By Lynn Räbsamen, CFA | Advisory Board Member, CFA Institute | Author, Artificial Stupelligence
I wrote about AI potentially replacing the $500,000 advisor earlier in June this year. It stirred up controversy in financial circles. It was also wrong about one thing. Not the argument. The arithmetic.
I said a $60-a-month tool like Hazel could finally make the small client profitable to serve. Automate the paperwork, collapse the admin cost, hand the mass-affluent a real advisor instead of a guess from ChatGPT.
The math was clean because I left out the hard part.
The $60 Is the Easy Number
Hazel reads a client’s tax return, pay stubs, statements and meeting notes, then drafts a strategy in minutes. Ask it a question and it answers by drawing on the firm’s conversations, emails, documents and CRM.
Read that last sentence again, slowly. Every word after “drawing on” is an assumption.
It assumes the conversations were logged. It assumes the emails are connected to the system, machine-readable, and consistent in how they name things, not filing “Bob” and “Robert J. Henderson” as two clients. It assumes they’re current enough to resolve contradictions, and clean enough to separate signal from “talk Monday.” Then it assumes the same of everything else: the documents filed where they should be, the CRM actually filled in, every source clean, current, and permissioned. Most firms are none of these.
The tool is only ever as good as the foundation it stands on. The industry has a name for a foundation that meets all those conditions: AI-ready data. And the demo always runs on the version the buyer does not have.
“The demo runs on clean data. The firm runs on fifteen years of nobody cleaning it.”
Almost Nobody Has AI-Ready Data
This is not a hunch. It is one of the best-measured facts in enterprise technology.
Gartner found that 63% of organizations either do not have, or are not sure they have, the data management practices that AI requires. AI-ready data, in other words, is rarer than the hype assumes. Informatica’s 2025 survey of chief data officers found that just 12% considered their data of sufficient quality and accessibility for AI.
Those are large companies with IT departments. A solo advisor with a half-filled CRM and most of the client context living in his own head is not the exception to those numbers.
Unready data does not produce a smaller benefit. It produces no benefit. MIT’s GenAI Divide study, published in 2025, found that 95% of organizations deploying generative AI saw zero measurable return. The cause was almost never the model. It was data readiness and broken workflow integration.
“95% got nothing back. The machine was never the problem. The foundation was.”
The License Is Not the Cost
So the $60 is real. It is also the least interesting number in the transaction.
$60 a seat is the price of the software. It is not the price of becoming the kind of firm where the software works. That price is paid in data work: extraction, cleanup, structure, governance, integration. McKinsey found in 2025 that the organizations getting real returns from AI were twice as likely to have redesigned their data and workflows before they ever chose a tool. The order matters. Foundation first. Software second.
For a firm that kept clean records, that work is a project. For one that didn’t, it is a wall, no matter how many people it employs. Someone still has to do it. On Informatica’s list of obstacles, a shortage of skills sits right behind data quality and technical maturity, and plenty of firms have a problem with all three.
That leaves two options, and neither one costs $60. Either someone inside the firm builds and governs the foundation, which most cannot, or a vendor comes in to do it, which they will invoice for. The number on the website is the tip. The foundation is the bill.
“You don’t buy AI readiness for $60. You buy the disappointment that comes from skipping it.”
The Verification Tax
There is a particular cruelty in pointing AI at unready data. It does not fail loudly. It answers.
It produces a fluent, confident, professional-looking output built on incomplete or contradictory inputs. Then someone has to read it, catch what’s wrong, and fix it. MIT has a name for this. The verification tax: the hours spent checking and correcting AI output, quietly eating the efficiency the tool was bought to deliver.
A tool that saves an hour and costs an hour of checking has a name too. It is a hobby.
“Confidently wrong isn’t a flaw in the demo. It’s the default setting of a machine pointed at messy data.“
The Divide That Actually Matters
Here is the part that should bother the industry most.
The instinct is to assume the big firms win this, because they have the budgets. They often don’t. Enterprise data is spread across decades of mergers, legacy systems, and a dozen CRMs that were never built to talk to each other. More money buys a bigger cleanup, not a faster one, and plenty of well-funded firms stall in exactly that swamp.
The obstacle was never the budget. It was the coordination, and no budget buys that quickly. Worse, coordination alone isn’t enough. Cleaning the data takes people with legacy knowledge, the ones who can look at two contradictory records and know which one is stale and which one still matters. That judgment lives in their heads, not in the system. And when those people retire or leave, the knowledge goes with them. No AI can reconstruct it, because it was never written down. It can only read what’s there. It cannot tell you what the person who left would have known.
The real divide is discipline, and it cuts across the size chart. A small firm that kept clean records from day one may be closer to ready than a global bank will manage for years. One CRM, one book, one decision-maker, and a contained problem a single data engineer can actually finish. The same firm that was sloppy from the start is in trouble, but even its mess is small enough for one person to see end to end.
“The advantage was never the budget. It was the discipline to keep the data clean before anyone called it AI.”
So the question for any firm, at any size, is not how much you can spend. It is how much you already cleaned up before you needed to.
A Closing Observation
For years, the advisor’s moat was the paperwork. AI took that. The lesson everyone drew was that the work is gone.
The work is not gone. It moved.
It moved from the visible, billable busywork to the invisible, unglamorous foundation underneath. The AI-ready data. The connected systems. The governance that decides what the machine is even allowed to see. That work has no demo, no headline and no $60 price tag. It is also, now, the only part that’s scarce.
The advisor who wins the next decade is not the one who buys the tool first. Everyone can buy the tool.
It’s the one who did the boring work first.
Wall Street thought AI made the advisor cheap. What it actually did was make the foundation valuable. The market still hasn’t put a number on that one, either.
For more insights about what AI can or cannot do, check out my book “Artificial Stupelligence: The Hilarious Truth About AI“.
Subscribe here to be the first to receive my insights.
