Earnix

Graded against Akur8 in the same pocket, which also sits at B, so the pocket stays internally consistent. The predictive models are what the platform exists to run and the vendor states plainly that predictive capability is not a recent addition but the thing the company was built on in 2001.

What holds it off A is the residue: strip every model and an enterprise rating engine, a rate deployment workflow and a policy administration integration layer remain, and those are products insurers buy in their own right. Its own framing supports the reading, describing itself as inserting a new software layer into an existing technology stack that adds intelligence to what is already there. The generative and agentic layers arrived later and partly by acquisition.

Asserts that gates exist and does not describe them, which is the settled B. The published vocabulary is guardrails, permissions, test coverage, governance, explainability and visibility for regulatory control, none of it specified. Two things are more concrete and are recorded without lifting the grade.

First, rate deployment runs through an analyse, simulate, approve and deploy sequence on role based workflows, so an approval step exists in the path, though it is described as a workflow rather than as a control on a model output and nothing says what an approver sees or may override. Second, an audit trail is stated to be generated automatically for every quote. The oxane lesson applies: a workflow that has people in it is not the same as an adjudication layer placed after the model with its scope stated.

Fourth clean instance of the hyperexponential precedent, after that vendor, provenir and SS&C. The company sells model governance as a product feature and publishes nothing about its own models. AI Studio is described as delivering guardrails, permissions and test coverage so agents behave predictably, and the platform is repeatedly described as explainable and auditable, with an audit trail per quote.

Those are shipped mechanisms applied over what the customer builds, which is what separates B from C. Held well short of A, and the contrast inside the pocket is instructive: Akur8 sits at A on this axis because its transparent modelling approach is the published product, whereas here explainability is asserted as a property and no validation methodology, accuracy measure, drift monitoring or versioning policy appears anywhere. Say it as before: a platform that sells model governance is not thereby transparent about its own models.

Clears the A bar on both routes and has a standing case study library. Named institutions with outcomes attached: Domestic and General pricing 40 times more plans per month, Warta among Poland's largest insurers deploying prices within hours, Co-operators externalising pricing into the platform alongside Guidewire.

Named executives on the record at named institutions: Ryan Potts, Director of Pricing and Profitability for Dealer Services at U.S. Bank, and Nofel Goulli, Deputy Chief Executive of BPCE Assurances and BPCE Vie. Further named customers include BGL Group, Gore Mutual and Hollard.

The strongest numbers remain anonymous in the usual way and are not what carries the grade: a leading United States auto lender at rate deployment from four days to one hour with a 16 percent margin increase and twentyfold return, and a United States auto insurer at 15 percent higher quote conversion. One analyst placement was found and deliberately not leaned on under the benchmark half life rule: a predictive analytics market leader designation from 2020, six years stale.

Nothing published on training data, retention or whether one carrier's portfolio experience informs models another carrier runs. The question is live rather than theoretical for a multi tenant pricing platform serving competing insurers in the same national markets, where pooled experience would be commercially valuable and competitively sensitive at once. Neither claimed nor denied. Akur8 in the same pocket reaches B here, so the gap is a disclosure choice rather than a category norm.

Nothing published. The exposure is larger than for a pure pricing tool because of the engagement product: real time next best action across service, distribution and digital channels operates on individual customer records to decide what offer a person is shown, and nothing states what is retained, how long, or what a carrier's data may be used for beyond that carrier's own decisions.

No security page, trust portal or enumerated certification surfaced across three searches. Written as a queued question rather than a silent refusal, and the check is specific and cheap: a company of this size selling to global carriers and a large United States bank will hold certifications, and the question is only whether it publishes them. Akur8 in the same pocket sits at B on this axis, so a favourable resolution would not be surprising.

A software supplier holding no licence and claiming none. One capability sits close to the line and earns nothing, which is worth stating because it will recur in this pocket: Filing Accelerator supports insurers preparing and submitting rate filings to regulators, and building a product that serves a regulated process is not the same as holding standing in it. The filing is made by the carrier in the carrier's name.

No fairness testing, no protected characteristic handling, no disparate impact analysis, and this is the most consequential gap on the record. Price optimisation and product personalisation in insurance are the single practice most directly targeted by consumer fairness regulation in this market: the United Kingdom regulator's general insurance pricing practices rules were made precisely because optimising price against a customer's likelihood of accepting rather than their risk produced systematic detriment to loyal customers.

This vendor sells exactly that capability, to carriers in that market among others, and publishes no fairness position beside it. It also sells personalised next best action to lenders across cards, auto and mortgage, where the same question arises under different rules. The contrast inside the pocket is direct and usable: Akur8 reaches B on this axis.

Nothing published on who bears the consequence of a wrong decision. The exposures are concrete and differ by side of the business: a mispriced rate deployed to market in an hour rather than four days is also a mispriced rate distributed at speed, an automated credit decision inside the lending product declines an applicant, and a personalised recommendation puts an unsuitable product in front of a customer. An audit trail records which of those happened and answers a different question from who is responsible for it. The record is silent on remedy, correction and appeal.

No model, family, provider or version named for the copilot, the studio, the agents or the orchestration layer. Notable because the generative capability was acquired rather than built: the company bought Zelros, a generative specialist for insurers and banks, and folded it in, and the acquisition is described in full while the technology inside it is described only as generative recommendation capability.

That is the Linedata pattern in a second pocket and within the same session, which makes it a pattern rather than a quirk. Named partners and integrations exist and earn nothing under the standing rule, being the lower rungs: Guidewire, Verisk, Duck Creek era policy administration systems.

Strong end of B, and held there for consistency with Akur8 rather than because the evidence is thin. Two integrations are named with specificity rather than as partner logos. Co-operators externalised its pricing and rating into Earnix integrated with Guidewire PolicyCenter, which is a live named deployment rather than a partnership announcement.

And the Verisk Electronic Rating Content integration for commercial lines is described as preserving deviations while cutting regulatory rating updates from months to days, which names an industry rating content standard and what the integration does to it. Naming a rating content standard belongs with naming a messaging standard or publishing a schema on the list of things that turn an integration claim into an artifact. Held at B: no interface documentation, no published connector inventory.

Delivered as software as a service and stated as such, and that is the whole of what is published. No hosting regions, no residency commitment, no single tenant or private option, no customer responsibility split, for a platform operating across the Americas, Europe, Asia Pacific and Israel and processing pricing data for regulated carriers and banks in each.

No pricing published. Licensing is described by third parties as enterprise subscription scaling with modules, policy volume and deployment size, with full platform or per module options, and none of that comes from the vendor. Recorded as a contrast within the pocket rather than a general observation: Akur8 publishes enough on its own site to earn a B on this axis, and Earnix, a larger and older company selling the same buyers, publishes nothing.

More than 80 customers across five continents, and unusually the breadth is across two regulated industries rather than one. Insurance side: personal and commercial lines, named carriers in Canada, Poland, the United Kingdom, South Africa and France. Banking side: consumer lending, unsecured loans, cards, auto finance, mortgages and dealer finance, with a large United States bank named.

Product coverage runs the full pricing lifecycle from modelling through rate deployment to customer engagement, plus underwriting and credit decisioning. Selling price optimisation to both an insurer and a lender from one platform is rare in this index.