Infosys Finacle

The Clearwater precedent, and the clearest case on the whole roster once the shipped capability is separated from the platform. Strip every model and core banking, lending, origination, payments, cash management, wealth and treasury continue to run for banks in more than a hundred countries, because they have for decades.

What makes this the sharpest instance rather than merely another one is what the artificial intelligence line actually contains: the two generative assistants the company names are a knowledge assistant that searches document repositories and a support assistant that resolves internal help desk tickets, and neither is a banking capability at all.

The banking specific generative use cases, covering customer engagement, loans processing, reporting and compliance, are described as co creation engagements undertaken with individual banks rather than as products. The third component is a platform for banks to build their own models. So the shipped artificial intelligence is document search, help desk automation, and tooling.

No gate described anywhere. The shipped assistants retrieve information and draft ticket responses, both of which are advisory by design, and design is not disclosure: nothing states a threshold, an approval step, a review requirement or what a person must sign off. Monitoring appears as a named lifecycle stage on the model platform, and monitoring a model in production is not a control on an action.

The second order question is the same one raised by the no code canvas on this roster and this is its second instance: the platform is explicitly intended to let business users as well as technical users create artificial intelligence solutions, and nothing published describes who reviews a model a business user builds, what testing it must pass, or who owns its behaviour once it is deployed against the bank's own customers.

Graded B on the Provenir and Oracle precedent, which credits governance tooling supplied over the customer's models. The platform covers the model lifecycle explicitly, build, train, deploy, monitor and optimise from a single interface, and production monitoring is the part that matters most for a supervised buyer because model drift is what examinations look for.

Pre trained models and machine learning techniques are offered alongside it, and the stated purpose is to let users create explainable solutions. Held firmly at B and no higher for two reasons. Everything faces outward at the customer's models, and the vendor publishes nothing about its own: no accuracy or error rate for the knowledge assistant retrieving from document repositories or the support assistant drafting ticket responses, no validation methodology, no drift figures, no versioning and no external assessment. And explainable is asserted as a property of what a user can build rather than defined as a method, which is the pattern recorded against Abrigo.

A deep named client set in the vendor's own material and no number attached to any of it. At least seven institutions are named with described programmes, including a Gulf digital bank, a Colombian digital bank, an Indian digital brand, a Thai social bank, a capital markets cash management build, a European liquidity management programme and an Australian core migration, and the Finacle chief executive is quoted by name on the last of those describing reduced cost of ownership and increased automation.

Scale is stated at more than a hundred countries and over a billion end customers. Held at B because the outcomes published are qualitative, the one client carrying a hard figure is identified only as a residential mortgage lender with roughly sixteen billion in assets under management, and no result anywhere is attributed to a named institution. Notably, none of the named client stories is about the artificial intelligence suite.

Silent on the substance. Responsible by design is the framing and it names ethics, trust, privacy, security and regulatory compliance as standards the approach ensures, without stating a single commitment about data. Nothing says whether customer data trains or tunes anything, whether anything is pooled across institutions, what the assistants retain, or what is excluded from a training corpus.

The question is sharpened by the architecture rather than softened by it, because the data platform exists to consolidate a bank's data into a lakehouse for model consumption, and the model platform then trains on it, so the boundary between one bank's lakehouse and the vendor's pre trained models is precisely the thing left undescribed.

Privacy appears only as one item in the responsible by design list and nothing behind it is published. No retention schedule, no deletion terms, no subprocessor list and no tenant separation statement for a platform delivered as a service to competing institutions.

The exposure is broad because the estate holds account, transaction, loan and wealth data, and because the data platform is expressly designed to consolidate that material into a lakehouse for downstream model use, which concentrates it rather than distributing it.

No certification, attestation report, audit period or trust portal was located in the material reviewed, so the grade follows the standing rule that a credential must be found and named rather than assumed from the scale of the parent group. Security appears only as a word inside a broader responsible by design phrase covering ethics, trust, privacy, security and regulatory compliance, which is an assertion of intent rather than evidence of assurance.

This is unproven absence rather than evidenced absence and the queued check is cheap, since a group of this size holds attestations at corporate level; the finding worth recording is that a buyer researching a core banking platform running a billion people's accounts does not encounter a security credential in the product journey.

A software vendor with no licence, registration or supervised standing of its own, operating as a product subsidiary of a listed technology group. Regulatory compliance appears in the published material as one of the standards its design approach is said to ensure for customers, which is a claim about what the software helps the buyer achieve rather than a position the firm holds.

The bank service provider question resolved elsewhere in this sweep was considered and is unsettled here: the company licenses software and also delivers software as a service, which places it on the boundary between licensor and processor, and it would need testing against specific service arrangements rather than assumed either way.

Artificial intelligence ethics is named as one of five standards the responsible by design approach is said to ensure, and that is the whole of it. No fairness testing, no protected characteristic treatment, no disparate impact analysis, no named external framework, no results and no position on any artificial intelligence regulation.

The contrast that makes this a C rather than a generous B sits on the same roster and in the same national market: TCS ships bias testing as a named component of its platform and takes B for it, while this vendor publishes the word ethics inside a list of adjectives. One ships a mechanism, the other publishes a value.

The exposure grows with the model platform, since it invites business users to build solutions on consolidated customer data with no fairness step described anywhere in the lifecycle it does document.

No recourse position published. The shipped assistants carry modest direct exposure, since retrieving a document or drafting a ticket reply does not decide anything about a customer, and that is recorded as mitigation rather than disclosure.

The real division sits under the model platform and it is the same three way split this sweep has now seen in four forms: when a bank's own business user builds a model using the vendor's pre trained components, on the vendor's data platform, and it produces a wrong outcome for a customer, responsibility divides between the pre trained component, the platform that made building it easy, and the institution that deployed it. Nothing addresses any part of that.

Two clouds named and no model, which is the clearest illustration of the standing rule anywhere in this sweep. One hyperscale provider is named as a general cloud partner and another is named as the platform the artificial intelligence suite runs on, and neither identifies a model, a provider of models or a version.

The suite is also stated to include a wide range of pre trained models, and none of them is named, described or attributed, which is a substantial omission for components a bank is invited to build production solutions on top of. The reference to the cloud provider is ambiguous as to whether it extends to that provider's managed model service, and an ambiguous claim earns nothing, so the check is queued rather than the benefit given.

Graded A on the basis that carried the other core suppliers here, with one piece of genuine architectural specificity that lifts it above a generic claim: the data layer is built on data models inspired by the Banking Industry Architecture Network reference architecture, which is a named industry standard for banking service definitions rather than a vendor's own abstraction, and it is paired with domain specific data marts.

Naming a reference architecture is the same class of evidence as naming a messaging standard, and it tells an integrator something actionable. The vendor supplies the core itself alongside origination, payments, cash management and treasury, deployment spans private, public and hybrid cloud and software as a service, and two named cloud partners appear in the published material.

Deployment shape is well answered and residency is not addressed at all. Four delivery models are named as choices, private cloud, public cloud, hybrid and software as a service, and two hyperscale providers appear in published material, one as a general cloud partner and one as the platform underneath the artificial intelligence suite. That is more than a generic cloud native claim.

Against it, nothing states region availability, data residency commitments or how customer data is kept inside a jurisdiction, which matters more here than for most because the customer base spans more than a hundred countries including markets with explicit localisation requirements. Graded C on the same basis as the other roster vendors that publish deployment choice without residency.

No pricing published for the core suite, the lending and origination products, the cloud and software as a service offerings or the data and artificial intelligence suite. No tiers, no bands, no metering basis. The gap is worth noting specifically for the model platform, since a platform sold on letting business users build their own solutions has an obvious consumption dimension and nothing indicates how it is charged.

Banks in more than one hundred countries, stated as serving over a billion people and millions of businesses, which is among the largest reach claims in the index. The named client set demonstrates the range rather than asserting it and spans unusually different institution types: a Gulf digital bank built around digital assets, a Colombian digital only bank, a digital brand launched by an Indian commercial bank for a young demographic, a Thai social bank targeting unbanked customers, a capital markets business building corporate cash management, a large European bank modernising liquidity management, an Australian commercial bank migrating its core, and a residential mortgage lender in Australasia. Retail, corporate, small business, wealth, capital markets and financial inclusion are all present.