- column
- TAX MATTERS
TIGTA finds shortcomings in IRS’s chat service
An audit by the Treasury Inspector General for Tax Administration (TIGTA) concluded that the IRS’s chatbot and live chat applications used by the Automated Collection System (ACS) lacked performance measures, reliable data, and adequate performance.
Related
Proof-of-stake rewards are gross income in year of receipt
Error in tax liability estimate does not void extension request or bar refund
Line items
TOPICS
On June 15, 2026, TIGTA issued an audit report, Opportunities Exist to Improve the Quality of Chat Applications, in which it concluded that the IRS expanded the chatbot and live chat applications operated by its ACS function without first establishing performance measures or ensuring the reliability of the data used to gauge whether those tools actually help taxpayers.
Background: The ACS function within the Small Business/Self-Employed (SB/SE) Division was the first IRS unit to deploy chat applications to interact with taxpayers, the goal being to steer them toward online and self-help options in lieu of the toll-free telephone line. Live chats are operated by human IRS assistors, who can exchange text messages, files, web pages, and other information with taxpayers in real time. In November 2017, the agency piloted unauthenticated live chats, which do not verify taxpayers’ identity and therefore are limited to answering general questions. It added authenticated chat in June 2019, in which an assistor verifies the taxpayer’s identity before disclosing account information, answers detailed questions, and takes actions relating to the taxpayer’s account. The IRS made the live chat program permanent in June 2025.
A separate automated chatbot offering preprogrammed responses in English and Spanish launched in December 2021. These automated sessions can answer general questions and direct taxpayers to other self-help options; notably, they do not use generative artificial intelligence (AI).
Congress directed the IRS to broaden its digital footprint through the 21st Century Integrated Digital Experience Act of 2018, P.L. 115-336, and the Taxpayer First Act of 2019, P.L. 116-25, the latter requiring the agency to build a comprehensive customer service strategy and to develop metrics and benchmarks to quantitatively measure chat applications’ performance. TIGTA’s audit also occurred against a backdrop of IRS contraction: Supplemental funding authorized by the Inflation Reduction Act of 2022, P.L. 117-169, had been pared from roughly $79.4 billion to approximately $26 billion by February 2026, and ACS live assistor staffing had fallen 20%, from 159 to 127, amid a broader IRS workforce reduction from approximately 103,000 to 77,000 employees between January 2025 and May 2025.
Findings: TIGTA found that the IRS made the live chat application permanent and expanded the program without adopting any performance measures of its ACS chat applications, including assessing their effectiveness, accuracy, or effect on toll-free call volume. Compounding the problem, the data that has been collected is unreliable, TIGTA stated. Although live assistors are supposedly limited to conducting three simultaneous chats, records showed 193 assistors handling as many as 603 chats at once. Officials traced the distortion in part to assistors’ leaving chats open until the application force-closed them after four hours.
The majority of live chats were recorded as unresolved. Of 635,684 chats reported, only 290,181 (46%) were classified as resolved, leaving 345,503 (54%) unresolved. Yet the IRS could not determine why, because assistors are not required to document the reason a taxpayer is disconnected or referred to the ACS telephone line, and when a live chat is abandoned, the taxpayer is not provided an opportunity to provide feedback.
Furthermore, what feedback the IRS did collect it never reviewed, TIGTA found. The IRS had not offered a survey at the close of a live chat since September 2022, and although the automated chatbot asks taxpayers if its information was helpful, IRS officials acknowledged that they neither collect nor review the responses — even though the chatbot assures taxpayers their input is being gathered to improve service. A live chat taxpayer survey had been planned since April 2025, but it still awaited Office of Management and Budget (OMB) clearance as of December 2025. TIGTA noted that OMB Circular A-11, Section 280, requires federal agencies to collect customer feedback and act on it.
Two further findings carried operational and disclosure risk. First, ACS conducted only nonevaluative reviews of live assistors (which cannot be used in their performance appraisal), on the theory that live chat remained a pilot program, with evaluative reviews not slated until April 2026. Second, and more pointedly, TIGTA’s review of a sample of 40 assistors found that 24 (60%) worked multiple chats concurrently, and half of those had more than one authenticated chat open at the same time, which heightens the risk of disclosing one taxpayer’s account information to another in violation of Sec. 6103.
Finally, testing of the ACS automated chatbot itself revealed quality lapses. Of 206 sessions tested, 29 (14%) needed improvement, including for such mistakes as directing taxpayers to broken hyperlinks. Of 53 common keywords tested, 44 (83%) went unrecognized by the bot or drew insufficient responses. The Spanish-language chatbot, unlike the English version, offers no keyword search box at all, an equity concern TIGTA flagged directly.
The IRS’s Transformation & Strategy Office, IT, and various other business units were working on a long-term AI strategy, but the Transformation & Strategy Office was eliminated in March 2025. The AI strategy was placed on hold indefinitely after changes in leadership at Treasury and the IRS.
Recommendations: TIGTA made nine recommendations directed at the director, SB/SE Division, Collection Inventory Delivery and Selection. They call for the IRS to develop a process to validate the accuracy and completeness of its chat data; establish assistor quantitative performance measures and indicators; correct a programming error that miscalculates how long a live assistor takes to handle a chat; prevent multiple or blank resolution codes; implement and review a live chat survey while expanding the chatbot survey to capture written feedback; institute evaluative reviews and formal management procedures for the live chat program; and repair the chatbot’s broken links, expand its word bank to recognize more keywords and questions, and add a search capability to the Spanish-language version. IRS management agreed with all nine recommendations, stating that it had already implemented corrective actions for four and would complete the remaining five.
- TIGTA Rep’t No. 2026-308-029, Opportunities Exist to Improve the Quality of Chat Applications (June 15, 2026)
— Thomas Godwin, CPA, CGMA, Ph.D., and John McKinley, CPA, CGMA, J.D., LL.M., are both professors of the practice in accounting and taxation in the SC Johnson College of Business at Cornell University in Ithaca, N.Y. To comment on this column, contact Paul Bonner, the JofA’s tax editor.
