Putting AI to Work

Where to put AI, and where to deliberately keep it out.

Most organizations are measuring AI adoption. Almost none are measuring what it displaced. I spent twenty years in corporate finance at HP, Hewlett Packard Enterprise and DXC Technology, and I now build the automation systems that free small teams from routine work. This page is what I give audiences after a talk: the tools, the constraints, and the three questions worth asking before you automate anything.

Download the working guide Invite Alicia to speak

A working guide for people who have real jobs.

Eighteen pages, no jargon, and nothing in it requires hiring anyone. It covers which tool to use for which job with current pricing, how to write a prompt that produces something usable, how to turn off model training on your own data, how to run AI locally on your own laptop if you would rather it never touch a cloud service, and what to do differently depending on whether you run a company, sell real estate, raise money, serve on a board, or teach.

Every price was checked in August 2026 and every one of them will change. Where a claim comes from research or a regulator, the source is named so you can check it rather than take my word for it.

Download the guide (PDF, 18 pages)

PDF, under 1 MB. No sign-up required.

Prefer the short version? The five-page quick start.

New: De-slop your AI writing — how to make the machine stop sounding like a machine.

Three questions, and you can stop reading after any one of them.

One

Am I using this to prepare for a conversation, or to avoid one?

You will know the answer before you finish asking, which is what makes the question useful. Simone Heng, who studies human connection, draws the distinction as a way station versus a destination: a way station is somewhere you pass through on the way to a person. A destination is where you stop. The person who rehearses a hard conversation with AI and then goes and has it does better. The person who has it with the AI instead does worse. Same tool, same afternoon, opposite outcome.

Two

Did the effort matter here?

Usually not. Let the machine draft the agenda, summarize the board packet, clean up the memo. But for a condolence note, an apology, or telling someone you are proud of them, the effort is the message. AI can give you better words in four seconds. It cannot give away your twenty minutes, and the twenty minutes was the entire gift.

Three

When did this thing last tell me I was wrong?

If the answer is never, you are not using a tool. You are keeping company with a flatterer. Research published in Science in 2026 found that across eleven major models, AI affirms user actions markedly more often than other people do — and that exposure made people less willing to repair a conflict and more convinced they were right.

The uncomfortable numbers.

95%

of enterprise AI deployments produced no measurable effect on the profit and loss statement

MIT Project NANDA, 2025

4.6% → 45%

share of responses rated empathetic: physicians versus an AI chatbot answering the same patient questions

JAMA Internal Medicine, 2023

2 in 3

success rate when organizations bought a tool built for a specific workflow, against roughly a third as often when they built it internally

MIT Project NANDA, 2025

$7.7B

reported losses by Americans aged 60 and older to internet crime in the most recent reporting year, up sharply year over year

FBI Internet Crime Complaint Center

None of these numbers say AI does not work. They say the technology is not the hard part.

The judgment is the product.

Give it to the machine

Drafting the memo, the agenda, the routine customer reply
Summarizing a two-hundred-page board packet before you read it properly
Turning a meeting transcript into decisions, owners and deadlines
Research and first-pass comparable analysis
Rehearsing an objection you expect to face
The form-to-system handoff nobody enjoys doing

Keep it human

Any communication where the effort is the message
Valuation judgment, program logic, and technical conclusions you will be asked to defend
Anything a regulator, auditor or funder will hold a person accountable for
The conversation that is the only reason two people on your team still talk
Final approval on anything consequential
Telling someone something true and difficult

The rule underneath both columns is the same one I used for twenty years in finance: supervision proportional to the cost of being wrong.

If your interest is the finance function specifically.

Automation inside a close process is a different problem from automation on a marketing team, because the output has to survive review. The question is not whether a model can produce the entry or the reconciliation. It is whether the control still exists afterward, whether the evidence path is intact, and whether a person is accountable for a conclusion they can explain.

That work — where technology acts, where people approve, and what evidence remains — is what Dahling Consulting does.

Read Automate the Work, Not the Accountability See the engagements

Put the conversation in the room.

Keynotes, luncheon talks, board briefings and workshops. The AI talk is built for mixed rooms — owners, executives, professionals and nonprofit leaders in the same audience — and adapts to the group. There is also a version for finance and accounting audiences that goes deeper on controls, evidence and what auditors are going to ask.

Discuss a speaking date

01Putting AI to Work: The Human Side of the Shiftfor mixed business, civic and professional audiences
02Automate the Work, Not the Accountabilityfor finance, accounting and audit audiences
03The Cost of Control Debtfor finance leadership and boards
04ASC 606 Beyond Compliancefor revenue and technical accounting teams

What I am doing about my own gaps.

A few weeks ago I applied to BlueDot Impact's AGI Strategy course. BlueDot is a London non-profit that grew out of a Cambridge reading group and has taken roughly six thousand people through courses on AI safety and strategy, a number of whom now work at the frontier labs and at the UK's AI Security Institute. The courses are free on a pay-what-you-want basis.

I applied because of the syllabus. I expected a course on AI risk to open with threats. Unit One is called Imagining a Better Future, and before you analyze a single risk you are asked to write down the future you actually want.

You cannot build a defense for a future you have never described. That turns out to be true of technology strategy as well.

bluedot.org

Say what you are trying to figure out.

If something in the talk applies to a decision you are facing, write to me directly and describe it in a paragraph. I will tell you what I think, including when the answer is that you do not need help.

alicia@dahlingconsulting.com