Just as MyTwoCensus was getting ready to launch our multi-part investigative series detailing the many problems associated with Harris Corp. and their failed attempt to create a handheld computer suitable for all aspects of the 2010 Census, Government Executive’s Brian Friel beat us to the punch and published this column:
The Right Stuff
As Census Bureau officials continue to salvage what they can from the bureau’s failed decennial automation project, it has increasingly become a real-time case study in core problems plaguing the federal government’s contracting practices.
The original $600 million contract, awarded to Melbourne, Fla.-based Harris Corp. in April 2006, would have allowed census workers to collect decennial data for the 2010 count by handheld device, rather than the old pen-and-paper way. The devices also would be used to update Census’ massive address list. Third, Harris would provide a variety of technology support services.
Two years went by, and then the entire contract went kaput. In 2008, Census and Harris officials ran to Congress with fingers pointed at each other as $200 million already sunk into the project basically went to waste: The handheld data collection project was a failure.
Now the Census Bureau has dropped the data collection and the major support services from the contract with Harris, leaving only the handheld-driven update of addresses. The new contract has a drastically reduced scope, but a significantly higher price tag. It will cost nearly $800 million.
The Commerce Department inspector general and other watchdogs have identified two big problems with the contract.
First, Census didn’t know what it wanted. As the IG noted in a March 2009 report, a significant problem was “the failure of senior Census Bureau managers in place at the time to anticipate the complex IT requirements involved in automating the census.” Its initial list of “requirements” in the contract grew and changed exponentially, adding layer upon layer of complexity. “Census changed requirements several times, which caused delays and increased costs,” the IG reported.
Second, Census set up a contract with Harris that allowed costs to spiral out of control. If the bureau had known what it wanted from the beginning, it could have written a fixed-price contract, which basically says: “Here’s what we want, here’s what we’ll pay you.” Instead, Census wrote a cost-plus contract, which basically says: “We’re not sure what we want, so we’ll pay you whatever it takes.”
In April, Vivek Kundra, the new federal chief information officer, told Congress these two problems are common across federal contracts. “The federal government doesn’t do a good job of defining what the requirements are,” he told Sen. Tom Carper, D-Del., at an April 28 Senate hearing. According to Kundra, if agencies do a better job figuring out what they want, they can set up more fixed-price contracts, which control spending more than cost-plus contracts. “Fixed-price should be most common,” he said.
Kundra identified a common problem that leads to “runaway contracts.” Every contract involving technology has two main sets of requirements. First, a set of business needs that an agency’s operational office defines. Second, a set of technical needs that an agency’s IT department defines. If the two groups aren’t working together to jointly define all the requirements — if one leaves the other out — then an agency won’t really know what it wants. “The way that happens is ensuring there’s a high degree of engagement from both the business side of the house and the technology side of the house,” he said.
In the Census Bureau’s case, officials realized they had that problem only after they already had sunk $200 million into their automation contract, and at a point when starting over was impossible. “By the time you find out the requirements have increased or the budget is out of control, it’s too late to make an adjustment,” Kundra said. “For far too long we’ve put good money after bad money.”
If you don’t know what you want but you pay for it anyway, chances are you’ll repeat that long-running mistake.