Picture it. An employer, let’s call her a CFO at a mid-sized manufacturing company in the midwest, decides it is time to go out to market on her group health plan. She has heard, from someone at a conference or a golf outing or a LinkedIn post she half-remembers, that certain brokers can get better rates. That some people have relationships, access, leverage in a market she does not fully understand. So she does what any reasonable person would do: she calls three brokers. Maybe five. She wants to cover her bases.
Each broker collects her census, her current plan design, her claims history if she can get it, and her renewal rates. Each broker prepares a request for proposal. Each broker submits that RFP to the same carriers in the market. And here is where the machinery reveals itself for what it is: a convergence of ACA non-discrimination rules, state rate-filing requirements, and carrier quoting systems means that identical data submitted by multiple brokers produces identical proposals. The carrier is not choosing to treat every broker the same. It is legally and operationally incapable of doing otherwise.
She has just dispatched a small army of professionals to retrieve identical information. Weeks will pass. Dozens of hours will be spent. And at the end of it, she will receive the same numbers, dressed in different fonts, delivered with competing promises of superior service. The market has spoken, and what it said is: there was never any leverage to begin with.
This is the RFP process. It has functioned more or less this way for forty years. And it is, from almost every angle you care to examine it, a system in the late stages of justifying its own existence.
The Black Box
While the broker waits, something is happening inside the carrier that very few people outside of it fully understand, and that the industry has never been particularly eager to explain. The underwriter takes the census: the age and gender distribution of the employee population, their zip codes, their dependent enrollment. She runs it through a base rate calculation. Then she pulls deidentified prescription data through a tool like GRX or Milliman, cross-references it against the group’s demographics, and begins extrapolating: what conditions are likely present in this population, what those conditions cost to treat, and what premium is mathematically required to cover that projected risk.
It is, at its core, applied statistics. Sophisticated, consequential, and increasingly automated applied statistics. The broker on the other end of this transaction has no visibility into the process and no ability to influence it. The rate that comes back is the rate the math produced. Full stop.
In the fourth quarter, when every employer in the country is simultaneously going out to market before their January first renewal dates, that math can take seven business days to complete. Often longer. Twenty-one business days is not unheard of during peak season. Three weeks for a calculation that, stripped of its institutional scaffolding, is a function of demographics and actuarial tables that a well-trained model could run in seconds. And after all that waiting, there is still the possibility of a DTQ: declined to quote. The carrier looks at the numbers and walks away. Weeks of work, for nothing.
The broker had no idea this was coming. She had no window into the underwriting. Which raises the question the industry has carefully avoided answering for decades: if the rate is driven by math she cannot see, influence, or meaningfully interpret, and the outcome is legally and operationally guaranteed to be identical regardless of who submits the paperwork, what exactly is her value in this transaction?
Speed. Increasingly, purely and embarrassingly, speed.
Who can get the census scrubbed and submitted first. Who has the carrier portal credentials organized. Who has the submission template filled out cleanly enough that it does not bounce back for a correction. The differentiation that remains in the RFP process is not intellectual. It is logistical. And logistics, as it happens, is precisely what artificial intelligence eats first.
Spreadsheet Theater
The quotes come back. Now the broker builds the spreadsheet: a side-by-side comparison of carrier proposals, premiums, deductibles, out-of-pocket maximums, network breadth, and a dozen other data points that matter enormously to the people who will actually use this coverage. The spreadsheet is assembled, formatted, branded, and delivered in a presentation. It is, in terms of the actual information it contains, identical to the spreadsheet every other broker submitted to the same employer.
The CFO from our earlier example is now sitting across from her third broker presentation of the week. The numbers are the same. They have to be. What differs is the pitch layered on top of them: the promises of white-glove service, the name-drops of carrier contacts, the suggestion, never quite explicit, that this particular broker has a special relationship with the market that the others simply cannot match. She has no way to evaluate any of it. She picks based on whoever made her most comfortable in the room, or whoever quoted the lowest fee, or whoever her CFO friend recommended. It is, for all its professional apparatus, a gut decision dressed in a spreadsheet.
This is the part of the process nobody in the industry likes to examine too closely. Because if you examine it honestly, you have to ask what problem, precisely, is being solved by having five different humans build the same Excel file.
The Question Nobody Wants to Answer
Here is the question the industry needs to sit with, not defensively, not rhetorically, but genuinely: if the rating methodology is actuarial math, and the comparison deliverable is a spreadsheet, and the regulatory and operational architecture of the market guarantees identical outcomes regardless of who submits the request, what part of this equation actually requires a human being?
I am not asking this to be incendiary. I am asking it because artificial intelligence is going to ask it, with considerably less patience, and the industry will be far better served by having developed an honest answer in advance. A model can scrub a census file in seconds. It can identify the optimal carrier mix for a given group’s demographics before the first coffee of the morning is finished. It can build the comparison spreadsheet, flag the outliers, surface the plan design tradeoffs, and present the analysis in plain language that a CFO can actually use to make a decision. It can do all of this simultaneously, for every group in a broker’s book, without a submission deadline or a Q4 backlog.
What it cannot yet do, and this matters, is exercise the judgment that comes from understanding a client’s culture, their tolerance for disruption, the political dynamics of a benefits committee, or the institutional memory of what happened the last time they switched carriers mid-year. That knowledge is real. It is also not what most brokers are currently being paid to deliver.
There is a version of this industry that leans into that distinction and builds something genuinely valuable on the other side of automation. There is another version that waits, defends the existing process, and finds out what happens when the technology stops asking permission.
The Reckoning
The RFP process is not broken in the way that things break, suddenly, visibly, with clear accountability. It is broken the way institutions break when they are never forced to justify themselves: slowly, quietly, and from the inside out. It has survived this long not because it serves everyone well, but because it serves the people with the most power to change it well enough that they have never had much incentive to try.
That calculus is changing. Not because the industry has decided to examine itself, but because the technology arriving in this space does not require the industry’s permission to begin replacing the parts of it that are purely mechanical. The census scrubbing. The carrier submission. The comparison spreadsheet. The timeline compression from twenty-one business days to something that does not require an apology.
What remains on the other side of that automation is a real and important question. But it is a question the industry can only answer if it is first willing to be honest about what the current process actually is.
The RFP is dead. It just does not know it yet.
Overnight Intelligence | Group Benefits in the Age of Machine | overnightintelligence.substack.com