Evaluating AI vendors in financial services, on the record.

One vendor handles what happens when the customer comes to the bank, the other what happens when the bank goes to the customer, and a buyer comparing them should know they are closer to complements than to substitutes. Glia runs the interaction layer for community and regional banks, credit unions and insurers, unifying voice, chat, video, messaging and screen sharing so a conversation moves between channels and between AI and human agents without losing context or forcing reauthentication. Personetics runs the intelligence layer, enriching raw transaction data, then reading income stability, spending patterns and upcoming obligations across more than 150 million active monthly customers to surface a specific need such as a predicted cash flow shortfall. The grade shapes follow. Glia takes A on oversight for exercising control before the assistant speaks rather than after, through a proprietary approvals framework that gives the institution authority over what its AI agent is permitted to say, and A on integration depth for running inside the cores, digital banking platforms and even the incumbent contact centre an institution already has. Personetics takes A on AI centrality, because removing the models leaves a transaction list, which is precisely the raw material the platform exists to convert. Both take A on operational evidence.

The structural difference is who builds the model, and everything else follows from it. At Akur8 the machine builds it: Transparent Machine Learning automates feature engineering, variable selection and geospatial smoothing to construct risk models the company states are produced up to ten times faster than manual methods, while keeping every coefficient legible to the actuary. At hyperexponential the carrier's own actuaries build it, authoring, deploying and versioning pricing models in a Python engine, with no model learning the price. That produces the sharpest grade split in this pairing. Akur8 holds an A on model risk transparency, only the second in this index, earned structurally rather than by documentation: every model it produces passes two external checks the vendor does not control, a certified actuary signing under professional obligation and a regulator receiving it as a rate filing with power to reject. hyperexponential holds an A on integration depth, with named partnerships, the carrier's policy administration system left as the system of record, and a portability commitment giving customers structured access to all data, configuration and code at any time, which very little in this index offers in writing. Worth knowing before treating this as a straight contest: the two companies hold a partnership covering specialty and commercial pricing.

The cleanest question in this pairing is whether the machine ever decides. Federato says yes and describes exactly when: the system drafts the best triaged deals and explains its reasoning while the underwriter adjusts and owns the relationship, a defined subset of deals that clearly fit strategy is fully executed by the AI under predefined rules and guardrails, and those automated decisions are randomly audited by underwriters for quality control. Naming which decisions are automated, what constrains them and what the sampling control is, in public, is what almost every other vendor leaves implicit, and this index treats it as the reference point for the axis. Kalepa says no and draws the line just as narrowly: Copilot reads, cross references and scores, the underwriter decides, with the reasoning behind every flag visible. Both earn A on oversight for opposite reasons, which is the point. The second split is architectural and equally deliberate. Federato takes A on integration depth by sitting on top of the policy administration core with native connections to the three dominant platforms in this market, replacing the underwriter desktop rather than the system of record, on an eight to twelve week implementation. Kalepa takes C by design, working on day one with no integrations at all.