Does Twentynext supply fraud detection software?

No. We are not a software vendor and do not sell a package with built-in fraud detection. We investigate the question, design the solution and build it until it runs in your own process on your own data.

How do you find fraud that has never been caught?

By not learning from labelled examples. A standard prediction model learns from labelled cases, and undetected fraud is, by definition, absent from those. Twentynext works unsupervised here: the model learns what normal looks like and flags what deviates significantly from it, with outlier detection and clustering.

Do we get a yes-or-no judgement per case?

No, a ranking: which cases deserve your investigators’ attention first. The threshold is set by your investigation capacity, not by the statistics. A hundred correct signals nobody looks at deliver nothing.

Does this also work outside healthcare?

Yes. The approach was developed with health insurers on claims data, but it is not healthcare-specific. Wherever large volumes of transactions flow and fraud is rarely labelled, the same principle applies: financial services, e-commerce, telecommunications or another industry.

Looking for expense management with built-in fraud detection?

Then you are looking for packaged software for claims processing, and we do not supply that. Our fraud detection starts where such packages stop: a model on your own transaction data, for patterns that built-in rules do not see. Not sure which route fits your situation? In one conversation you get a straight answer, even if that answer is: buy a package.

AI fraud detection: spotting patterns early

AI solutions for early detection and prevention of fraud

Fraud detection is the automated identification of transactions that deviate from the normal pattern: claims or payments that stand out statistically. Modern fraud detection uses machine learning for this: supervised where labelled fraud cases exist, unsupervised where they do not. In practice, the latter is the rule.

Why unsupervised

The hard thing about fraud is that you only know what you have caught before. A standard prediction model runs aground on exactly that: it learns from labelled examples, and the fraud nobody spotted is missing from the training data. So the model gets no list of fraud cases. It learns what normal looks like in your data and flags what deviates from it, with techniques such as outlier detection and clustering.

Where the real work sits

A model that flags deviations is the easy part. The work sits in what comes after:

How many alerts can you handle? The threshold is set by your investigation capacity, not by the statistics. A hundred correct signals nobody looks at deliver nothing.
Why is this a deviation? An investigator has to see what makes a case stand out. Without that grounding, an alert is not picked up.
What happens with what you find? Confirmed cases are the first real labels you have. They go back into the model.

Beyond healthcare

The approach was developed with health insurers on claims data, but it is not healthcare-specific. It applies wherever large volumes of transactions flow: financial services, e-commerce, telecommunications.

Fraud detection software or a model of your own?

Whoever searches for fraud detection mostly finds packages. Sometimes that is the right answer, and sometimes you pay years of licence fees for rules that fail to detect your specific fraud patterns. This is how to weigh it up:

Off-the-shelf software

Quick to start and proven on standard processes. You get generic detection rules and an annual licence; your process adapts to the package. Strong on known fraud patterns in common processes, limited as soon as your situation or your data differs from what the package was built for.

A model of your own, in your process

Runs on your own data and in your own process. No yes/no alarm but a ranking, so your investigators see the biggest risks first. No annual licence: the solution is yours. It does require a data foundation and an implementation process; we say that honestly.

In doubt? In one conversation we tell you honestly whether software is enough for your situation. That conversation takes 30 minutes, and sometimes the answer is: buy a package. We respond within one business day.

Schedule an introduction →

What you get

Not a package, but a working solution

Twentynext is not a software vendor. We investigate the question, design the solution and build it until it runs in production.

What we deliver

A solution that runs in your own process, on your own data. No licence for someone else’s software, and no report with recommendations.

Where we start

With the business question, not with the model. Which decision needs improving, and what information does it depend on? Technology is the last step, not the first.

Who builds it

The same people who design it. From data analysis and architecture to production and management, without handing over to another party.

How we approach a project

Solutions we have built