TCS BaNCS

The Clearwater precedent, and the vendor's own naming makes it easy: the artificial intelligence line is a separately branded companion to the core and securities platforms rather than the platforms themselves, described as pairing the existing estate with a set of models, generative capability and pre built agents.

Strip every model and core banking, lending, payments, custody, corporate actions and market infrastructure connectivity all continue to run for the reported three hundred and seventy institutions, because that is what the platform has done for decades. The learned line is real, named, dated and attached to named pilots, which is what carries the build.

Governance is claimed and no gate is described. Audit logging appears as a named platform component and it is retrospective traceability rather than a control on action: it records what happened, it does not decide whether an action may proceed. Nothing published states a threshold, a confidence band, a default configuration, a review requirement or what a person must approve.

One oversight question is genuinely new here and is recorded because no other vendor on this roster raises it: the platform ships a no code agent design canvas explicitly so that institutions can build their own agents, and it is presented with frontline staff as intended users. That distributes agent creation into the business, and nothing published describes who reviews an agent before it reaches a customer, what testing it must pass, or who owns its behaviour once built. Delegating construction without describing approval is a distinct gap from delegating a decision.

Graded B on the strength of a measured figure with a stated denominator, which is rarer in this index than any amount of methodology language: securities event classification accuracy of 99.2 percent across twenty five million records. A denominator is what makes an accuracy claim readable, and most vendors here publish neither. Audit logging is named as a platform component, which supports traceability of model output after the fact. Held at B and not A on several counts.

The figure comes from controlled trials rather than production, no validation methodology, backtesting, drift monitoring or versioning is published, and no external assessment of any model appears. The company publishes substantial practitioner material on model risk management platforms naming the United States supervisory guidance and the United Kingdom prudential regulator's model risk principles, and that is category content addressed to the buyer's obligations rather than a statement about its own models, so it earns nothing here. Recorded because it is a direct contrast with Abrigo on this roster, which sells to institutions examined against that same guidance and never names it.

This vendor has more named institutions than almost anything on this roster and more published numbers than almost anything on this roster, and the two sets never meet. Named and attributed: a chief technology officer at a United States regional banking group quoted by name on the agent canvas, custody modernisation at a named European clearing bank with its leadership on record, a named Gulf digital bank launch, a named global custodian piloting corporate action agents, and an expanded alliance with a named securities services business across more than one hundred markets.

Published numbers: securities event classification accuracy of 99.2 percent across twenty five million records, onboarding reduced from forty eight hours to fifteen minutes, and a projected multi million annual saving on query resolution. Held at B because every number carries either an anonymous institution or a forward looking verb.

The one figure attached to a named institution is what that institution expects rather than what it has measured, and the strongest measured figure belongs to a controlled trial with no institution named at all.

Silent on the substance. Responsible practice is claimed as the purpose of the artificial intelligence platform and nothing behind it addresses data. No statement of whether customer data trains or tunes any model, whether anything is pooled across the institutions on the platform, what an agent retains, or what is excluded from a training corpus.

The question has particular weight for a platform holding custody positions and corporate action data for competing custodians on shared infrastructure, and for pre built agents that were trained on something the vendor does not identify.

Privacy by design is named as a component of the artificial intelligence platform and that is a principle rather than a position. No retention schedule, no deletion terms, no subprocessor disclosure and no tenant separation statement for a hosted platform serving competing institutions. The data residency support built for one jurisdiction is a genuine commitment and is credited on the deployment axis rather than double counted here. Published material on protecting personally identifiable information in back office operations is category commentary about the buyer's problem rather than disclosure about the vendor's own handling.

No certification, attestation report, audit period or trust portal was located from the platform material, so the grade follows the standing rule that a credential must be found and named rather than assumed from the size of the parent company.

This is unproven absence rather than evidenced absence and the queued check is cheap: a group of this scale operating hosted services for custodians and clearing banks certainly holds attestations, and the reason none surfaced is that they sit at corporate level rather than on the product pages. Recorded as a finding in its own right that a buyer researching a platform that holds custody records does not meet a security credential anywhere in the product journey.

No licence, registration or supervised standing held by the firm appears in the material reviewed. The bank service provider question resolved elsewhere in this sweep was tested against this vendor and is not settled either way: the statutory examination authority attaches to performing services for a depository institution rather than to licensing software, and this company both licenses platforms and runs hosted and managed operations for institutions, which places it on the boundary rather than clearly on one side. Recorded as an open question rather than a silent refusal, and it would need testing against the specific service arrangements rather than the company as a whole.

The first grade above C on this axis anywhere on this roster, and it is a thin B held deliberately at the floor of the band. Bias testing is reported as a named component of the artificial intelligence platform alongside audit logging and privacy by design, which is a described mechanism rather than an aspiration, and this index has consistently graded a named mechanism above silence.

Everything that would make it substantial is missing: no fairness metric, no protected characteristic named, no test methodology, no results, no threshold for what a failed test would prevent, and no named external framework. The sourcing is also one step removed, appearing in dated launch coverage attributed to the company rather than in a product specification.

Queued and specific: the platform's own product page was not opened and is where this would either firm up toward A or fall back. Worth stating plainly that a bias testing component shipped inside a banking platform is a meaningful market signal even at this thinness, because the rest of this roster publishes nothing at all.

No recourse position published. Two exposures sit under the silence and the second is unusual. The conventional one: agents touching onboarding, query resolution and underwriting signals reach customers, and nothing states whether a customer learns a model was involved or how a wrong outcome is contested.

The unusual one follows from the no code canvas: when an institution builds its own agent on the vendor's platform using the vendor's models and design surface, responsibility for that agent's behaviour divides three ways between the model supplier, the platform that provided the building blocks, and the institution whose staff assembled it. That is the agent marketplace liability gap recorded against Send and Eltropy, arriving in a new form, and it is answered here no better than there.

No model, provider or version is named for any part of the platform. The capability is described only in categories, machine learning and deep learning alongside generative capability and pre built agents, which identifies the technique and not the supplier. The omission is sharper for a platform selling a no code canvas than for most, because an institution assembling its own agents is taking on a dependency it cannot assess without knowing what sits underneath the canvas. The parent group publishes named partnerships with several large technology suppliers in other industries, and none is identified as underpinning this platform.

The vendor supplies the system of record on both sides of its estate, the core banking platform and the securities and custody platform, with an enterprise payment hub between them, which is the basis that carried Azentio, SBS, Kiya.ai, Finastra and Oracle.

What lifts it clearly rather than marginally is the market infrastructure layer: connectivity is described in named messaging standards, ISO 15022 and ISO 20022, and against a named national depository rather than as generic interoperability. Corporate action processing across securities feeds is itself an integration discipline, since the data arrives from custodians, depositories and market feeds in inconsistent form. A no code design surface sits over the top for assembling agents against that estate.

The best residency disclosure found on this roster, and it is concrete rather than adjectival. A custody and cash processing offering for Saudi Arabia is sold as a jurisdiction specific service that supports data residency, connects to the local depository over named messaging standards, and covers the local Sunday working week.

That last detail is the tell that this is a real localisation rather than a marketing claim, because accommodating a different working week is an operational commitment nobody writes unless they have made it. Software as a service is offered across the wider stack.

Held at B and not A because one jurisdiction is described in this detail and nothing comparable is published for the rest of a global base of roughly three hundred and seventy institutions, so a buyer elsewhere learns that the vendor can do residency and not what it will do for them.

No pricing published. The delivery model is stated as software as a service across the stack, which tells a buyer the shape of the commercial relationship and nothing about its cost, and no tier, band, module structure or metering basis appears anywhere.

Agent pricing is a live unanswered question specifically, since the platform is sold on institutions building their own agents through a no code canvas and nothing indicates whether that is licensed by agent, by user, by consumption or with the platform.

Around three hundred and seventy institutions globally, and the buyer range is unusually wide even by the standards of this tier because it spans two different industries rather than one. On the banking side, retail banks, regional lenders, community banks, credit unions and co-operatives, plus a greenfield digital bank in the Gulf. On the securities side, global custodians, clearing banks and a securities services business operating across more than one hundred markets.

Insurance sits alongside both. Coverage of custody and corporate actions is not an adjacent market to core banking, it is a separate operational discipline with separate buyers, and holding both is what puts this at A rather than B.