Why Benefits Enrollment Is Still Stuck in 2015
Ask a benefits broker how open enrollment actually goes and you'll get one of two answers. Either a polished pitch about their technology stack, or — if they trust you — the truth.
The truth involves spreadsheets. Paper fax confirmations. Carrier portals that time out. Nights in October that no sane person would choose.
I've been in enterprise software for 37 years. I've watched entire industries get rebuilt from the ground up — supply chain, payroll, tax filing, loan origination. Each time, the pattern is the same: the technology arrives years before the organizations are willing to trust it. And then one day they are. And the holdouts look ridiculous in retrospect.
Benefits enrollment is deep in that middle period right now. The tools exist. The workflows don't.
It's the same AI readiness gap I see across every industry — the technology is ready, but the organizational infrastructure isn't. The data sits in silos nobody's willing to connect. And brokers are eating the friction.
A calendar convention we never questioned
Here's something I find genuinely maddening: roughly 70% of benefit elections happen during a single three-to-six week window each fall. That's not a technical constraint. That's a calendar convention we inherited from paper-based administration and never seriously questioned.
During that window, the same brokers who spent all year building relationships are suddenly doing data entry — manually keying employee information into carrier portals that weren't designed to talk to each other.
Chasing down department heads for headcount updates. Re-entering the same demographic record into four different systems because none of them share a common identifier. I've heard brokers describe it as "controlled chaos." That's generous.
The downstream cost of manual entry errors
Research estimates manual data entry errors during enrollment create downstream reconciliation costs that can run $5,000+ per affected group annually — before compliance exposure. When an employee's dependent gets dropped because a form was misread, or a coverage tier is entered wrong, the cost isn't just financial. Someone doesn't get their medication covered. That's not an abstraction.
And then January comes. The crunch ends. And everybody goes back to pretending the system works.
A tax on broker time
I want to be specific about what "fragmented" actually means in practice, because the word gets used so casually it's lost its bite.
A mid-sized broker handling worksite benefits across a diverse client portfolio may be maintaining active logins to a dozen or more distinct carrier portals. Each portal has its own data format. Its own export logic. Its own two-factor authentication. Its own session timeout that kicks you out mid-entry. Most of them were built in the early-to-mid 2010s and have received cosmetic updates since then.
There is no common data exchange standard that carriers have collectively adopted for enrollment.
ANSI 834 transaction sets — the EDI format that was supposed to solve this — are technically supported by most carriers but inconsistently implemented. A broker's ops team will spend real hours every week massaging files to make them compliant with carrier-specific quirks. That work produces no client value. It's pure overhead.
Costs are going up. The administrative burden of managing those costs is going up. And the margin brokers work on isn't expanding to absorb either.
A slow grind
The Affordable Care Act's employer mandate has been the law for over a decade. You'd think we'd have smooth compliance processes by now.
We don't.
ACA compliance for applicable large employers requires tracking employee hours across often-irregular work schedules, maintaining accurate records for measurement and stability periods, and generating correct 1094-C and 1095-C filings annually. Any broker or TPA supporting these clients is involved in some portion of that workflow. And because the underlying HR and payroll data is messy — people change status, move between full-time and part-time, get rehired — the reconciliation work is constant.
IRS Section 4980H penalties can reach $2,970 per full-time employee per year (2024 figures) for employers who fail to offer minimum essential coverage. That's not a rounding error.
The cruel irony is that the data needed to get compliance right exists somewhere. It's in the HRIS. It's in the payroll system. It's in the carrier's eligibility file. Nobody's connected it in a way that gives compliance logic continuous access to ground truth.
And what it can't do alone
I want to be careful about this section, because the AI hype cycle has done real damage to serious conversations about operational improvement.
NLP for unstructured forms
Natural language processing extracts structured data from unstructured enrollment forms.
Anomaly detection
Models flag enrollments that look statistically inconsistent with household size before errors propagate downstream.
Document processing
Intelligent document processing replaces the manual re-keying of paper forms.
It's not a capability gap — it's a governance gap
Benefits data is among the most sensitive data an employer holds. It carries HIPAA obligations. It intersects with ADA considerations, ERISA fiduciary responsibilities, and ACA reporting requirements. An AI agent operating on this data needs to operate within a defined set of constraints that reflect all of that legal and regulatory context simultaneously. Not sequentially. Not in separate systems. Together.
Most AI deployments in this space today are point solutions. A chatbot that answers plan questions. An OCR tool that processes paper forms. They're better than nothing. But they're not connected to each other, and they're not governed in a way that makes a compliance officer comfortable extending them real authority.
What the market needs — and what I believe is technically achievable right now — is AI that operates inside an explicit governance framework. One where the rules aren't hardcoded into the model but are maintained as auditable, updateable policy logic that the AI references at runtime. Where every decision an AI agent makes can be traced back to a specific rule, a specific data input, and a specific authorization boundary.
That's different from "AI with guardrails," which usually means someone added a content filter. I'm talking about AI that treats regulatory and operational constraints as first-class inputs — not afterthoughts.
That's compliance as a competitive moat, not a cost center.
The brokers I respect have stopped asking "should we"
The brokers I respect aren't asking "should we use AI?" anymore. That conversation is done. They're asking how to deploy it in a way that doesn't create new liability, doesn't require them to become data scientists, and actually reduces the operational drag that's been bleeding their margins for years.
The answer isn't another carrier portal with a chatbot bolted on. It isn't an enrollment platform that automates data entry but leaves compliance logic to humans reviewing spreadsheets at midnight in November.
The answer is governed AI — agents that can act with meaningful autonomy on enrollment and compliance workflows because they're operating inside a framework that makes their actions auditable, defensible, and correctable.
Benefits enrollment is not a hard problem technically. It's a hard problem institutionally. The data exists. The rules exist. What's been missing is the architecture to bring them together in a way that regulated industries can actually trust.
That architecture is being built. And when it's in place, the October crunch — the portal logins, the manual re-entry, the reconciliation nightmares — is going to look exactly as avoidable as it always was.
Tired of the controlled crisis?
We built the governance layer specifically for this problem.
