The Internal Revenue Service made progress on its customer service and systems last year, but it’s still facing challenges with processing tax returns on time, according to a new report.

The report, released Thursday by the Government Accountability Office, noted that the IRS set a 13-day processing goal for individual paper returns but instead averaged 20 in 2024. In addition, IRS responses to taxpayer mail continued to be delayed, with 66% of them considered to be late at the end of filing season. The IRS has a web page showing the receipt date of taxpayer mail it is processing, but the page didn’t provide timeframes for when taxpayers should expect a response.

The release of the report comes as the IRS began another tax season on Monday while dealing with a hiring freeze imposed by President Trump in an executive order signed on the day of his inauguration, singling out the IRS for an even longer period when it won’t be able to hire. That move has prompted the IRS to rescind some of its job offers amid uncertainty over the more than $20 billion in budget cuts that Congress recently approved as part of a deal to avoid a government shutdown, on top of an earlier $20 billion in budget cuts.

During the 2024 filing season the IRS processed 98% of the nearly 174 million individual and business tax returns it received, as of April 19, 2024, according to the GAO report. However, the IRS continued to face challenges with timely processing of paper returns. For example, the IRS did not meet its 13-day goal for processing individual paper returns, instead averaging 20 days. In January 2024, the GAO reported that the IRS faced similar challenges processing paper returns during the 2023 filing season and recommended that the IRS determine the cause and address processing shortfalls. The IRS agreed and changed its reporting methodology in June 2024 to account for days in which the IRS is still awaiting taxpayer responses. However, the IRS has not yet documented the cause for the shortfalls.

“The 2024 filing season marked significant achievements by the IRS as we continued to modernize our operations by replacing aged equipment and transitioning our process to a more digitally integrated model,” wrote IRS acting commissioner Douglas O’Donnell in response to the report. 

He was named acting commissioner after Danny Werfel announced he would be resigning on Jan. 20, Inauguration Day. Trump had named a former Missouri congressman, Billy Long, to replace Werfel, even though Werfel’s term wasn’t set to end until November 2027.

“We remain focused on improving service to taxpayers, offering them more in-person and online resources as part of our effort to deliver another successful tax season,” wrote O’Donnell. “Taxpayers and tax professionals saw additional improvements in our operations and service in 2024 that made it easier for them to prepare and file taxes.”

AP-focused financial automation platform Stampli released an AI-driven procure-to-pay solution that works through dynamic conversations with the system itself. 

Stampli Procure-to-Pay, which uses the company’s AI assistant Billy the Bot, has the ability to turn plain conversational text into structured purchase orders and financial documents, replacing rote data entry. Stampli observed that procurement often involves dynamic conversations between multiple stakeholders, such as finance teams, employees and vendors. These conversations, though, tend to take place on separate channels, emails and chat tools versus the procurement platform itself, which breaks the audit trail. 

The conversation-first approach, on the other hand, allows the software to capture and contextualize all procurement discussions within each transaction’s workflow. It can support an existing intake process, approval workflow or business requirement. The system uses AI-powered approval flows guided by historical patterns. It automatically suggests preferred items and vendors based on an analysis of purchasing patterns. The program generates formatted purchase orders directly within customers’ ERP systems. 

“Only Stampli unifies all procurement processes, documents and discussion into a single intelligent workflow,” said Stampli CEO and co-founder Eyal Feldman. “Everything happens within Stampli: every step, every approval, every budget review, and every conversation, in perfect harmony with your ERP.”

For example, according to the CEO in a later email, a user could describe their office supply needs in plain language. Billy the Bot analyzes the text and extracts the item names, quantity and context, then maps them to stock-keeping units, descriptions and costs. This process requires no special training and is designed to encourage employees to use the system and not try to bypass or ignore it, which creates challenges in terms of compliance as well as workflow efficiency. 

Stampli said its new solution can be implemented within weeks. Once deployed, it will adapt to existing processes, reducing the amount of necessary change management. 

The announcement comes a few months after the release of another AI-powered solution, Stampli Cognitive AI, which the company said features “human-level” purchase order matching. Using large language models combined with nearly a decade’s worth of data on invoice-processing workflows, the model is said to understand the nuanced context of financial operations and replicate human decision-making processes.

Cloud ERP solutions provider Acumatica announced several updates and enhancements for 2025 that align with the company’s new AI-first product strategy, which involves looking at business problems from the ground up and then determine how applied AI can address these problem scenarios.

“Acumatica’s AI capabilities are designed to automate and eliminate error-prone manual processes and enable businesses to intelligently streamline workflows across industries,” said Acumatica chief engineering officer Miten Mehta. “From Generative AI assistants to smart automated sales workflows tailored to unique operational needs, our tools integrate seamlessly into existing workflows, making them immediately impactful. We’ve created a stable, secure foundation, ensuring businesses can confidently integrate AI technologies into their processes—whether AI-powered anomaly detection in manufacturing or intelligent assistants in construction—driving real, measurable results.”

Two major components of this will be enhancements to Acumatica Labs as well as the release of Acumatica AI Studio. Acumatica Labs will include a new customer preview program that enables early access to new features for testing and direct feedback, as well as advanced kitting, order orchestration, customer special orders, case closure notes, B2B ordering, document templates and AI-powered anomaly detection for those who sign up. 

Meanwhile, Acumatica AI Studio—a new feature in the larger platform—will let businesses automate their workflows without having to code. Chief Technology Officer Mikhail Shchelkonogov, during his presentation, showed how he could use it to automate report production, using the example of support case reports. After their engineering staff helps a customer with a technical issue, they must take all the case activities, analyze them, and report on things like what the problem was, what the root cause of that problem was, and how it was fixed, among other things. This is a complicated multi-step process. But Shchelkonogov, with the AI Studio, created a single button that directs the AI to gather all the information from the support case and then send it to the large language model to produce the report itself. He noted the AI even knew he spent 4 hours and 15 minutes working on that particular support case. 

He explained how he set up the button himself through the AI Prompt Editor that is part of the AI Studio. Effectively, the button activates a prompt he had already prepared which specified things such as the items he wanted to include in the closure notes, the format of the report itself, and additional information to be used to enrich the text (e.g. product descriptions from the Acumatica website.) He then tested the results, refined the prompt to his satisfaction and completed the automation. 

“Number one, I did not write a single line of code to do that. Number two, it was very fast. Number three, think of how many actions I automated. Now you can take it and go to any screen in Acumatica, think of the scenario you want to implement, and do it by yourself,” said Shchelkonogov. 

The AI Studio also provides data-driven insights in a secure environment. For example, during his presentation Doug Johnson, vice president of solution architecture, raised the example of an Alaskan company that makes scratch off lottery tickets. The insights feature combed through hundreds of thousands of records to find which of their salespeople were discounting too heavily, which were selling below margin and other performance statistics. They also used it to determine who was overdue with their bills (Johnson noted that billing can be challenging in Alaska due to mail delays, so standard patterns won’t work). They were able to find this information by entering a simple data query, which gathered the information and put it on a dashboard “so instead of 100,000 or a million records, you only have to look at 15.”

“Another thing here is the [AI determined] average days to pay. It varies wildly by customer. That is what AI really helps with, so they can focus on the right problem instead of focusing on people who just have mail issues at their place,” he said. 

Acumatica also announced several industry-specific solutions for the distribution, manufacturing, construction, retail and professional services industries. For professional services, Acumatica plans to implement a native ProjectManager integration for resource planning and project scheduling.

At this time, Acumatica hasn’t shared specific release dates for these updates.