Six Enrollments, 55 Contractors, and Zero Integrators

FICTION · An Unfair Advantage Intervention

Fiction. Real history. Sourced below.

Nothing about the HealthCare.gov diagnosis is contrarian. Test under load, name an integrator, make scope the variable, degrade gracefully — every post-mortem says it and every post-mortem is right. What makes the case worth studying is the second half nobody tells: a small team applied those same ordinary practices and had the site working in weeks, which proves the original failure was never technical. And then the government did the one thing that actually mattered, which was to stop renting the capability.

Jack opened the file. “This one’s a layup for blame, which is why I want us careful. Everybody knows it was a disaster. I want the mechanics, the real numbers, and the ending — because the ending is where the argument is. Timothy, engineering. And Sarah, I want you arguing against the framework by the end, because I don’t think it fully holds here.”

Full fiction disclosure and sourcing note

This is a work of fiction. Jack Whelan, Timothy Matthews, Sarah Jones, David Martinez, Mark Mitchell and Eugene Spark are fictional characters from The Unfair Advantage by Todd Hagopian, and Cartwell is the book’s fictional company. They analyze the real, documented history of the 2013 HealthCare.gov launch from the present day; they never advised any agency or contractor, never met or spoke with any real person, and no dialogue is invented for anyone. Cost figures vary by accounting method and are given as reported ranges. The historical facts are accurate and sourced as of 2026; the framework analysis is invented.

• Six
• Eleven Hundred
• Fifty-Five, and Zero
• The Anti-Consultant Manifesto
• Three Waves, and the Third One Lasted
• Sarah’s Objection
• From the Operator’s Chair
• The Launch-Readiness Playbook
• By the Numbers
• Frequently Asked Questions
• About the Stagnation Assassin

Six

“Six people enrolled on day one,” Timothy said. “October 1, 2013, roughly 4.7 million unique visitors, and six completed enrollments. The site crashed within hours, accounts wouldn’t create, plans wouldn’t load, and it came offline that weekend for emergency repairs.”

“Before we pile on,” he added, “the warnings existed. Months out, a senior technical official at the agency reportedly told colleagues he wanted to make sure the launch wasn’t a third-world experience. The people closest to the build could see it coming. Knowing didn’t change the date, and that gap between knowing and acting is the actual subject here.”

Jack: “So this isn’t a case where the diagnosis is disputed.”

“It isn’t. Every post-mortem reaches the same conclusions and they’re all correct. Which is what makes it interesting — because if the lessons are obvious and everyone agrees on them, the question isn’t what went wrong. It’s why obvious lessons don’t get applied by competent people with enormous resources.”

Eleven Hundred

“The number that should have stopped it,” Timothy said. “A stress test roughly one day before launch showed the site bogging down at about 1,100 simultaneous users. Eleven hundred, for a national marketplace the entire country had been told to visit.”

David came in. “And the governance knew. The readiness assessment that should have happened in March slipped to September, weeks out, and the system went live without the required performance verification and approvals. That’s the operational tell. When you slip the readiness review to the last moment and launch anyway, the review was never a gate — it was a formality you intended to walk past.”

“Which makes testing theatre,” Eugene said. “My question is always the same: did anyone in that building have authority to halt the launch based on a test result? If the date is fixed regardless of what the tests say, you’re not testing for protection. You’re testing for the record. And no contractor slows down to fix what can’t move the date.”

Timothy took the capacity numbers. “The system was designed for roughly 50,000 to 60,000 simultaneous users — the government’s own stated expectation. On launch day it took as many as 250,000 at once. Off by around five-fold, with no graceful degradation, no queue, no waiting room. Just cascading failure: timeouts, dropped sessions, corrupted applications reaching insurers.”

“And a bottleneck nobody had to build,” Sarah said. “The site required a full account, including identity verification, before you could browse plans or see prices. What did people actually want? To window-shop. So the design funnelled essentially all traffic through the heaviest, most fragile subsystem at the front door. That’s not an act of God. That’s a product decision manufacturing a chokepoint, and no amount of capacity planning relieves it.”

Fifty-Five, and Zero

“Structure,” said Mark. “Commonly cited as fifty-five contractors, with the lead website contract to CGI Federal — a contract that grew from about $94 million to roughly $292 million before launch. Total program costs run somewhere between $1.7 and $2.1 billion depending on whose accounting you accept. The tidy ‘half a billion’ figure people quote is wrong in both directions: larger and murkier.”

“Now the zero,” David said. “No designated lead systems integrator. The agency, CMS, effectively tried to play that role itself without the technical capability to do it. CMS thought CGI was the integrator. CGI didn’t share that understanding. So nobody owned the whole, and every contractor could truthfully say their component met spec while the assembled system failed — because integration was no one’s job.”

“Made worse by the contract economics,” Mark added. “Largely cost-reimbursement contracts, where the government pays allowable costs whether or not the thing works, against requirements that kept changing into the summer. Contractors reportedly didn’t have firm specifications until around March 2013, six months out. Changing requirements, cost-plus economics, no integrator, fixed date. That’s not a project. That’s a guaranteed overrun with a launch party scheduled on top.”

The Anti-Consultant Manifesto

Eugene went to the board. “Here’s why I think the framework fits, and I want to be careful because the name sounds glib. The Anti-Consultant Manifesto isn’t ‘never hire outside help.’ It’s a position about where capability has to live: the organization that owns the outcome must hold enough capability to direct, evaluate, and integrate the work, or it has outsourced not the work but the judgment.”

“CMS outsourced the judgment. Fifty-five vendors, each competent within scope, none accountable for the whole, and a client without the technical capability to be the integrator it was implicitly acting as. Every failure downstream follows. Nobody could evaluate whether the pieces would work together, because evaluating that requires the capability they’d contracted away.”

“Notice what this predicts that the ordinary reading doesn’t,” Eugene continued. “The standard post-mortem says the launch failed because of bad testing and no integrator. True. But those are symptoms of a client that couldn’t have run good testing or been an integrator if it had wanted to. You cannot demand a readiness gate you lack the expertise to assess. The stop button Eugene keeps asking about requires someone competent enough to know when to press it.”

Jack: “And the fix?”

“The fix is the proof. The tech surge pulled engineers from Google, Red Hat and Oracle, brought in Jeffrey Zients to run the effort, and finally named a real integrator, QSSI. And the repairs were entirely ordinary: add end-to-end monitoring, which barely existed at launch; add capacity and caching; build a waiting-room queue so overload degrades gracefully; fix the registration bottleneck. Within weeks a site that fell over at a thousand users handled tens of thousands concurrently. Largely stable by December. Roughly eight million enrolled by the close of the first open-enrollment period in spring 2014, above the original projections.”

“So the original failure wasn’t engineering difficulty,” Timothy said. “The same problems yielded to standard practice the moment capable people had authority. That’s the Manifesto’s whole claim, demonstrated inside eight weeks.”

Three Waves, and the Third One Lasted

“Run the rescue through Three-Wave Implementation,” Timothy said, “because it’s unusually clean, and because wave three is the part everyone forgets.”

“Wave one, stabilize. Monitoring first — you cannot fix what you cannot see, and at launch there was essentially no end-to-end visibility into production. Then capacity, caching, and the queue, so that overload became survivable rather than fatal.”

“Wave two, fix. The registration bottleneck, the architecture decisions that had routed every visitor through identity verification, the corrupted-application problems reaching insurers.”

“Wave three, institutionalize. And this is where the case stops being about a website. In 2014 the government created 18F inside the GSA and the U.S. Digital Service in the White House, with Mikey Dickerson — one of the rescue engineers — as the first USDS administrator. Agile methods, modular contracting, a digital services playbook, and a standing route for technical talent into government.”

“That’s the Manifesto made permanent,” Eugene said. “The rescue proved capability had to live inside the institution. Wave three built the institution to hold it. Most organizations stop after wave two, declare the crisis over, and rebuild the same conditions. This one didn’t, and it’s the reason the case has a genuinely good ending rather than a merely survivable one.”

Sarah’s Objection

“I’ve been asked to argue against this,” Sarah said, “and I actually do disagree, so let me.”

“The Manifesto reading is clean and it quietly excuses the thing that mattered most. Suppose CMS had held all the capability in the world. The date still couldn’t move. October 1 was set by the law and the coverage calendar, and politically a slip would have been read as proof the whole statute had failed. Fix the capability problem entirely and you still have an immovable date, which means quality is still the variable, which means you still ship something broken while everyone in the building knows it.”

Eugene: “You’d say capability wasn’t the binding constraint.”

“I’d say it was a binding constraint and not the binding one. A competent internal integrator with real authority would have produced a better-scoped launch, maybe a phased rollout, maybe a smaller working core. That’s a real improvement. It is not the same as a working launch, because the fundamental trade — fixed date, therefore variable quality — sat outside anyone’s technical authority. Notice the rescue team got something the original team never had, and it wasn’t only competence. It was permission to take the time.”

Jack let that sit. “So where does the room land?”

“That both are true and the framework is the more useful half,” Timothy said, “because the date was genuinely fixed and nobody in that building could change it, while the capability gap was theirs to close and they chose not to. You work the constraint you control.”

“Agreed on the practical point,” Sarah said. “I just want the record to show that ‘they should have had a stop button’ is easy to say about a launch where the stop button would have been overruled by a statute. That’s the part that doesn’t transfer, and pretending it does makes the lesson feel cheaper than it is.”

From the Operator’s Chair

I’m Todd Hagopian. Early in my career I sat in a launch review where the trade-show date had been announced to customers, the product wasn’t ready, and every person in the room knew it. We shipped. It was recoverable and it was humiliating, and what I remember most clearly is that nobody lied — everyone said the truth out loud and it changed nothing, because the date had already been made real in a way the readiness never was.

That’s why I don’t read HealthCare.gov as a government story. It’s the default failure mode of any organization where the date is more real than the readiness.

Three things. Make scope the variable rather than quality — if the date genuinely can’t move, cut features to protect a working core, because a smaller thing that works beats everything shipped broken. Give one person real authority to halt or de-scope, and make sure they’re competent enough to know when — a stop button held by someone who can’t evaluate the evidence is decoration, which is Sarah’s point turned into a requirement. And own your integration. You can outsource components. You cannot outsource ownership of whether the whole thing works, and the most reliable predictor of a launch disaster I know is that nobody internal has both the capability and the authority to make the pieces fit.

The part I’d actually put in front of a leadership team is the rescue. The fixes were standard practice applied by capable people with air cover. That tells you the original failure was structural rather than technical — and that a two-billion-dollar humiliation bought a lesson you can have for free.

The Launch-Readiness Playbook

• A stop button with a name and the competence behind it. One person with explicit authority to delay or de-scope, who can evaluate the evidence themselves. Authority without capability is decoration.
• Make scope the variable. If the date can’t move, cut features to protect a working core rather than shipping everything in a broken state.
• Test at the peak you might hit, then believe the result. Provision for worst plausible demand and assume your estimate is low. A test one day out that fails at 1,100 users is information, not a formality.
• Don’t manufacture your own bottleneck. Map the user’s actual path and check you aren’t routing everyone through your heaviest subsystem before they get value. Let people window-shop before registering.
• Name the integrator, internally. One accountable owner with the capability and authority to make components work together and reject those that don’t. It cannot be a gap between contracts.
• Design to fail gracefully, and instrument from day one. Queues, waiting rooms, preserved sessions, end-to-end monitoring. Overload should be survivable, and you cannot fix what you cannot see.

By the Numbers

• 6 — successful enrollments on launch day, out of roughly 4.7 million visitors.
• ~1,100 — simultaneous users at which a stress test bogged down the site about a day before launch.
• 50,000–60,000 vs ~250,000 — the concurrent load the system was built for against the peak it actually met.
• 55, and 0 — contractors commonly cited, and designated lead systems integrators before the rescue.
• $94M → ~$292M, $1.7–2.1B — the lead website contract’s growth, and total program cost estimates.
• ~8 million — enrollments by the close of the first open-enrollment period in spring 2014, above original projections.
• 2 institutions — 18F and the U.S. Digital Service, both created in 2014, with a rescue engineer as the first USDS administrator.

Frequently Asked Questions

Why did HealthCare.gov fail at launch?

Several causes compounded. The system was built for roughly 50,000–60,000 concurrent users and met about 250,000; a stress test a day before launch had already bogged down near 1,100 users; a registration-first design routed all traffic through the most fragile subsystem; there was no lead systems integrator, with the agency implicitly playing a role it lacked the capability for; cost-reimbursement contracts and requirements changing into mid-2013 undercut accountability; and an immovable statutory date meant launching without verified readiness.

What was the registration bottleneck?

The site required a full account including identity verification before users could browse plans or see prices. Because most visitors simply wanted to compare and get a quote, this funnelled essentially all traffic through registration and identity — the heaviest and most failure-prone subsystem — at the front door, creating a chokepoint that additional capacity could not relieve.

How was HealthCare.gov fixed?

Through a “tech surge” that brought in engineers from Google, Red Hat and Oracle, named Jeffrey Zients to run the effort, and finally designated a real integrator in QSSI. The repairs were standard practice rather than exotic engineering: end-to-end monitoring, added capacity and caching, a waiting-room queue for graceful degradation, and fixing the registration bottleneck. The site was largely stable by December 2013.

Did HealthCare.gov ultimately succeed?

Yes. Roughly eight million people enrolled by the close of the first open-enrollment period in spring 2014, exceeding original projections. The more lasting outcome was institutional: in 2014 the government created 18F within the GSA and the U.S. Digital Service in the White House, with rescue engineer Mikey Dickerson as the first USDS administrator, embedding agile methods, modular contracting and in-house technical capability across federal agencies.

Was the fixed launch date the real problem?

It is genuinely arguable. The October 1 date was set by statute and the coverage calendar and could not move politically, which meant quality became the variable — a trade no technical authority could override. But the capability gap was within the agency’s control and was not closed, and a competent internal integrator would likely have produced a better-scoped or phased launch. Both were binding constraints; only one was theirs to fix.

About the Stagnation Assassin

Todd Hagopian is VP of Product Strategy and Marketing at JBT Marel, having previously worked across Berkshire Hathaway, Illinois Tool Works and Whirlpool. He has shipped products against immovable trade-show dates, board commitments and fiscal-year promises, and writes as The Stagnation Assassin about what that pressure does to organizations — including the Anti-Consultant Manifesto’s central claim, that capability has to live inside the institution that owns the outcome. He is the author of The Unfair Advantage, a Forbes contributor, and an award-winning keynote speaker. Connect with Todd on LinkedIn or bring him to your stage via his speaking page.

The rescue used ordinary practices and worked in weeks, which means the failure was never technical. You can outsource components; you cannot outsource ownership of whether the whole thing works — and the stop button only functions in the hands of someone competent enough to know when to press it.

The discipline of keeping capability inside the institution that owns the outcome is in The Unfair Advantage.

Connect on LinkedIn · Book a keynote.