INSTANDA

The company's own framing is precise and settles this grade: the platform is described as artificial intelligence augmented rather than artificial intelligence native, and augmentation is what the removal test finds. Strip the models out and the franchise survives intact, because the franchise is no code product configuration, letting business teams define product structures, rates, coverages and customer journeys without engineering involvement, and that capability predates the current wave by roughly a decade.

The two learned products launched in 2026 sit on top of it, one for large scale commercial underwriting with item level categorisation, one for complex cases requiring human judgement, plus query and wording assistants. Both are recent, both are additive to a working platform, and neither is load bearing for the core proposition. Marketing language about embedded native artificial intelligence with no bolt ons describes integration quality rather than dependence.

This vendor scopes its artificial intelligence toward the underwriter rather than around them, and it does so explicitly enough to grade above the lane norm. The 2026 operations platform is described as addressing complex cases that require human judgement rather than full automation, positioned to complement the administration system by taking on work that cannot easily be automated, which is a vendor declining an autonomy claim its competitors are making loudly.

The large scale commercial product is described as underwriter first and as supporting human decision making, with the assistants scoped to querying policies and drafting clause and endorsement wording rather than to deciding. Two reservations keep it below the top band. No named autonomy modes, referral triggers or approval requirements are published, so the constraint is a product philosophy rather than an enforced control a buyer can verify. And item level categorisation across tens of thousands of assets in real time is high volume automated classification that shapes pricing, which the human judgement framing does not cover.

Auditability is the published property and validation is the missing one, and they answer different questions. A record of what the system did supports supervisory review and incident reconstruction. Model risk management asks how the capability was validated before it operated and how degradation is caught afterwards, and two passes located none of it: no accuracy or error rate for item level categorisation, no benchmark, no validation methodology, no drift monitoring, no revalidation cadence, and no model documentation a carrier could place in its own inventory.

The categorisation claim is the one most in need of a number, since insurance grade categorisation at individual item level across tens of thousands of assets is precisely the sort of high volume classification where a small error rate compounds into material mispricing, and the company publishes the capability without publishing how well it performs.

Evidence is broad in claim and thin in attribution. The stated position is more than 80 clients whose operations reach more than 150 countries, and that phrasing needs care from a buyer, because it describes the territories client operations touch rather than the number of markets in which the vendor has deployed, and those are materially different facts. Named customers are scarce, with Great Plains Casualty the principal example located in two passes.

The strongest published outcomes are anonymised, describing a North American technology enabled insurer launching more than 50 specialty products across multiple jurisdictions in two years, and a global reinsurance group launching 30 products with new minimum viable products in eight weeks.

The second carries a claim of accelerating product development one hundred fold, which is an extraordinary figure published without a baseline, a definition of what was measured or a named attesting party. Partnership announcements with a workflow automation vendor and a claims specialist are recent and carry no deployment evidence yet.

Claims exist and evidence for them does not. The company states its artificial intelligence involves no black boxes and no bolt ons, and third party coverage of the 2026 underwriting platform reports explainable and auditable governance intended to serve customers, underwriters, auditors and regulators. Two passes located no vendor published document behind either statement, and no artificial intelligence policy, data handling statement or model documentation page.

The unanswered stewardship questions are the standard ones and none of them is addressed: whether client or policyholder data is used to train or improve any model, whether a client's data informs capability sold to another client, what happens to that data on termination, and where inference occurs relative to the client's chosen tenancy.

That last one matters more here than elsewhere, because a buyer who has paid for a dedicated single tenant stack would reasonably expect model processing to respect the same boundary, and nothing published says whether it does.

Two passes located no privacy policy content covering carrier or policyholder data, no data processing agreement, no retention schedule and no subprocessor list. A trust portal exists on a third party trust platform, which is a better arrangement than nothing and is gated behind an interactive application, so its contents could not be verified and a buyer will only see them after engaging.

One architectural control is published and does bear directly on this axis: tenancy is a buyer choice, with a dedicated single tenant stack available on request, which lets a carrier whose own policy or regulator requires physical separation obtain it rather than argue about logical isolation.

Certification to the international information security management standard is held, and that governs how information is managed rather than what may be done with it, so it does not substitute for the privacy commitments that remain unpublished.

Security is documented on a dedicated published page rather than promised on request, and the headline credential is current in the way that matters. Certification is held to the 2022 revision of the international information security management standard, stated as covering global operations, and citing the current revision rather than the superseded one is a detail worth crediting because many vendors still advertise the older version.

Supporting practices published include a secure software development lifecycle, continuous protection, and a management system aligned to the same standard. A trust portal is maintained on a third party trust platform, which indicates document provision is at least organised. Three qualifications hold it below the top band. An additional security certification is referenced without being named. No service organisation control report of either kind was identified in two passes. And the trust portal is gated behind an interactive application, so nothing in it could be verified independently.

The company holds no insurance licence and does not claim one, the ordinary position for a platform supplier. Its published regulatory capability is configuration rather than compliance service, with the platform stated to scale across channels, regions and states using configurable components while complying with local regulation, and one customer describing regional compliance customisation as a benefit of the no code approach.

That places responsibility with the carrier configuring the product, which is the honest allocation and also means the vendor evidences nothing about the outcome. Two passes located no rate, rule and form maintenance capability, no filing support, and no published positioning against the European artificial intelligence regulation, which is a notable omission for a vendor headquartered in the United Kingdom selling into European markets across life and health, where risk assessment and pricing sit in that regulation's high risk schedule.

Explainability is claimed prominently and fairness is not addressed at all, which is the pattern across this tier and slightly sharper here because of the vocabulary used. Published material promises no black boxes, and third party coverage reports explainable and auditable governance serving auditors and regulators, so the transparency limb is at least asserted.

Two passes then located no bias testing description, no fairness metrics, no disparate impact analysis, no model card, no responsible artificial intelligence statement and no governance documentation. The exposure is concentrated in life and health, a line this vendor supports and one where underwriting decisions carry the heaviest regulatory weight in both European and United States frameworks.

A vendor building products for that line while publishing nothing on fairness testing leaves its carrier customers to construct that evidence themselves, at the point where their own regulator will ask for it.

Recourse is unaddressed in both directions, and the human judgement framing does not close the gap as cleanly as it first appears. The complex case platform is positioned around work requiring human judgement, which locates decisions with underwriters for that product, and it is a product philosophy rather than a platform wide guarantee, with no published statement that any output requires approval before it takes effect.

The large scale commercial capability sits outside that framing entirely, categorising tens of thousands of assets at item level in real time, and an applicant priced on a miscategorised asset experiences an outcome no person examined. That applicant has no relationship with this vendor, no notice that automated categorisation shaped the price, and no published route to obtain or contest it.

Between vendor and carrier nothing published allocates liability for a categorisation error, a wording assistant producing a defective endorsement clause, or a query assistant returning an incorrect policy position.

Infrastructure provenance is published and model provenance is not. The platform's hosting is named openly at a major cloud provider's platform services layer, which tells a buyer where the software runs and implies without stating that model services may come from the same provider's catalogue. The models themselves are undocumented.

Two passes located no provider, model family, version, hosting arrangement or fine tuning approach for the query assistant, the wording assistant or the item level categorisation capability. The company's own language cuts against clarity here rather than toward it, since describing the capability as embedded native artificial intelligence with no bolt ons implies models built in house without ever saying so, and that phrasing would read identically whether the underlying capability were proprietary or licensed. A buyer cannot assess concentration risk, notification on version change, or where inference occurs relative to its chosen tenancy.

The estate covers policy administration and digital distribution as a single proposition, which is a wider distribution surface than most cores in this lane offer, spanning quoting, policy management, renewals, quote to bind, self service and channel connectivity for carriers, managing general agents and brokers alike.

Configuration reaches deep enough to define product structures, coverages, rates and fees without code, and the architecture is described as channel centric and interface first for connecting existing components. Integration is extended through partnership rather than built, with a workflow automation vendor covering service and claims workflow and a claims specialist partnership covering claims management, and that is the structural qualification on this grade: claims is reached through partners rather than owned, so a carrier consolidating the full lifecycle onto one vendor cannot do it here. Reinsurance, distribution management as a discipline and specialty operations are likewise absent from the published estate.

Two published facts lift this above the lane default. The infrastructure is named openly, with the platform architected and deployed on a major cloud provider's platform services layer and the company describing itself as fully cloud native, which is a disclosure many peers withhold. More consequentially, tenancy is a buyer choice stated in public: multi tenant by default, with a dedicated stack created for a customer who requests single tenancy.

Publishing that option rather than negotiating it privately lets a carrier with a separation requirement establish feasibility before entering a sales process, and few vendors in this lane do so. What remains unpublished is the residency layer above it.

Two passes located no region list, no statement of which geographies the platform runs in, and no data residency commitment a carrier could contract for, which matters given operations described as reaching more than 150 countries and a client base spanning five continents.

Pricing is absent from every vendor surface, consistent with this lane, and the commercial argument rests on speed and on removing engineering dependency rather than on stated cost. Published claims include product launches in weeks rather than months, business teams owning the adjustment lifecycle without developer involvement, and scaling across channels, regions and states through configurable components.

Nothing published indicates the charging unit, whether it follows premium, policies, products configured, transactions or subscription, and nothing indicates whether the two artificial intelligence products released in 2026 are included with the platform or licensed separately, which is the live question given both launched as distinct named offerings rather than as platform features.

Tenancy is the one commercially relevant variable the company does publish, since a dedicated single tenant stack is available on request and a dedicated stack normally carries different economics from shared infrastructure, with nothing stating whether it does here.

Breadth is the genuine strength of this record on both dimensions that matter. Lines of business span property and casualty including complex commercial, life and health, and group protection, with the platform described as product agnostic rather than configured per line, and a specific capability for commercial and non admitted business where high asset volumes have historically limited underwriting.

Buyer types run wider than most of this lane, covering carriers, managing general agents and brokers rather than carriers alone, which matters because those parties carry different obligations and the software has to satisfy each. Geographic reach is claimed globally, with configurable components for complying with local regulation across regions and states, and the company operates across five continents.

What keeps it below the top band is depth of evidence behind that reach, since more than 80 clients is a modest base from which to claim coverage of this many segments, and the country figure describes where client operations extend rather than where deployments sit.