How Riskified connects risk across the purchase journey
On 3 November 2026, Riskified announced an expanded version of its Risk Intelligence Platform. The company is bringing events from account registration and login through checkout, promotion use, returns, payment disputes and customer-service interactions into one risk layer. [1 · Riskified · Risk Intelligence Platform expansion · 3 November 2026] [2 · Riskified · unified platform design · 3 November 2026]
An identity-clustering engine sits at the centre. Riskified says probabilistic machine-learning models link fragmented accounts and individual events to a presumed identity, while behavioural models add the history of its activity across the company’s global network. [1 · Riskified · Risk Intelligence Platform expansion · 3 November 2026] [2 · Riskified · unified platform design · 3 November 2026]
The platform is intended to feed that assessment into five areas: account protection, order approval, policy and returns abuse, payment disputes and risk checks during customer-service conversations. Merchants can place their own rules on top of Riskified’s assessment. [1 · Riskified · Risk Intelligence Platform expansion · 3 November 2026] [2 · Riskified · unified platform design · 3 November 2026]
Riskified says $1.5 trillion in transactions has been received across its network since 2013. This is a vendor-stated data volume, not a published validation of model quality: the announcement does not show signal coverage, identity-linking accuracy or differences across countries and retail categories. [1 · Riskified · Risk Intelligence Platform expansion · 3 November 2026] [2 · Riskified · unified platform design · 3 November 2026]
What changed and what remains a promise
Riskified’s official explanation explicitly calls the platform a combination of existing solutions and newer capabilities. The event is therefore more accurately described as an expanded common risk-management layer than as a wholly new product whose full set of functions is available to every customer at once. [2 · Riskified · unified platform design · 3 November 2026]
One newer element, Agent Identity Risk Intelligence for shopping and service requests through AI assistants, was introduced on 8 October. At that time, Riskified scheduled a limited beta for enterprise merchants for December 2026 and placed general availability later. [3 · Riskified · Agent Identity Risk Intelligence · 8 October 2026]
The 3 November announcement gives no pricing, market list, migration timetable or independent benchmark. It does not disclose correctly linked account rates, mistaken merges of different people, false declines of legitimate orders, manual-review time or the unified platform’s effect on fraud losses. [1 · Riskified · Risk Intelligence Platform expansion · 3 November 2026] [2 · Riskified · unified platform design · 3 November 2026]
Before deployment, merchants should record a baseline for each stage: approval of legitimate orders, fraud losses, account takeovers, returns and promotion abuse, dispute recovery, manual reviews and response latency. Only that comparison can show whether the common layer reduces total risk cost or merely moves errors between teams. [1 · Riskified · Risk Intelligence Platform expansion · 3 November 2026] [2 · Riskified · unified platform design · 3 November 2026]
Sources
- Riskified · Risk Intelligence Platform expansion · 3 November 2026 — Primary announcement of the expanded platform, its signal-linking mechanism and functions for accounts, checkout, policies, disputes and customer service.
- Riskified · unified platform design · 3 November 2026 — Official explanation that the platform combines existing products with newer capabilities; it describes probabilistic identity linking and tools for managing rules.
- Riskified · Agent Identity Risk Intelligence · 8 October 2026 — Primary source on the agent-focused component: a limited beta for enterprise merchants is planned for December 2026, with general availability expected later.
Expert commentary
The confirmed fact is that Riskified has combined data and decisions from several stages of the customer journey in an expanded platform. The logic is sound: an account takeover, suspicious payment and post-delivery dispute may be episodes in one scheme. A shared screen and graph, however, do not prove that a merchant will catch more fraud or block fewer legitimate customers. [1 · Riskified · Risk Intelligence Platform expansion · 3 November 2026] [2 · Riskified · unified platform design · 3 November 2026]
Probabilistic linking does not “know” a person with absolute certainty. The model compares signals and assigns the most likely connection. That can expose several accounts run by one fraudster, but it creates the reverse risk: family members, colleagues or users of a shared device may be merged. The cost of an error rises when one conclusion affects checkout, returns and support. [1 · Riskified · Risk Intelligence Platform expansion · 3 November 2026] [2 · Riskified · unified platform design · 3 November 2026]
For the industry, the change is a move from separate fraud filters to risk management across the customer journey. Competitors will be judged on how consistently they carry context between teams and how well decisions can be audited. Merchants need more than a score: they need to change a rule, test it in shadow mode and see which event contributed to a decline. [1 · Riskified · Risk Intelligence Platform expansion · 3 November 2026] [2 · Riskified · unified platform design · 3 November 2026]
For customers, a shared risk layer could reduce repeated checks: a trusted shopper may clear checkout faster and receive a refund sooner. The opposite scenario is also possible, with an incorrect label following the person from login to dispute. A clear review process, response deadlines and a rule against treating every anomaly as proven fraud are therefore essential. [1 · Riskified · Risk Intelligence Platform expansion · 3 November 2026] [2 · Riskified · unified platform design · 3 November 2026]
Agentic commerce sharpens the issue because purchases made through an AI assistant may lack familiar browser signals. Riskified proposes linking the agent’s action to a person and their history, but the relevant component is still approaching a limited beta. Until results emerge, claims for the wider platform should not be projected onto live agent-originated orders and returns. [1 · Riskified · Risk Intelligence Platform expansion · 3 November 2026] [3 · Riskified · Agent Identity Risk Intelligence · 8 October 2026]
A sensible evaluation requires several quarters and a control group of stores or comparable order flows that retain the previous rules. The main measures are legitimate-order approval, fraud loss, false declines, manual reviews, return-resolution time, disputes won and blocks successfully overturned by customers. Improving one metric while another deteriorates would show a redistribution of risk rather than a net gain. [1 · Riskified · Risk Intelligence Platform expansion · 3 November 2026] [2 · Riskified · unified platform design · 3 November 2026]