 |
| Michael Coverstone |
Docket.tax draws on records from the court's e-filing and case management system, Dawson, and layers AI-powered features on top of that data without compromising sensitive taxpayer information, said Michael Coverstone, a counsel at Kostelanetz LLP who also runs his own Florida-based practice.
Those functions include alerts for new petitions, concise summaries of opinions and calendar reminders for case deadlines, Coverstone told Law360.
"The calendar feature reviews the filings, pulls out the due dates and then populates a calendar for the user to follow," he said.
Docket.tax is an evolution of how Coverstone said he used to track cases — manually extracting information from Dawson's PDF docket sheet for each case and inputting it into spreadsheets.
Dawson's capabilities are limited, he said, because "the court is very serious about taxpayers' private information." The court launched the online filing system in 2020. Its name comes from the court's longest-serving judge, Howard A. Dawson Jr., and doubles as an acronym for "docket access within a secure online network."
Coverstone said the Internal Revenue Service's latest effort this year to settle conservation easement disputes initially inspired him to build an AI-assisted monitoring system for those cases in the Tax Court — called the CE Docket Explorer — that others in the tax practice can also use.
"There really wasn't a mechanism to track all these cases," he said.
The project quickly evolved into Docket.tax, which Coverstone developed with a small team to capture cases beyond conservation easements.
With a Docket.tax account, people can select any docket they want to track, sign up for real-time email updates and get summaries of new opinions. The platform can also filter cases by the assigned judge, as well as attorneys for the IRS and taxpayers.
As of July 20, Docket.tax was tracking nearly 64,000 cases, and the opinion archive contained about 15,900 opinions, going back to the early 1990s, Coverstone said.
Among the open cases, he said, there are roughly 2,300 attorneys representing petitioners, each handling an average of 6.4 cases, compared with about 1,090 IRS attorneys averaging 28 cases apiece.
He said he also learned from the platform that the attorney with the largest caseload, at 500 pending cases, was Patrick McCann of Chamberlain Hrdlicka White Williams & Aughtry, where he previously worked.
"There's a lot of cool Tax Court nerd data out there ... because the court is a very niche area," he said.
Here, Coverstone shared more details with Law360 below about Docket.tax.
This interview has been edited for clarity.
How does Docket.tax work?
All the information in Dawson is built in the public domain. In overly simplified terms, Docket.tax organizes that information differently and makes that information more accessible to our subscribers. It is a kind of data aggregation, data analysis and information-gathering tool that takes this information in Dawson and then puts it in a more user-friendly fashion.
What are Dawson's shortcomings?
Dawson gives practitioners exactly what they need as an electronic filing system, which it does really well. While the public can search for certain information in Dawson, it can be very difficult to find. It's not a legal tech software.
The Tax Court is in the business of making sure that there's an e-filing system that practitioners can use and in the business of adjudicating cases. They're not a tech company that would have the resources or the time to develop these types of things.
What role does AI play in Docket.tax?
The AI features add value to the information that is already in Dawson. For example, every time a case gets calendared, a pretrial schedule is issued. So the Docket.tax system will use some of the artificial intelligence to review the pretrial schedule and then populate those dates in the tracking software to help practitioners and petitioners use the tool to track their cases. AI is also used to put together plain-language summaries of the cases and identify the various issues that are discussed in the cases.
We also have some things in development, including using AI to summarize and analyze some of the orders that are issued. We hope to implement some of our AI tools to analyze which orders have substantive information and which orders are just run-of-the-mill extending the deadlines. We've spent a lot of time trying to develop the large-language-model prompts to get us the best information that is accurate and correct.
How does Docket.tax handle information it generates that is incorrect?
For the calendar entries, we have a process in place to make sure that people are not 100% relying on these deadlines and due dates without some verification. Our interface is set up to review all the dates and verify them, because a lot of them are AI making guesses or assumptions on what the correct dates are. But our interface allows the user to correct information.
Our interface also has features to allow users to flag an issue that is glaringly wrong. Then we'll go back and look at our system to make sure there are things we can do to make sure that the result is more reliable.
But at the end of the day, the filings and the opinions themselves are the best sources of information. We're just trying to help people summarize that information and make their lives a little easier.
AI is a great tool, but you have to verify.
What challenges did you encounter in developing this system?
Just the sheer amount of information that was out there was very daunting. There's a lot of stuff in Dawson. Then, just really working on the back end and the coding architecture has been a lot of the work behind the scenes to make sure everything in our system is reliable and doing what we want it to do.
The second-biggest piece is working on the AI and the various large language models to make sure we're getting the best results. We've spent a lot of time testing different models, whether it's Anthropic, Google Gemini or ChatGPT, to figure out what gives us the most reliable results with our prompts. That's been a big endeavor behind the scenes to make sure it results in a very reliable product.
What are the biggest limitations users should keep in mind when using this tool?
AI provides a lot of value, but it also gives us a lot of limitations. [There have been] improvements in AI we've seen over the past few months, but AI does make mistakes. We acknowledge that. So that's what we've been most focused on — making sure that the information our product is providing is as accurate as possible. We don't want to mislead anybody, and we want to make sure it's correct.
Do you see the Tax Court eventually adopting similar technology?
I think probably not. I think Dawson is a very well-thought-out platform for what it does. But I think the court itself has very limited resources, and the court needs to keep its resources focused on what it needs to do, which is to adjudicate, move cases and write decisions.
The court and Dawson have taxpayer information in their system. I think the court needs to make Dawson a little less user-friendly to protect taxpayer information, which is the right thing to do.
We don't have any of that information in ours. We have what the Tax Court has approved for release. So that gives a lot of flexibility as a private enterprise to kind of develop these new legal technologies.
What are your future plans for the platform?
We do have plans to reach out to low-income taxpayer clinics and give them an opportunity to use Docket.tax. Whether it's free access — or if it's not free, it will be at a significant discount. This will be a helpful tool for clinics to connect clients who need representation with pro bono counsel.
Getting low-income taxpayers with pro bono assistance to resolve their case will hopefully also ease the Tax Court's workload and help the tax community at large. If we can make things easier for people to do pro bono work, we're excited about that.
--Editing by Aaron Pelc.
For a reprint of this article, please contact reprints@law360.com.
