At Tinubu’s 2025 Customer Advisory Board meeting (CAB), one of the most anticipated presentations came from Mitchell Wein, Executive Principal of Datos Insights. Wein, a renowned expert and researcher in the worlds of Surety and insurance shared his findings on how AI is changing Surety operations now, and his predictions for how it will do so going forward. After Wein’s talk, we wanted to learn more about how he sees AI’s role in Surety, so we sat down with him for an interview.

Wein has identified numerous ways that AI is affecting Surety, and he’s certain it will become even more significant, not just because the technology continues to be refined, but also because of “the evolution of the regulatory environment for Surety. Digitization isn’t available in all states and venues yet because some still require wet signature and physical seal, but I think we will get to the point where 100% of the states allow fully digital processes,” he says. “For example, what the federal government is doing to move away from physical checks and require two-factor authentication is going to reinforce the need for everything to be digital.” 

Below Wein outlines the ways AI is already influencing Surety underwriting, fraud detection and more, plus how the technology will further transform the industry going forwards. 

The Present

Underwriting Efficiency and Accuracy

“The area of Surety where AI is having the biggest impact is underwriting. We see it being leveraged in a couple of different areas, including early adopters using it in proof of concepts for risk analysis for larger construction bonds. The underwriting process for large bonds involves gathering disparate pieces of information which are often semi-structured or unstructured. This could include public information, nonpublic information provided by a company or supplier, photographs, various types of data sent by email. All this information must be unpacked and put in a format the underwriter can work with, so they can review it and price the bond. AI can unpack and map all that data, and place it in the right spots for analysis, which saves a huge amount of time.

AI can also look at the criteria the underwriter is using to come up with pricing and make recommendations. We don’t see autonomous utilization of Gen AI, because it’s a probabilistic technology that can still make errors. But paired with human intervention, Gen AI can really speed up the underwriter’s work.”

Detecting Fraud

“AI is very effective at fraud detection because it can pick up patterns in disparate transactions, for example, if different entities wish to establish Surety bonds and they only appear to be unrelated. AI can derive patterns such as common owners of different corporate entities that actually have the same address or are in fact the same people, which can suggest fraud. A lot of people are starting to use it for that analysis. 

The AI is looking at external data as well as internal data. It can scan the internet, it’s looking at data that’s been collected from the opening of other Surety bonds that an underwriter typically won’t scan. To be able to mine all that data in real time and then refine terms in a bond is something an underwriter could logistically never do.”

Determining Default Risk

“AI is useful for predicting the probability of default. AI can take a company’s financial statement, plus predictive indicators from third parties and the internet, including the nature of the company, its history, the history of its principals and articles about them. An underwriter could theoretically find and look through all that data, but it wouldn’t be timely. 

Then as the underwriter goes through the information and does the analysis, the AI can highlight possible default triggers, then recommend additional financial guarantees to add to the bond to cover those specific concerns. That would be for interpretation by the underwriter and inclusion after review. It has a huge productivity impact.”

Smarter Form and Document Management

“One challenge for those issuing small bonds is that it can be hard for bond libraries to keep up with the changes of forms from different counties, states and other issuers. [At Tinubu, the bond library is constantly updated, so it largely avoids this problem, but that is not the case for all Surety platform providers]. An agent may submit a bond, and find the bond is rejected because it was on the wrong form, or there’s a missing field. To avoid that, you can have AI use the bond library, and the website of the form issuer, and other websites because there may be, for example, articles on rules that have changed in a particular county. Then AI would identify which is the correct version of the form. This process can be embedded in the agent’s digital workflow. 

AI can also map the data that’s been submitted to the agent or broker into the form, determine if a piece of information is missing, then autogenerate an email back to the appropriate individual asking for the information.”

The Future

Litigation, Claims and Payouts

"We’re already seeing AI utilized for litigation assessment in insurance generally, but Surety is an area for that: sometimes bonds end up in litigation situations. You can use AI to help determine, for example, whether you’d want to litigate or settle. It can take into account factors like what venue it’s going to go in, who the judges are in the venue, what the inflation is around the awards there. 

Document review is another great use of AI: Gen AI is excellent in summarization. In fact, that’s the area where it makes the least mistakes. AI can get complex documents and parse them to help determine what they are communicating. Is this a true claim? Should the bond be called? If, for example, a vendor hasn’t provided the services and the bond is getting pulled, do we have a vendor that can step in and do the work? What is my suite of vendors in my various venues that can be deployed? What are their workloads? The automation of document review, the claims process, providing backup vendors where appropriate, whether to settle or to litigate: those are all areas of opportunity for Gen AI.”

Portfolio Analysis

“AI could be used for risk assessment across the portfolio for some of the larger carriers, such as US multinationals. Some of them have fronting partners in other countries essentially writing the bond and then reinsuring it back to a centralized entity in the United States that takes on the risk and receives a fee. The challenge with those operations is understanding portfolio risk. You might have several fronting partners that happen to be operating in the same area and not be fully aware of that. So if I write too many construction bonds in the Amazon where there’s an area of fires, I could have to pay out a lot of bonds if something gets out of control. 

Understanding portfolio level risk across a bond portfolio, and either pricing for that or assessing it and passing on an opportunity is something that’s not easily done—vendors have had mixed results historically—and AI could really help with that.”

SLMs and WIPs

“Something we haven’t seen yet in Surety that I could imagine happening is the use of small language models [SLMs], which are custom-built AIs, built with open source libraries, trained in a very specific domain space. This is conjecture, but somebody could create a model that’s specifically trained to a particular niche in Surety bonds. Perhaps you’re doing financial statement analysis for firms that build, for example, nuclear bunkers, which is a highly specialized analysis of high risk, large dollar bonds. If you had a custom small language model trained specifically for that domain niche, you’d potentially have an edge in terms of writing bonds in that space. I could see that evolving.

In the short term, Gen AI could be used to analyze Work in Progress [WIP] schedules that include project performance trends, backlog, cost to complete and remaining gross profit. We don’t have specific examples of this happening yet, but our opinion is that this should happen soon, perhaps in later 2025 or 2026. Remember, most insurers have just moved from AI proof of concepts in 2023 and 2024 to production implementations, so it is still early days.”

Despite it being “early days” for AI, Wein sees its role growing. As an example, he mentions SuretyDIGIT, an industry-wide coalition for the further digitization of the industry. “They’re starting to apply Gen AI around appropriateness of seals, powers of attorney and other areas,” he says. “That’s a pragmatic use of AI with broad industry support.” As Surety digitizes further, says Wein, “I think, frankly, the vendors that plug in their digital end-to-end solutions with AI are going to end up being the winners in the long run, because there will be growth opportunities.” 

Mitchell Wein is an Executive Principal in the Insurance Practice at Datos Insights. He has expertise in international IT leadership and transformation as well as technology strategy for banking, insurance (life, annuities, personal, commercial, specialty), and wealth management. Prior to joining Datos Insights, Wein served in senior technology management positions at numerous financial institutions including AXA and AIG. He has an MBA in Information Systems and a B.S. in Finance from Fordham University.