IntellectAI

The strongest C in this pocket and the reasoning matters more than the grade. The model layer is named as a product in its own right rather than described as a feature, embedded models are one of the six stated properties of the underlying architecture, and the platform is positioned as intelligence led rather than as a platform with intelligence added.

Against that, the removal test still resolves the same way as it does for Backbase: strip the models and a full lifecycle wealth platform remains, covering onboarding, portfolio construction, execution, custody connectivity, post trade servicing and reporting, and the parent has been shipping banking software for three decades. Graded consistently with Backbase rather than above it.

The second A on this axis in the index and it closes the gap Backbase left open. The company states that every recommendation carries explainability showing its reasoning, its data sources and a confidence level, with the models augmenting rather than replacing human judgement and advisers retaining control of client interactions.

A published per recommendation confidence level is the specific thing this index recorded as missing from Ruleguard, Temenos and Backbase, all of which described governance without ever surfacing how certain the system was. Backbase earned its A by naming the enforcement mechanism and where it sits in the execution path; this one earns it by surfacing provenance and certainty at the point of every decision. What is still absent is recorded: no sampling audit of automated outputs, no stated action threshold, and no description of what happens when confidence is low.

The first B on this axis anywhere in the incumbent wealth and banking pocket, and it turns on one property. Publishing a confidence level alongside every recommendation, together with its reasoning and data sources, gives an institution's validators a live behavioural signal across the whole output stream rather than a single explicable case, which is the property that earned Sardine a B for exposing how its controls actually behave. Off an A by a wide margin.

No accuracy, precision or recall figure, benchmark, validation package or independent measurement was located, and nothing describes how the confidence score is calibrated against realised outcomes. An uncalibrated confidence score is a number the model emits, not a measurement of whether it is right, and the distinction should be put to the vendor directly.

A named customer executive quoted by name and title, the head of wealth product and private banking at a Thai bank, plus a second named institutional voice, which clears the bar. Independent weight comes from inclusion as a notable vendor among 23 in a Q2 2026 analyst landscape of digital wealth management platforms, cited specifically for agentic automation, compliance monitoring and portfolio analytics.

Off an A because no quantified outcome attaches to any named client: no adoption figure, adviser productivity measure, or effect on revenue or cost. Industry awards for use of artificial intelligence were not credited, on the same basis as the submission based recognitions declined elsewhere in this index.

The most informative C on this axis in the pocket, and it lands exactly where Red Oak did. The company describes a three layer knowledge structure that separates public market data, the institution's own product and compliance rules, and private client information, and names multi tenancy as a platform property. That is a real architectural account of where client data sits relative to the models, and more than any peer here offers.

It stays at C because it is presented as capability architecture rather than as a boundary commitment: nothing states that private client information from one institution never informs an output served to another, nothing describes retention, and nothing says whether models are tuned per tenant. Restated as a commitment it would be a B, and the architecture appears to support one.

No privacy programme, retention position, data processing terms or subject rights framework was located. The knowledge structure explicitly designates private client information as an input layer to the models, which makes the absence more pointed than usual: the company has described where identified individuals' financial data sits in its architecture without describing how long it stays, who may query it, or what the individual can see or correct. Deployments across Asia, the Middle East and Europe span materially different data protection regimes and none is addressed in public material.

No certification, attestation, report type, audit scope, penetration testing summary or trust centre was located for the platform or the model layer. A vendor selling full lifecycle wealth infrastructure to banks across three continents will hold formal credentials and produce them in procurement, and none are published where a buyer evaluating the platform can find them. Temenos and Avaloq, the two closest peers by shape, both publish theirs.

A software vendor holding no financial licence of its own, with no supervisory programme, sandbox admission or regulator run assessment of the model layer located. Its clients carry the authorisations. The parent is a listed company subject to securities disclosure obligations in its home market, which is a reporting regime rather than a financial services licence and says nothing about the products.

Explainability is genuinely disclosed and fairness is not, which is the same split recorded against Backbase and is worth stating plainly because the two are routinely conflated. Showing the reasoning and sources behind a recommendation tells a user how an output was produced; it does not establish that outputs are even handed across client segments, wealth bands or demographics.

No bias testing, fairness evaluation, model documentation or independent assessment or certification of an artificial intelligence management system was located, and a platform generating portfolio restructuring recommendations and contextual nudges across retail and private banking is making distinctions between clients by design.

Nothing published describes liability, indemnity or recourse when a generated recommendation is wrong. The explainability disclosure creates an asymmetry worth naming: the adviser sees the reasoning, the sources and the confidence level, and the client who receives the resulting advice sees none of it and is not told a model produced the restructuring recommendation. Transparency delivered to the intermediary is not recourse delivered to the affected party, and nothing describes a contest route for the latter.

The model layer is named as a distinct platform, which is a step beyond describing it as proprietary, and nothing is disclosed about what runs inside it. No base model, provider, version or hosting arrangement is named, and no statement was located on whether third party foundation models sit underneath. An institution recording model versions against advisory decisions, which is standard practice under model risk oversight, cannot do so from public information.

Full lifecycle coverage in one platform, from client acquisition and onboarding through portfolio construction, advisory, execution, custody connectivity and post trade servicing, on an architecture explicitly built for integration: event driven, microservices based, interface led, cloud native and headless, with open ecosystem connectivity to institutions and marketplaces named as a capability. Multi tenancy, multi currency, multi asset and multilingual operation are stated properties. Headless architecture in particular means the platform is designed to be consumed by a client's own front end rather than imposing one.

Cloud native delivery with multi region and multi tenant operation named as platform properties rather than left to inference, which is more than most of this pocket offers and matters for institutions in Asia and the Middle East operating under local hosting expectations.

Off an A because no residency regions are enumerated, no residency commitment is published, no on premises path is described, no subprocessor list was located, and nothing states where the model layer executes relative to the client data it reads.

No pricing, band, module rate or minimum is published. The platform is composable and sold by configuration across a wide feature surface, which makes the units of purchase and how they combine into a deployment cost impossible to estimate from public material. Pricing follows a sales engagement.

Segments and geographies are named specifically, covering relationship managers at retail and private banks, trust banking and wealth managers, with multi currency, multi asset, multi region, multi tenant and multilingual operation stated as platform properties rather than aspirations, and deployments across Asia and the Middle East. The parent reports more than 325 customers worldwide.

Held at B rather than A on the PerformLine and Saifr discriminator: none of the breadth is anchored for this platform specifically. No client count, no assets on platform and no adviser count was located for the wealth product itself, and the parent figure spans four business lines including insurance and transaction banking.