Thinklytics

Employers Holdings Inc. · Financial Services · Austin, TX · 16 weeks

Policy data quality score raised from 61 to 94

A mid-market insurance carrier spent $1.2M on an AI underwriting platform but couldn’t launch it because the policy data quality scored only 61 out of 100. We improved the data quality to 94 within 16 weeks, enabling the carrier to proceed with their $8.4M underwriting project.

Challenge

The carrier spent $1.2M on an AI underwriting platform that demanded a data quality score of 80 across 10 categories. Their policy data scored 61. The vendor’s fix would cost $2.8M. They needed a faster, more affordable solution.

Approach

We analyzed the vendor’s data quality rubric and found eight failing dimensions. We fixed six by building automated pipelines that standardized address formats, resolved duplicate policy IDs, filled missing coverage start dates, and normalized coverage code mappings. The remaining two issues involved legacy records and needed manual review workflows.

Outcome

By week 16, the policy data quality score hit 94. We launched the AI underwriting platform three weeks after delivery. The carrier projects $8.4 million in annual savings from improved underwriting efficiency. Our total cost was $340K, a fraction of the vendor’s $2.8 million estimate.

Vendor quotes didn’t even come close to the real problem. Here’s what we actually found.

The vendor was ready to scrap the whole policy system just to fix some data quality issues. That felt like using a sledgehammer to crack a nut. So, we rolled up our sleeves and got to work. We pinpointed eight specific data problems causing all the headaches. Then, we fixed each one with targeted solutions. No need to tear everything down.

We nailed 6 of the 8 key data fields without lifting a finger.

Here’s the deal: we had eight problem spots to tackle. Six of those? Totally fixable with automated data pipelines. So, we set up nightly jobs, hit the switch, and just like that, six areas were compliant in four weeks flat.

How We Fixed 17,200 Messy Records, No More Old-School Hassles

We jumped into the data and found around 14,000 old policies with coverage types that were a complete mess, super unclear. Plus, there were 3,200 records where the beneficiary info just didn’t make sense. To fix this, we created review queues and pulled in four analysts to handle the workload. Over eight weeks, they went through everything and cleaned it all up nicely.

Results

  • $8.4M Annual underwriting value added
  • 61 to 94 Data quality score
  • $340K vs. $2.8M vendor quote
  • 16 wks Delivery to go-live

We had $1.2 million stuck in a platform we couldn’t use because our data wasn’t ready. The vendor told us it would take another $2.8 million to fix it. Then Thinklytics came in and handled the whole thing for $340,000 in 16 weeks.

Chief Actuary, Mid-Market Insurance Carrier

In the client's terms

The question

Where does this number come from, and can the path be trusted?

The situation

An AI underwriting platform the carrier had already paid $1.2M for required a data quality score of 80 across ten dimensions. Policy data scored 61. The vendor quoted $2.8M to remediate.

What we found

Eight dimensions were failing, and six of them were mechanical rather than structural: address formats, duplicate policy IDs, missing coverage start dates, coverage code mappings. The vendor had priced all eight as if they were the same kind of problem.

What it produced

The score reached 94 by week 16 against a threshold of 80. The platform launched three weeks after delivery, with the carrier projecting $8.4M a year in underwriting value. Total cost was $340K against the $2.8M quote.

The roadmap

Automated pipelines fixed the six mechanical dimensions, which is work that runs once and then keeps running. The remaining two involved legacy records where no rule could decide, so those went to a manual review workflow rather than being forced into automation that would have guessed.

The work behind this result

Related case studies

Frequently asked questions

Do you have to replace a policy system to fix data quality?

Not usually. This carrier's vendor quoted $2.8M and a platform replacement. Profiling found eight specific failing dimensions, six of them rule-based, which were fixed with automated pipelines for $340K in total, taking the score from 61 to 94.

How long does data remediation for an AI underwriting platform take?

Sixteen weeks here against a vendor threshold of 80, reaching 94. The platform, already paid for at $1.2M, launched three weeks after delivery with the carrier projecting $8.4M a year in underwriting value.

Which data quality problems can be automated and which cannot?

Address formats, duplicate policy IDs, missing coverage start dates and coverage code mappings are rule-based and were automated here in four weeks. Around 14,000 legacy policies with ambiguous coverage types and 3,200 with unreconcilable beneficiary data went to human review queues instead.