Consultants, AI and the Knowledge Gap: Advice Can Be Outsourced. Accountability Can't.
Consultants, AI and the Knowledge Gap: Advice Can Be Outsourced. Accountability Can't.
Whether a consultant or an AI tool makes you better or lazier depends less on the tool than on how much you already know about the problem. Three levels of knowledge, three very different jobs for outside help — and one thing that never moves.
Picture a carpenter with thirty years on the tools. Hand him an architect's drawings for a renovation and inside ten minutes he'll tell you which beam is undersized and which detail will leak in the first big storm. Hand him a commercial lease to sign off, and he's out of his depth by clause four.
Same person. Same intelligence. A completely different relationship with outside help.
That picture matters because of a view you'll often hear from executives, usually just after a consulting invoice has landed: managers bring in consultants on things they don't understand, or don't feel confident enough to own. Lately the same line gets pointed at AI.
My view: it's partly true, and it's not the whole truth. (Declaring an interest up front — I run an outsourced finance practice, so weigh my opinion accordingly. I'm also an accountant who would be genuinely useless fixing a bath.)
The variable that matters is how much you already know
Whether outside help — a consultant, an adviser or an AI tool — makes you better or lazier depends mostly on where you sit relative to the problem. There are roughly three places to sit, and the help does a different job in each.
|
You know nothing: the navigator
Guides you through territory you can't read. You can't check the answer, so choose the helper on accountability.
|
You know a little: the sounding board
Shows you what could be done better. The most useful tier, and where confident mistakes slip through.
|
|
It's your own field: the co-pilot
Makes sure you don't miss anything. You'll argue with it, and still get there faster.
|
At every level: the decision stays yours
The help can change from tier to tier. Who answers for the outcome doesn't.
|
What really changes as you move up is your ability to check their work. That's the whole game.
When you know nothing: outside help is the navigator
Back to the bath. If the tap body behind the tiles fails, an accountant having a go alone is a genuine risk — to the house and, in NSW, to the law as well: plumbing and electrical wiring are classed as specialist work that needs a licence regardless of the cost of the work. An AI chatbot can explain what a tap cartridge is and which questions to ask the plumber. That's real value. It can't do the job, and it can't be held responsible if the job goes wrong. The carpenter with the lease is in the same spot: a lawyer — or, for a first pass, an AI tool — will walk him through the risk far better than he could alone.
At this level you can't verify the answer, so you choose the helper on something else: accountability. A licence, professional indemnity insurance, a name that answers for the result. This is where the case for a human expert is strongest and for AI alone is weakest — not because AI is less capable, but because nobody stands behind its answer.
The finance version: a finance manager asked to approve a system data migration design or a cyber incident response plan. Getting help here isn't weakness. Doing it alone would be.
When you know a little: the most useful place, and the most dangerous
Give the carpenter those architect's drawings again. He knows what's buildable better than most people who draw plans, but not the engineering calculations or every council requirement. An engineer, or a well-prompted AI tool, can show him options he hadn't considered. The accountant reviewing a supplier contract sits in the same spot: payment terms, price escalation and termination fees are comfortable ground; indemnities, limitation of liability and IP ownership much less so.
This middle tier is where outside help adds the most, and where it can go wrong most quietly. You know enough to feel confident, and not enough to catch a confident mistake.
The clearest Australian example is now a year old and still worth remembering. The Department of Employment and Workplace Relations commissioned Deloitte, under a contract of around $440,000, to carry out an independent assurance review of its Targeted Compliance Framework; the review ran from December 2024 to June 2025, and the department released the final report in August 2025. A University of Sydney researcher in health and welfare law, Chris Rudge, later identified up to 20 errors, including references to academic works that don't exist and a fabricated quote attributed to a Federal Court judgment. What first tipped him off was a book attributed to a Sydney University colleague that he'd never heard of — because it didn't exist. A revised version, published in October 2025, disclosed that a generative AI tool chain had been used in part of the work, and Deloitte agreed to repay the final instalment of the contract; the department said the review's substance and recommendations were unchanged. The department's page now shows a further corrected version, posted in February 2026.
Notice who caught it: not the client, not the pre-release checks, but someone who knew the field. It's one case rather than a statistic, but it shows the middle-tier trap cleanly — a polished document, a trusted brand, and readers who couldn't easily tell which parts were solid.
The discipline here is unglamorous. Keep asking why until you can explain the answer in your own words. Check the two or three claims the decision rests on, at the original source. And if the area matters to your role, study it — not to become the expert, but to become a better reader of the expert.
When it's your own field: a co-pilot so nothing gets missed
Now the experienced carpenter asks an AI tool to check his materials take-off and cutting list, or gets another senior chippie to walk the job before he quotes. Every so often he'll push back — that fixing is wrong for the timber — and he'll be right. He still finishes faster, with fewer things forgotten, than he would alone.
For a finance leader, that's AI or an external adviser working through the month-end checklist, stress-testing reforecast assumptions, or reading draft management reporting pack commentary for gaps. You'll find mistakes and you'll argue with them — yes, even with the AI. That's healthy. If you never disagree with your co-pilot, the likeliest explanation isn't that it's perfect. It's that you've stopped reviewing.
Think of it as a team of experienced people helping you steer the ship. They'll spot the reef you missed and suggest a better course. You're still on the wheel.
Consultants and AI are not the same kind of help
The critique lumps them together, but they fail in different ways.
A good consultant brings context, judgement about your situation, and accountability: a contract, professional indemnity cover, and a firm that can be held to account — as the DEWR case shows, poor work can end in rework and a partial refund. The costs are the fee, and a commercial incentive to find more scope.
AI is cheap, immediate and endlessly patient with basic questions — excellent for getting oriented in unfamiliar territory and as a checklist in familiar territory. But it carries no accountability, delivers errors in the same confident tone as facts, and raises a data question: depending on the product and your settings, what you paste in may be retained or used to train future models. The Office of the Australian Information Commissioner recommends, as best practice, not entering personal or sensitive information into publicly available generative AI tools. In our sectors that covers client, participant and staff data, so check your tool's data-use and training settings before anything identifiable goes near it.
Advice can be outsourced. Accountability can't.
So is the critique right? Partly. If a manager commissions a report to avoid owning the answer, that's a real problem, and swapping the consultant for AI just makes the avoidance cheaper.
But the fix isn't to ban outside help. It's to match the help to how much you know, and keep the decision where it belongs. One test works at every level: without the consultant or the AI in the room, could you explain the recommendation to your board in your own words, and defend it when challenged? If yes, you used the help well. If no, you outsourced the decision, not just the work.
It isn't ideal to take whatever you're told, even from people who charge a great deal to be called experts, and getting something right in an unfamiliar area still takes some study of your own. But used with your eyes open, outside help makes almost everyone better at the job.
Note: The carpenter and accountant examples in this post are illustrative and not based on any specific individual or organisation.
This post is general commentary based on publicly available information and does not constitute legal or tax advice. Always seek independent professional advice before acting.
Senior finance help that matches what you already know
Some organisations need a navigator for the finance side of the business. Others need a second set of senior eyes on work their team already owns. PFL provides senior-level outsourced finance, management reporting, and AI automation for Australian NFP, NDIS, and SME organisations — and the decisions stay with you.
Start a conversation with PFL →Department of Employment and Workplace Relations — Secretary statement on progress under the Targeted Compliance Framework Integrity Assurance program
Department of Employment and Workplace Relations — Targeted Compliance Framework Assurance Review: Final Report (update history)
Associated Press — Deloitte to partially refund Australian government for report with apparent AI-generated errors (October 2025)
NSW Fair Trading — Home building licence check (licensing of specialist work)
Office of the Australian Information Commissioner — Guidance on privacy and the use of commercially available AI products
Comments
Post a Comment