bondIT

The machine learning is the product, not a layer on top of one. FRONTIER is an optimisation engine that constructs and rebalances fixed income portfolios against client constraints in minutes rather than days, and SCORABLE is a predictive rating transition model analysing more than 250 variables daily across 3,000 plus rated issuers.

Strip the models and nothing purchasable remains: there is no underlying data product, order management system or research library that would stand on its own. Consistent with the A grades held by capital markets modelling peers Axyon AI, Boosted AI, MDOTM and Theia Insights.

Recommendation and construction are automated, the decision is not. The platform generates optimised portfolios and rebalancing proposals in seconds and Scorable flags probable rating transitions, but the portfolio manager or advisor selects, approves and executes, and click to trade is presented as a human action. The explainable AI framing is functionally part of the oversight design, since the stated purpose is to let a user understand the rationale before acting on it.

What is not described is the boundary at scale: mass customisation across tens of thousands of accounts implies some proposals are applied with limited individual review, and no threshold, exception route or sign off model is published for that mode.

A predictive product carries an obligation to publish a hit rate, and this one does not, which is the same finding recorded against Blue Fire AI. Scorable states a twelve month horizon for upgrade and downgrade probability across 3,000 plus issuers, which makes accuracy directly and cheaply measurable against realised rating actions, yet no precision, recall, calibration curve, false alarm rate, lead time distribution or backtest appears.

Explainable AI is offered in place of measurement and the two are not substitutes: knowing which variables drove a score does not establish that the score is right. The portfolio optimiser carries the separate risk that optimising against historical data selects for what fitted the past, and no out of sample or realised versus modelled study is published.

The strongest evidence profile in this pull and it clears the A bar on independent parties with money at stake, not on vendor claims. BNY Mellon led the December 2022 Series C of roughly 14 million dollars and took a board seat. BNY Mellon Pershing then launched BondWise, a fixed income research and trading tool built in collaboration with bondIT and shipped on its NetX360 plus advisor platform, which is a custodian putting its own client-facing product on this engine.

Liquidnet integrated Scorable Credit Analytics into its fixed income electronic trading platform in December 2023, reaching a stated 700 plus fixed income member firms, with Liquidnet's global head of fixed income product quoted by name. A further named partnership with First Rate targets US wealth management. What is still missing is a quantified performance outcome: no client reports a measured improvement in returns, risk or advisor productivity in figures.

The unanswered question is competitive separation, and it is sharper here than for most vendors because of the distribution model. Asset managers, private banks and broker dealers using this platform compete with each other directly, and the same engine also sits inside an execution venue where 700 plus member firms trade against one another.

Nothing published states whether portfolio positions, universes, constraints or trading intentions from one client inform models, defaults or analytics served to another, nor whether internal models a client onboards remain segregated. Compare Aiera, which earned a B here for publishing entitlement aware access and consumption metrics designed to preserve content value.

No published privacy posture. The exposure is narrower than for consumer facing vendors because the platform reasons over securities and portfolios rather than individuals, but it is not absent: portfolios are built to individual client requirements, investment parameters, goals and constraints, and mass customisation across tens of thousands of accounts means underlying client profile data flows through the system. Nothing states what client identifying data is held, where, or for how long.

No trust centre, no named certification, no SOC 2 or ISO 27001 attestation found in public material. Notable given the customer set: a custodian and an execution venue both integrated this platform into client facing systems, which means each ran a third party risk assessment that is not reflected in anything the vendor publishes. Passing an institutional review privately is not the same as publishing evidence a prospective buyer can read.

No licence, regulated entity or supervisory relationship named anywhere, and the perimeter is real rather than theoretical. Generating specific investment recommendations delivered to end clients through advisors and private banks touches suitability and appropriateness duties under MiFID II in Europe and best interest obligations in the US, and the vendor positions itself explicitly on making recommendations consistent and client specific. The regulated party is the institution rather than the technology provider, which is the standard arrangement, but no public material engages with where the vendor's responsibility ends.

Explainable AI is marketed heavily and positioned against what the vendor calls obscure black box solutions, but explainability is a transparency property and not a fairness measurement, and no governance framework, model inventory, review board or bias testing is published.

The exposure specific to this product is issuer coverage rather than protected classes: Scorable covers rated issuers, so unrated, smaller, emerging market and non English reporting issuers are structurally under-served, and a credit signal available for 3,000 rated names and absent elsewhere tilts capital toward the already covered. Nothing describes coverage limits or how absence is presented to a user.

No warranty, service level, accuracy commitment or remedy published, and the path from a model error to a loss is short and direct. A rating transition signal that misses a downgrade leaves an institution holding a deteriorating bond, and a portfolio optimiser that misconstrues a constraint produces an allocation an advisor delivered to a client as a recommendation.

The harmed party in the second case is an end investor who never encountered the vendor, has no way to know an external engine shaped the recommendation and nothing to appeal to. There is a second party with no relationship to the vendor as well, in the pattern noted for Blue Fire AI: the issuer flagged with a high downgrade probability, which cannot see or contest the assessment while subscribers reprice its debt.

A short and mostly self owned chain, which is the substance of the B rather than a disclosure document. The models are classical machine learning and explainable AI built in house, with no foundation model provider at the centre and no dependency on a third party generative API, so the failure and pricing exposure that comes with that dependency does not apply.

Provenance for the credit analytics half is unusually traceable because it was bought rather than built: Scorable was acquired from a Berlin startup in November 2020 in a stock swap reported at 12 to 16 million dollars, so the rating transition models predate their current owner.

Underlying market and reference data sources are described only as a vast array including financial statements, fundamentals and capital market data, with no vendor named, and the data agnostic positioning means the input mix varies by client.

Distribution is the strategy, in the pattern Aiera earned an A for. Downstream connectivity into existing portfolio management and trading systems, API integration positioned for institutions of all sizes, onboarding of a client's own internal models into the platform, execution through electronic trading venues with click to trade, and two deep embeds where the engine reaches users through somebody else's system: Scorable inside Liquidnet's fixed income trading platform, and the BondWise tool on BNY Mellon Pershing's NetX360 plus. Being data agnostic and running a client's proprietary models rather than mandating its own is the integration decision that lets it sit inside incumbent stacks instead of competing with them.

Presented as a cloud SaaS platform reached by API, with no published deployment options, hosting regions, data residency commitments or single tenant alternative. This is a live question for the buyer set rather than a formality: private banks and custodians in Switzerland, Singapore and the Gulf routinely require in jurisdiction hosting for client portfolio data, and the vendor sells into all three regions.

No pricing, rate card, tier structure or minimum published. Third party profiles describe the model only as clients being charged for access to the platform's tools and capabilities. The unit of consumption is not disclosed either, which matters here because the platform is sold both direct to firms and embedded inside partner platforms such as Liquidnet and Pershing, and those two arrangements almost certainly price differently.

Genuinely broad across institution types rather than deep in one. Named buyer categories across its own material and trade coverage: institutional asset managers, wealth managers, private banks, banks and custodians, fixed income brokers and broker dealers, financial advisors, family offices and asset owners.

It states it is data agnostic and sized for financial institutions of all sizes via API integration, and the distribution through a custodian platform and an execution venue reaches firms it does not sell to directly. The boundary is functional rather than institutional: everything sits inside fixed income investment management, with no lending, insurance or payments surface.