12 weeks, 122 agencies: How AI transformed public spending oversight in Oklahoma

The problem

By 2023, most of the public money spent by the US state of Oklahoma was bypassing the checks meant to oversee it. Like many governments, Oklahoma asks its agencies to purchase through a central office that ensures each purchase is fair, competitive, and properly recorded. The state legislature's independent spending watchdog, the Legislative Office of Fiscal Transparency, examined the 2022 figures and found that outside purchases exceeded $3 billion, compared with about $538 million that went through it. It called the process slow and unclear and warned that weak oversight created "financial and legal risks to the State."

Two things had opened the gap. Agencies could use a long list of legal "exemptions" to skip the central check, and they could put smaller purchases on government payment cards, similar to corporate credit cards, which also avoid review. Meanwhile, the checking itself was painfully slow. A single complex purchase had to pass a 61-point checklist and could take up to 150 days to approve.

Catching problems was almost impossible. The job fell to a team of just six people within the Office of Management and Enterprise Services (OMES), the state's central finance and administration agency, who were tasked with monitoring thousands of transactions every month across all 122 agencies. Working by hand, they could review only about eight agencies a year. At that rate, it would have taken roughly 15 years to look at all of them, by which point any problem would have long been buried.

Then the pressure became official. The state governor issued Executive Order 2023-04, a formal directive, and the state's chief legal officer, the Attorney General, ruled that OMES was legally required to review every purchase claiming an exemption. OMES now had to review all 122 agencies within the year. The way to do that was to hire private auditors, at an estimated cost of $12.4 million.

The solution

OMES found a cheaper way, using software it already could run in-house. A senior OMES leader, Jerry Moore, had come across a tool from a company called Celonis that performs process mining.

Process mining is a way of seeing how work really flows through an organisation. Every time a staff member requested a purchase, approved it or paid for it, the state's financial software quietly recorded that step. Process mining reads all those digital records and reassembles them into a map of the route each purchase actually took. It is a little like an x-ray of a process: it shows how things really run, which is often not how the official rulebook says they should.

When OMES connected the tool to its financial system, the map was startling. There were around 3,000 different approval routes for purchases. More importantly, the team now had a live view of every purchase order, the formal request an agency raises to buy something, across every agency at once. Instead of digging through records months or years after the money was gone, they could see a purchase that broke the rules in real time, identify which agency or buyer was responsible, and step in before the order went out.

The core technique here is process mining. Celonis adds a layer of AI on top to point out where a process has gone off track and suggest how to fix it. In 2025, the state added an AI assistant called Process Copilot to make it easier to ask about those findings in plain language.

The effect was quick. As the state's chief information officer, Joe McIntosh, put it, "Within one week of implementing Celonis, we had data-driven insights that helped us understand our key financial challenges and how to address them." Checks that used to take staff days of searching could now be pulled up in seconds.

The impact

1. Every agency reviewed in months instead of years

OMES switched the system on in early June 2023 and completed its review of all 122 agencies by 1 July. In under 12 weeks it had checked more than 24,000 purchase orders, worth $4.58 billion, before the money left the state. The same work by hand would have taken about 15 years, and a year on, the effort appears to have satisfied the watchdog's concerns.

2. Money that could be saved by buying smarter

The tool showed that many agencies were not using the state's shared contracts, which are deals the state negotiates once, at better rates, for everyone to use. Instead, they were buying from single suppliers without seeking competing offers. By catching these and pointing buyers back to the shared deals, OMES has identified more than $174 million in potential savings. This is money that could be saved, not money already banked. Later figures from 2025 are larger and measure different things, including $190 million in payment-card transactions.

3. Faster purchasing

OMES says the average time to complete a purchase fell from 110 working days to 46, mostly by making the hold-ups visible and holding agencies to account. As Jerry Moore put it, the old timeline was "so long that vendors would forget that they even put in a bid." (For comparison, the watchdog's own report put the by-the-book average at 95.8 working days.)

4. Staff freed from manual checking

Six staff members who had been checking payment card transactions by hand were reassigned to other work because the software could do those checks faster. As Moore framed it, "Anytime you make a government operate more efficiently, you're able to do more with less."

OMES has described the in-house approach as $11.4 million cheaper than the outside auditors it had been ready to hire.

Lessons learned

• The real problem was that no one could see what was happening. Oklahoma's agencies were not, for the most part, deliberately breaking the rules. The system had simply grown too tangled to follow, with 3,000 different approval routes. The watchdog had said as much, concluding that the state first needed one shared system for recording all spending before it could enforce anything. Once the process was visible, compliance and consistency improved on their own. For any government struggling with procurement, better visibility can achieve more than another layer of rules.

• Spotting problems early beats finding them late. A normal audit looks backwards, reviewing what happened months or years ago, when the money is already spent. Here, the team could flag a purchase that broke the rules much earlier. For finance teams elsewhere, that move from cleaning up afterwards to preventing problems is the biggest change the technology offers.

• Keeping the skill in-house means it stays. The alternative was to pay outside auditors around $12.4 million for the year. The software did the job for less, and the know-how now belongs to the state instead of walking out the door when a contract ends.

• The hardest part was people, not technology. Janet Morrow, who leads the OMES team responsible for this work, has said the most difficult part was change management: getting more than a hundred agencies to understand the tool and how it would be implemented.

• It only works if the spending data is already in one place. The tool reads records from the state's financial system, so it requires that every agency's transactions be in a single system. A government whose spending data is scattered across many systems would first have to consolidate it. That is the main thing another administration would need before trying something similar.

This case study was written with assistance from artificial intelligence.