After the 2012 presidential election, I recall reading in some detail about the differences between the Obama and Romney campaigns’ digital efforts. The stories focused on the tablet applications to be used for supervising voter turnout and other details of election day, but also referred to the IT operations throughout the campaign cycle. The Obama campaign had developed all of their efforts in-house, tested with scores of volunteers, hosted with Amazon Web Services for flexibility, and had a great success. The Romney campaign had farmed out their efforts to consultants, done almost no testing at all, and found early on that Tuesday morning that they had laid a big goose egg. I will admit to an unseemly Schadenfreude at that point, but like most, came away with a great confidence that the Obama team “got it” in the digital realm. Apparently not.
In fact, one could almost think that the Romney campaign had designed the Obamacare website, healthcare.gov. The work was farmed out to multiple consultants and supervised by those who had no experience and simply didn’t comprehend the task at hand. Horror stories related by a friend who temped with the developer early this year bore this out.
Could I have done better? Well, the job was too large in scale for me to consider bidding on, but damned straight I could! Not that that’s saying too much, a properly staffed approach would have been far better.
Within hours of approval of the Affordable Care Act, President Obama should have chosen a champion to create the website for those states choosing not to create their own. That champion should have been someone with significant experience in databases, electronic marketplaces, or other-large scale networking projects. (Wikipedia and Google, among others, would know how to build a site that could be scaled from hundreds of simultaneous users to hundreds of thousands.) Most important, that champion should have been an employee of the government, reporting to an appropriate level, not an outside contractor.
A small team should have been assembled to design the flowchart for the site. Within a month the first version of the website should have been online. Not open to the public, and limited in function, but up and running.
A plan should have been established to connect to the IRS or the Social Security Administration to establish eligibility for discounts. At that point this could have covered a very limited population, say single persons in a few states, but something to test against. As the months went by, the interface could gradually include a wider range of users.
At least two insurance companies should have been recruited. One of these should have been chosen because their database was typical of other companies while the other was chosen because it wasn’t. There are several ways the databases could have been deployed. One possibility was to allow the insurance companies to access the database created in the exchange. Another would have been to establish a joint database for each insurance company and use native synchronization. Least useful would be to have a daily batch export from the exchange to the insurance companies, leading to the possibility that when a user made a change to his application the insurance company would end up with a duplicate record. Needless to say, this last approach was chosen. I doubt that this would have been the case if testing had been done at every stage of development.
Commerce sites are built using relational databases, mostly using the Structured Query Language or SQL. The nature of these systems is that the data itself is strongly structured. There are purposes for which nonstructured data is the rule, search engines frequently use tools that do not require an enforced structure. Transactions, inventory, mailing lists, and a host of other common data perfectly fit the traditional relational database. Insurance companies love them. Nonstructured databases are a relatively new development and apparently have some sex appeal or something. Those in charge of healthcare.gov chose a nonstructured database, making it much less likely that experienced developers would be available and that existing database code could be used to rapidly get the site working.
The site wouldn’t have needed to be pretty, just functional so the flow of user interaction with the site could be tested. A properly coded site can be given an entirely new appearance with a new style sheet the night before going public if need be. Because the Affordable Care Act was already law, there should have been no need to keep the site secret. The testing should have expanded as development progressed, first to key members of Congress and the administration, and their aides, then to a wide range of folks from the general public. At some point, truly expert useability experts should have been contracted, someone like Jakob Nielsen of the Nielsen Norman Group. Not to mention people knowledgable about access by the disabled, and not just to tick off the boxes required by law but to actually make it work.
There were things that shouldn’t have needed testing. Most notably, the question of whether or not a user needed to establish an account and verify his identity before shopping for plans. Rudimentary usability testing would have shown that this was a poor approach. On the other hand, having team members familiar with purchasing would have led to the same decision without testing. I’m not talking about complex Defense Department procurement here, merely sufficient experience and wit to go to a deli for a ham sandwich and come out with something involving bread and protein. Despite this low bar, they got this one wrong too.
The site is finally getting fixed, ever since Obama appointed a single person to take control. The vultures are circling, demanding that at least some heads roll from a great height. Before the Senate changed the filibuster rules this wasn’t likely because there was little or no chance of ever getting confirmation for a replacement if Kathleen Sibelius had gotten the axe, now this is possible. But I would say that while there are lessons to be learned and applied, no career-ending blame needs to be laid on anyone.
If I Were King, I would realize that the fault should be charged against “the system”. The structure of government made it unlikely, perhaps impossible, to assemble a staff to build this site on the understanding that most of the team would depart within weeks of the site going live. “The system” says this sort of work needs to be defined in advance so competitive bidding can be used to dole the tasks out to contractors. “The system” is wrong. If nothing else, nobody has ever defined a great website in advance, the best sites evolve over time based on lessons learned building out the initial plan. Obama and his advisors know how to do this kind of work, what they didn’t know was the importance of insisting that this site would be built the same way they built their campaign sites.