Census Bureau scrapped its $600m handheld computers for paper — and saved money

The 2010 census was to be America's first paperless count: enumerators would carry custom handheld computers instead of pencils, and the Census Bureau awarded Harris Corp a $600 million contract to build them. The Government Accountability Office had flagged procurement and management risk on the programme — the Field Data Collection Automation (FDCA) effort — as early as January 2005.

It went the way GAO feared. Requirements churned, the schedule slipped, and by 2007 the programme's life-cycle cost estimate had risen to $647 million with more increases anticipated; Harris blamed the growth on extra requirements. Field tests in May and June 2007 under census-like conditions found the devices slow and inconsistent at processing data. By March 2008, GAO told the Senate that overrun estimates for fixing the problems ranged from $600 million to $2 billion.

In April 2008 the Bureau pulled the plug on handhelds for the door-to-door follow-up and sent enumerators back to paper. The Commerce Department said the rework would add as much as $3 billion, pushing the total count past $14 billion. Deputy Census Director Preston Jay Waite — described as the architect of the 2010 census — retired within weeks amid the criticism, and administration officials conceded the Bureau had been unprepared to manage the contract, having never clearly spelled out its technical requirements to Harris.

Then the twist. In 2009 the Commerce Inspector General and GAO both ranked the 2010 census among the federal programmes most likely to fail. Instead, in August 2010 the White House reported the census was on schedule and 22 percent under budget, returning $1.6 billion in operational savings.

Overruns estimated at $600m–$2bn on the handheld program; the paper fallback added up to $3bn, pushing the count past $14bn

The Bureau never fixed clear technical requirements before signing, so requirements kept churning and every change cost money.

GAO recommendations went unimplemented for three years while the programme's baseline quietly eroded.

Realistic field testing came two years into the contract, when discovering the devices' limits was expensive to fix.

Keeping a paper option was treated as a boring contingency, but it was what made a safe retreat possible.

An automation plan that cannot survive field testing is a liability. The census survived because a paper fallback existed and leaders cut losses two years early — not because the technology got fixed.

The 2010 census went ahead on paper and was completed on schedule, finishing 22 percent under budget with $1.6 billion returned to the Treasury. The White House credited record mail-back participation — each percentage point of mail response saved about $85 million in door-to-door follow-up. The episode became a standing exhibit in GAO's warnings on federal IT procurement, cited as recently as the Bureau's planning for the 2020 count.