What AI Gets Wrong About Your Retirement, and How We Catch It
Written by David L. Blain, CFA — CEO, BlueSky Wealth Advisors
Earlier this year, I ran the same retirement planning case through three leading AI platforms. Each produced a polished, professional set of recommendations. Yet two of the three missed a planning opportunity worth an estimated $150,000 to $200,000 in lifetime tax savings, and neither gave any indication that something might be missing. All three answered the question as it was posed. None paused to consider whether the question itself was complete.
That distinction is important because it highlights where the real value of financial planning lies. If you’ve ever wondered whether you could simply describe your situation to an AI and trust the plan it hands back, the answer is nuanced: the most valuable planning work often comes from asking the right questions before any recommendation is made. It means identifying assumptions, uncovering blind spots, and understanding what may be missing from the picture. That is the work an advisor is trained to do.
Three capable tools, three different gaps
The case I tested was representative of the households we work with. The family, whom I’ll call the Brennans, held roughly $3.9 million in net worth and $407,000 in combined income, carried seven distinct planning complexities, and stood about four years from their first retirement date. Each platform received the identical case study and the same five-part planning question, word for word, so any difference in the results reflected the tool rather than the input.
The costliest omission concerned a fairly technical election on the husband’s deferred compensation. His plan was scheduled to pay out as a single lump sum, stacking a large amount of income into one year and into the highest tax brackets. The rules allow that payout to be re-elected, well before it begins, into a series of installments spread across several years instead. Applied correctly, it reshapes the wave of income that arrives when that compensation pays out, and in the Brennans’ situation, that single move accounted for the six-figure difference. One platform identified it, worked through it, and went a step further, raising a question about an inherited IRA that changes the entire withdrawal strategy. The other two never raised it. One built its whole approach around managing the income concentration rather than asking whether that concentration was necessary in the first place.
The gaps differed in kind, which is the instructive part. One tool produced by far the most useful format, a detailed spreadsheet model showing three scenarios side by side, and it was the only one to test whether the portfolio actually lasts. Its base case had the accounts running dry at age 84. That is very likely the single most important figure the whole exercise produced, and only one of the three tools thought to look for it. Yet that same model rested on a Social Security estimate that was almost certainly too high, and that one faulty input flowed through every projection in the file.
Another platform’s plan was the strongest on strategy, but it listed a key payout in one year on its timeline and a different year in the written summary. A small thing in isolation, but not something I could ever place in front of you when the document disagrees with itself. A third did the most precise tax work of the group, yet offered no scenario comparison and said nothing about estate or insurance considerations.
Every one of these tools is genuinely capable. Not one of them produced a plan I would have handed you as-is.
Even the best setup still fills in the blanks
Some of these tools were configured to behave like specialists rather than general assistants, and that does raise the quality of the starting point. But even in that mode, each one supplied assumptions no one had asked for: life expectancies, rates of return, inflation, the timing of major decisions, all delivered with the same confident polish as the rest of the work. A better starting point doesn’t remove the obligation to read what follows with care. If anything, seeing several capable attempts side by side is clarifying, because the points where they disagree tend to be the very assumptions worth questioning.
How we use these tools on your behalf
A few practices separate a useful instrument from a genuine risk to your plan, and we treat none of them as optional.
We confirm every rule-based figure against the original source before it ever reaches you. Required withdrawal ages, Medicare premium thresholds, Roth conversion rules, the mechanics of Social Security: we verify each against the IRS, the Social Security Administration, or the equivalent authority. A model will state a wrong number with the same assurance it brings to a right one, and these figures change over time.
We treat every output as raw material until a qualified advisor has reviewed it, verified it, and signed off. Our firm stands behind your plan; the software does not. Whoever’s name is on your plan owns every figure in it, both the ones the system calculated and the ones it quietly assumed.
We protect your information. We never enter your real names, account numbers, or personal details into a platform that hasn’t been vetted and approved. Describing a situation in the abstract, as I’ve done here with the Brennans, costs nothing and removes an entire category of risk.
And we hold ourselves to a written standard for all of this: the tools we permit, the verification we require, and how we handle your data, the same kind of discipline regulators expect of us.
What doesn’t change
AI has earned a real place in how we build retirement income plans. It’s faster, it’s an able partner in thinking, and at its best it surfaced a strategy two other tools missed entirely. Within the very same set of files, it also produced a number that didn’t hold up, contradicted itself on a key date, and introduced several assumptions without ever flagging them. Both things are true at once, and holding them together is most of the job.
The technology moves the work forward. The judgment, and the responsibility that comes with it, stays with the person whose name is on your plan.
