Neo4j aims its Palantir Gotham alternative at bank and insurer fraud desks

The four stages have names, and the second one is the stage investigators work in. Signal looks for suspicious patterns buried in connected data, and Alert hands investigators deduplicated warnings with the context behind each flag ^[7]. The context attached to each flag decides how many systems an analyst has to open before a flagged payment becomes a case file.

Underneath the stage names, the work is a join. The software brings data held in separate systems into one graph that analysts can query to trace links across accounts, transactions and devices, and Neo4j calls that kind of multihop reasoning a core strength of graph databases ^[6]. "Every fraud involves a network," said Michael Down, global head of financial solutions at Neo4j ^[11]. He said storing relationships natively lets a graph platform hop between data points quickly enough to surface suspicious behavior ^[12].

The market case around the launch comes from Interpol. Financial fraud cost victims an estimated $442 billion globally in 2025, according to an Interpol threat assessment published in March ^[3]. Interpol said in July that a single operation this year produced 5,811 arrests in 97 countries and territories and $293 million in intercepted funds ^[4]. Divide the first figure by the second and the annual loss estimate is roughly 1,500 times what that operation intercepted ^[18]. The loss estimate covers a global year, and the interception figure covers one police action.

Klarna is one of several fintechs using Neo4j for AI projects, the company said ^[14]. For a firm already running the database, the question this launch poses is whether a packaged four-stage workflow beats the internal one its own team has wired together over the years.

A pilot that takes twenty closed cases and splits each into two numbers is what would settle that: minutes spent assembling the picture across systems, and minutes spent judging it. A graph join only shortens the first number. If assembly runs under a third of case time, what a buyer is paying for is the Decide stage, which logs the outcome of each case and preserves the relationships and provenance behind the call as longer-term evidence ^[9].