Software Rules Tax Preparation, But at What Cost?

It’s mid-April, which means it’s the end of tax season in America again, when those who haven’t yet filed their income taxes scramble to beat the impending deadline. This year, like every year, more of those filers will use software to help them prepare their taxes than ever before.

It’s been thirty years since the Internal Revenue Service began embracing technology in a big way: In 1986 the agency piloted a program for electronic filing. The initial project required an IRS employee to manually turn a modem on each time returns were received, and it could only process certain simple returns. From 25,000 returns in that pilot year, the program grew rapidly: to 4.2 million returns the first year the program went nationwide, in 1990; to 68 million in 2005, when electronic filing surpassed mailed returns; and to over 125 million last year, or more than 85% of all individual returns.

Today, computers are ubiquitous throughout the proccess of taxation. Since 2010, the IRS no longer mails out 1040 forms—even if you want to still fill out paper forms, the agency expects you to download and print them for yourself. 

The rise of electronic filing has been mirrored by the growing role and influence of tax prep software. In 2015, over 50 million people filed self-prepared electronic returns, accounting for 1 in 3 individual filings. While more taxpayers still rely on tax professionals, the balance continues to slowly shift toward software-assisted self-filing (in 2006, only 15% of returns were done that way).

In some ways, taxes are a natural domain for computer assistance. Tax legislation can mostly be modeled as a set of rules and criteria that apply under certain conditions. But the problem is that most tax codes were not written with automation in mind, so there’s a lot of work required to translate them into a technical specification. (As my colleague Robert Charette has noted, the Standard Federal Tax Reporter, which explains the U.S. tax code to accountants, has grown to over 70,000 pages). Not to mention the dozens of state and local tax regulations.

The upfront investment required to build a comprehensive abstraction layer on top of such large collection of requirements is a large barrier of entry to new competitors. That partially explains the success of Intuit’s TurboTax, which dominates the consumer market, processing more than twice as many returns as its nearest competitors, H&R Block and TaxAct, combined. Together, the three account for nearly 90% of returns filed electronically by individuals.

There are a number of reasons consumers choose software like TurboTax, with convenience and cost near the top of the list. (Disclosure: I’ve used TurboTax for many years, including this year). But not everything that’s good for TurboTax is good for its customers, and certainly not for the IRS.

For one thing, TurboTax has a vested interest in making sure the tax code stays complex or becomes even more complex over time. They have lobbied heavily against initiatives like California’s return-free-filing.

There’s also evidence that the sheer scale of TurboTax’s customer base has given them a wealth of valuable data, allowing the company to understand taxes as well as—and sometimes better—than the IRS. That came to light last year when TurboTax was forced to temporarily stop processing state returns after an unprecedented increase in fraudulent returns. A pair of whistleblowers claimed that TurboTax ignored its own internal fraud models, which were more reliable than those at the IRS. Similarly, I suspect that TurboTax has a large enough sample size of data to accurately reverse engineer IRS auditing risk models (which allows them to confidently offer audit protection for an additional fee).

Finally, there’s a danger to filers dependent on tax-preparation software: The more we rely on software like TurboTax, the more we risk falling into the complacency of the automation paradox, where we no longer know enough about how taxes work to hold it accountable or do our own sanity checks. Maybe we would be better off with a simpler underlying protocol than a user-friendly abstraction layer.

In any case, best of luck to those of you who have yet to file!