Straight answers to fair questions

It complements any business system. Your claim notes — sentence data — are pulled from wherever they're stored and processed through our platform. Nothing is installed in your claim system, and nothing about it changes.

The best analogy is Excel: just as numerical data is pulled out of business systems and evaluated in a spreadsheet, sentence data is pulled out and evaluated in our technology.

No. There's no new software for your team and no workflow changes. We work entirely from the claim notes your adjusters already write, in any claim system.

Our starter engagement is deliberately simple: 5 core metrics measuring the operational activities you choose, a 5-month commitment, a single claim system, and a simple implementation. Each metric is configured to your protocols, your terminology, and your nuances.

Results arrive as monthly reads of every file — so you start seeing your team's numbers from the first cycle.

Every metric traces back to the exact notes it came from, so you can validate any result yourself. The same measurement is applied to every adjuster and every period — a transparent, repeatable methodology, not a black box.

No. We use your data — and only your data — to identify and measure events and activities for you. Nothing is pooled, borrowed, or shared.

Yes. We work with narrative documentation wherever it's captured and stored. As you'd expect, a poorly scanned or imaged document produces noise that has to be removed before effective grammar parsing can occur — we account for that in processing.

Our focus is documentation represented by typed text characters. If your solution requires handwriting recognition, we can license a third-party tool and plug it into our platform.

Coded data has two big constraints. First, codes are always backward-looking — built on historical or expected activity — while you care about what's happening right now. Second, codes compress away most of the knowledge about any topic.

A competitor working with sentence data will have far greater insight into any given issue — and have it much faster — than a business relying on coded data alone.

The critical difference is context.

Text processing and most AI tools focus on words: clusters, themes, and patterns found through frequencies and statistical horsepower. Sentence data processing identifies granular, factual information in the context of an individual sentence, all the sentences in a document, and all the documents within an entity — a claim, a policy, a patient.

That context is what lets us distinguish whether an event has happened, might happen, or has not happened.

NLP has become a very broad term — voice recognition, speech interpretation, pattern identification — and it means different things to different people.

We prefer to focus on the business answer you're receiving rather than the label on the method. Generally speaking, we use a variety of tools and strategies for identifying and extracting events and activities in context — happened, did not happen, or may happen — as needed to answer our clients' business questions. If you want to dig into the how, we're happy to.

Big Data touches all types of data — structured and unstructured — usually to surface themes, trends, and patterns across it. SDRefinery is focused on finding specific, granular facts that matter to a business user.

Consider Google, the ultimate Big Data engine: search a phrase and it scours hundreds of millions of records to return 4,000 for you to review. Big Data's expertise is finding the 4,000 records. Ours is reading and interpreting what's inside them.

Machine learning is broadly about using the computer to make decisions. Our process is about using the computer to improve the productivity of people.

Machine learning also depends on well-built training sets of specimen documents. Our technology helps users create comprehensive, accurate training sets more effectively and efficiently — so the two approaches complement each other.