming quickly.
The workable position: disclosure matters when it affects what the client is buying. Using AI to reorganize your own notes is like using a spellchecker. Using it to produce the substantive analysis a client believes came from your expertise is a different thing, and quietly not mentioning it is a position you would not want disclosed later.
If a contract, a professional rule, or a client’s own policy addresses it, that governs. If your answer to “would I be comfortable if they knew?” is no, you have your answer.
Billing
If you bill by the hour and AI compresses six hours into two, bill two. This should not need saying, and it does. The efficiency belongs to your client unless your engagement says otherwise; the alternative is straightforward overbilling and it is the kind of thing that surfaces eventually.
The longer-term implication is that hourly billing sits uneasily with a technology that collapses hours, and this is a live conversation in several professions.
Rolling it out to a team
Start with a pilot, not a policy. Three or four willing people, real work, four to six weeks. You will learn more about what actually helps in a month of use than in a quarter of evaluation.
Write the policy before you scale. Cover which tools are approved, what information stays protected, what must be verified, what must be disclosed, and who can answer questions. A template in the resource library will help you begin.
Buy the business tier. The consumer product is not the same purchase. The difference is administrative control, data handling terms, and the ability to answer a client’s security questionnaire honestly. For any organization touching client data, this is not optional.
Train people in judgment. Product features may be obsolete in six months, while verification, prompt construction, and knowing what to delegate determine whether AI helps or embarrasses you.
Expect uneven adoption. Some people will transform their work in a fortnight and some will never use it. That is normal, and forcing the second group produces theater rather than results.
Measuring whether it is working
Time saved is a seductive and incomplete metric. Track three things:
Rework. If AI-assisted output requires more correction downstream, you have moved work rather than saved it.
Things now getting done that were not before. Usually the largest real gain, and invisible in a time-saved calculation.
Error escapes. Anything wrong that reached a client. This number must be zero, and it is the reason verification discipline is not optional.
What not to delegate
Decisions about people. Hiring, firing, promotion, discipline, performance ratings. Use it to prepare and to check your reasoning; never to decide.
Anything you cannot check. This occupies the bottom-right cell of the delegation grid. In a professional context this is where the reputational risk lives.
Relationships. The apology, the difficult conversation, the condolence, the personal thank-you. People can tell, and the discovery costs more than the time saved.
Final accountability. It goes out under your name. That is the whole arrangement, and it has not changed.
Case Study: The policy that came second
A mid-sized engineering firm found that roughly a third of its staff were already using consumer AI accounts for work: none of it approved, none of it configured, some of it involving client drawings and specifications.
The instinct was a prohibition. What they did instead: acknowledged the existing use without penalty, bought a business tier with proper data terms, and gave everyone thirty days to move.
The training was two hours and covered almost no features. It covered what may never be entered, how to verify, and the one rule that turned out to matter most — that any figure, standard, or code reference appearing in client-facing work is checked at the source, every time, regardless of how confident the output sounds.
The time savings mattered, but the larger result was a change in reading behavior. Engineers began reading the full text of standards and specifications they had previously skimmed because a tool that summarizes and answers questions makes a two-hundred-page document manageable.
Exercise 20.1: Find your volume-reading problem
Identify material you should read and do not: regulatory updates, competitor material, customer feedback, long reports.
Run one through an AI tool this week. Ask for the summary, then for what it thinks you would find most consequential, then verify that part at the source.
Exercise 20.2: Find your professional guidance
Search for your professional body’s current guidance on AI use. Read it; it is usually short.
Note anything that surprised you, and anything you are currently doing that it addresses.
Exercise 20.3: Run the disclosure test
List three ways you currently use, or would use, AI in client work.
For each, answer honestly: would I be comfortable if the client knew exactly how this was produced? Anything that gets a no needs either a change in practice or a disclosure.
Learn More About Artificial Intelligence
Learn more about the CLEAR, TRUST and LEARN methods for Artificial Intelligence