After HealthCare.gov stumbled through its launch, government investigators went to the agency that had built it and asked a simple question. Who was in charge?
Several staff named positions that were empty. Asked which division had led the project, staff named four different divisions.
The site had opened on October 1, 2013, as the federal marketplace for people in states without one of their own who needed to shop for health insurance under the new law. Outages began within two hours. By the end of the day, six people had managed to submit an application and choose a plan.
Six.
Not the traffic
The crowds were bigger than planned, and that was the first explanation the public heard. The Inspector General of the Department of Health and Human Services, which studied the project for its 2016 report, concluded that the traffic was not the whole story. The site had core performance problems of its own.
And behind those, the report put one thing first. “Most critical was the absence of clear leadership, which caused delays in decisionmaking, lack of clarity in project tasks, and the inability of CMS to recognize the magnitude of problems as the project deteriorated.” CMS is the Centers for Medicare & Medicaid Services, the agency that ran the build.
The warning signs were old. In December 2011, a CMS technical official complained that there was no single leader to make decisions and sign off on the business requirements. Mid-level staff, the report says, had trouble coordinating and lacked the authority to decide.
Early in 2013, after criticism that nobody was clearly leading, CMS did name someone: its chief operating officer. But the assignment was never formally announced. The job came with no clear responsibilities. And the person given it already had a very large job.
So even the fix for “nobody is in charge” had been half done.
A name beside every line
A task with four owners has none. When everyone is partly in charge, every hard decision waits for someone else.
Project managers have a plain tool for this, the RACI chart. For each piece of the work it records who is Responsible for doing it, who is Accountable for making sure it gets done, who should be Consulted, and who should be Informed. The point of the chart is the A. Every piece of work needs one name beside it, and everyone needs to know whose it is.
Apple shortened the idea to three letters. Inside the company, Adam Lashinsky reported in Fortune in 2011, the person who owns a thing is the “DRI,” the directly responsible individual. “Any effective meeting at Apple will have an action list,” a former employee told him. “Next to each action item will be the DRI.” When someone needs the right contact on a project, the question is “Who’s the DRI on that?”
Picture the meeting that doesn’t work this way. Seven people agree that the client’s complaint needs a reply by Friday. The minutes say “Team to follow up.” On Friday afternoon, each of the seven assumes one of the other six has written it. Nobody asked, because nobody’s name was on it.
Fill in the four letters and the same Friday runs differently. One account manager is R and writes the reply. The team lead is A and decides what the client is promised, so the decision stops there instead of circling the table. The engineer who knows what went wrong is C, asked before anything goes out. Sales is I, told once it’s sent, so nobody is surprised on the next call with that client. Four letters, four different jobs, and only one of them is “it’s mine.”
HealthCare.gov did, eventually, get its name. In August 2014, three months before the second year of sign-ups, CMS hired a chief executive for the marketplace. The report notes that the post was unusual for the agency, “but it answered the call for a single leader.”
Who answers for the bot
AI raises the question in a new form. When an AI agent does the work, whose name goes beside the A?
Air Canada tried an unusual answer. A passenger had asked the airline’s website chatbot about bereavement fares, followed its advice and was then refused the refund it described. When he took the airline to a small-claims tribunal in British Columbia, Air Canada argued it could not be held liable for information from its agents, including a chatbot. “In effect,” the tribunal member wrote, “Air Canada suggests the chatbot is a separate legal entity that is responsible for its own actions. This is a remarkable submission.”
He was unpersuaded. “It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot.”
The tribunal ordered Air Canada to pay $812.02.
That settled who answers to the passenger. It doesn’t settle who answers inside the airline. A company’s name in the A column is no more use than “Team to follow up.” The U.S. government’s AI risk framework, from the National Institute of Standards and Technology, asks for “accountability structures” so that “the appropriate teams and individuals are empowered, responsible, and trained” for managing AI risks.
The chatbot can be Responsible for an answer. Beside the A there still has to be a person, or a named role, with the power to change what the chatbot says.