Huscarl’s announcement that it raised $5.6 million to develop an “autonomous AI actuary” caught my attention. The company is targeting corporations and their insurance captives, and says every study is reviewed and signed by a credentialed human actuary.
For me, it raises a practical question about AI in actuarial consulting: could a small actuarial team complete work that once needed the resources of a much larger firm?
I think there is a real opportunity. AI can take on some of the preparation that absorbs an actuary’s time, giving a smaller team more capacity to investigate the business behind the numbers. How much additional work that team can take on will also depend on the review required.
For MGAs, carriers and captives choosing actuarial support, that makes the reviewing actuary’s involvement worth a closer look.
Where AI can save time in actuarial work
An actuarial analysis often starts with information that needs considerable preparation. Loss runs may arrive in different formats, and fields need to be mapped before the data can be used together. Someone also needs to investigate missing values or unexpected changes.
AI tools can assist with extracting information and preparing code to process it. They can also help flag records that deserve attention. The usefulness of that assistance depends on the task and how the output is checked.
For a smaller actuarial firm, reducing repetitive preparation could make a broader assignment manageable. It could also leave more time to explore a result with the client before the report is finalized.
I’d be careful about assuming that all the preparation time saved becomes additional capacity, though. A new process can introduce different errors, and the actuary needs time to understand where those errors are likely to occur.
Why actuarial data validation still needs judgment
Consider a field labeled “total incurred” in two claims files.
In one source, it might include claim expenses. In the other, it might exclude them. An automated process could map both fields to the same column and combine them without producing an obvious error.
The resulting exhibit could look perfectly reasonable. The calculation might work exactly as instructed. Yet part of the experience would be measured on a different basis, which could distort the loss experience used in a pricing or reserve analysis.
An actuary reviewing that data needs to establish what each field contains and whether the sources can be brought onto a consistent basis. That may require a conversation with the people who produced the files. A column heading alone may leave too much unresolved.
This is an existing concern in actuarial practice. ASOP No. 23, Data Quality, addresses the suitability and consistency of data, along with the disclosure of significant limitations. It does not require an audit of the data.
My view is that faster preparation makes it even more important to give these questions enough attention. I’d use some of the saved time to investigate differences that a rushed review might miss.
Faster preparation can mean more review work
Someone who has worked through an analysis manually may have noticed unusual records along the way. When more of that preparation is automated, the reviewer needs a way to see what happened during those steps.
For example, I’d want to know how the process handled a missing expense field. Did it leave the value blank, treat it as zero or exclude the record? Each treatment could affect the analysis differently. The reviewer needs to be able to trace that decision back to the source.
AI may help with the review itself, including identifying inconsistencies across a large dataset. An experienced actuary still needs to assess those findings and decide what further investigation is warranted.
That means some of the efficiency gain may need to go back into review, particularly while a new workflow is being established. The engagement should allow time for that work.
Questions to ask an actuarial consultant using AI
If I were choosing actuarial support, I’d want to understand how the firm uses AI and who takes responsibility for the resulting analysis. These are the questions I’d start with.
Who will review the work, and when will they become involved?
Ask about the reviewing actuary’s experience with the assignment and whether they will be involved while the data and assumptions are being evaluated. Find out how much access you will have to that person when questions arise.
How are differences between data sources investigated?
The consultant should be able to explain how field definitions are checked and discrepancies are resolved. For a program using several claims sources, ask how they establish a consistent basis for the analysis.
Can the team trace a result back to the underlying information?
Ask how they document changes to the data and the assumptions used. If an indication changes materially, there should be enough supporting detail to investigate why.
How much time is allowed for review and follow-up?
A quick turnaround needs to account for questions that emerge during the work. Ask what happens if the actuary finds a data issue that needs clarification from your claims team.
AI could give smaller actuarial firms more room to do this work well. Before appointing a consultant, I’d ask to speak with the person who will review the analysis and discuss how they would handle an unexpected result in your book.