Neobank fraud budgets grew 357%, roughly as fast as the volume their loss rates imply

Divide each year's average loss by its loss rate and you get the transaction value the average institution handled. At 0.90 basis points, $800,000 of losses implies about $8.9 billion moved ^[15]. At 0.72 basis points, $2.84 million implies about $39.4 billion ^[16], roughly 4.4 times as much ^[17].

Prevention spending grew at almost the same pace. On the published figures of $1.8 million and $8.02 million it rose about 4.5 times, or about 346% ^[18]. The headline 357% would need a starting budget nearer $1.75 million ^[23], so the base is presumably rounded.

Put the budget over the volume and it barely moves: about 2.0 basis points of transaction value in both years ^[19], or 1.97 on the lower starting figure ^[24]. Add back what still got through and the full fraud bill, prevention plus losses, slipped from about 2.9 basis points of volume to about 2.75 ^[20].

A 4.4-fold jump in volume per institution in one year has three possible sources ^[17]. The same banks may have grown that fast. The panel may have changed, with larger institutions answering this year. Or the loss rate sits on a different base from the dollar averages, and the division above does not hold. The report summary does not say whether the same institutions answered in both years. One detail points toward a sample effect, since the rate for what appears to be the same year differs between two PYMNTS studies. An earlier study put the neobank loss rate for 2025 at 1.1 basis points, against 0.8 for its full sample ^[11]. This report's starting figure, for a group that also includes digital-only banks and fintechs, is 0.90 ^[2].

I'd put the weight on the first two readings, and either one means the 357% mostly measures size ^[1]. Only the third leaves room for prevention having become a much larger cost per dollar moved. If the full report shows the rate and the dollar averages come from different populations, this account is wrong, and the 357% is a straight cost increase.

The money is going into technology ^[14]. Every institution in the survey uses behavioral analytics and device fingerprinting, and 96% use machine learning ^[7]. Of the machine-learning users, 91% said it enables a proactive fraud strategy, up from 19% in 2025 ^[8]. Revolut's PRAGMA model, as PYMNTS reported in April, was trained on 207 billion data points covering 25 months of customer activity ^[9]. Those systems now face fraud that arrives with valid credentials. Account takeover through compromised credentials alone accounts for 28.2% of fraud dollars ^[6].