Case study · AI & Data Engineering
An AI claims assistant that reads documents, flags fraud risk and settles simple claims in hours, not weeks.
Results
01The challenge
Harbor Mutual’s adjusters spent most of their day re-typing data from repair estimates, medical bills and photos. Simple claims waited weeks in a queue, customers churned after slow payouts, and fraud reviews relied on gut feeling rather than data.
02Our solution
We built a claims data platform on Snowflake, then trained models that extract fields from documents, estimate damage from photos and score fraud risk. A large language model drafts adjuster summaries with links to its sources, and low-risk claims under a set value are settled automatically — with every decision logged for auditors.
03Key features
Invoices, estimates and reports read and checked in seconds.
Risk signals explained in plain language for every claim.
Claim summaries and next steps, each linked to its evidence.
Low-risk claims paid automatically within agreed limits.
04Services delivered
05Tech stack
Chosen for this product’s needs — see our full tech stack.
06Timeline
Total duration · 7 months
3 weeks
Claims shadowing, data audit and value case.
8 weeks
Pipelines, data quality checks and a single claims model.
10 weeks
Extraction, fraud and damage models with human review.
8 weeks
Pilot with one team, then every line of business.
07The team
Softysta didn’t sell us AI for its own sake. They found where our adjusters were losing time, proved the value in a pilot and built something our compliance team signed off on first time.
Taking on new projects
Tell us what you’re building. We’ll help you shape it, scope it and ship it — whatever stage you’re at.
You’ll talk to a senior engineernot a salesperson.
Prefer to reach out directly?
What happens next
