Fiserv
Fiserv sells the software many United States banks run on, and it has built an AI line on top of that position. The estate spans account processing and digital banking, card issuer processing and network services, payments, electronic commerce, merchant acquiring and processing, and the Clover point of sale and business management platform. The banking segment alone reports $2.4 billion of revenue across more than 3,500 financial institutions, digital payments runs just under $4 billion and serves 41 of the 50 largest United States banks, and the financial solutions side reports more than 6,000 clients globally.
The AI line is agentOS, launched 14 May 2026 as an operating system for agentic AI in banking and stated as widely available from August 2026. It runs natively across the company's own core, payments, issuer processing and servicing platforms rather than beside them, which is the point of it: an institution already running Fiserv systems can deploy agents into those workflows without a separate integration. Six financial institutions co developed it and two were running agents in beta at launch. Strategic collaborations with OpenAI and Amazon Web Services are named openly.
Its distinguishing feature is a marketplace. Institutions can run agents Fiserv built, build their own, or deploy agents from third parties, all inside the same governance, identity and audit controls. Four first party agents launched, covering commercial loan onboarding, daily operational analysis and reporting, deposit intelligence and anti money laundering triage, alongside nine third party agents spanning risk management, regulatory reporting, deposit operations and back office reconciliation.
The company is candid about a difficult period. Executives have publicly attributed higher than desired core banking attrition to past service failures, missed product deadlines and forced conversions, with a co president stating the company has a service problem rather than a technology problem, and have committed to supporting all cores with no forced migrations. Fiserv is listed on Nasdaq and is a member of the S&P 500.
Capability Axes
Capability grades
15 of 15 axes rated · 6 graded A or B
Account processing, card issuing and merchant acquiring all predate the agents and run without them. The May 2026 agentic platform is described as extending existing technology investments, and its four first party agents work around the core, from loan onboarding to financial crime triage.
Human oversight and escalation paths govern every agent, whether first party, bank built or one of nine third party agents in the marketplace. What a third party agent must pass before listing, and who answers when it errs inside the guardrails, is unstated, though agents already triage money laundering cases and onboard commercial loans.
Pilots are said to deliver measurable gains today, and no accuracy, saving, baseline or observation period backs the claim. The AI figures the company does publish describe its engineers' adoption and the share of code written, not how agents perform in a bank.
Public filings verify the scale, with a banking segment reporting $2.4 billion of revenue and a digital payments business near 4 billion. None of the six co developing institutions behind the agents is named, though executives have candidly admitted that service failures drove core banking attrition.
Platform level kill switches let an institution halt agents, not just watch them, alongside deterministic guardrails, full action audit and real time anomaly detection. No model card, evaluation method or statement on what client data trains or contextualizes the agents is published.
Deposit, card and merchant records for a large share of the United States population sit in these systems, yet no privacy program, retention schedule or subprocessor list is published and the Gramm Leach Bliley Act (GLBA) is never cited. At the agent layer, role based permissions and infrastructure level data masking limit what an agent sees, which concerns agents rather than what the company holds.
Outside procurement, nobody can confirm a certification, attestation report or penetration test, since none is published. The agentic platform's security is described in real detail, from identity bound execution to traceability, so the company can publish specifics when it chooses.
As a listed registrant the company files quarterly reports with risk factors, which is corporate rather than financial services regulation. Its own examinations as a bank service provider, any EU AI Act position and fit with model risk expectations are all unaddressed.
Published governance answers who an agent is, what it may do and whether it can be stopped, not whether it treats people evenly. Money laundering triage has documented uneven effects across geography and name origin, and loan onboarding raises fair lending questions, yet no fairness testing covers any agent.
Responsibility for an agent error splits three ways nobody has settled: first party agents point to the vendor, bank built ones to the bank, and third party agents running on the vendor's controls to no stated party. Terms, warranty and service levels are unpublished, and nothing tells an affected customer how to contest an agent decision.
OpenAI and Amazon Web Services are named as strategic collaborators, letting a buyer assess the provider's own terms directly. Neither model family nor version is mapped to any agent, and the nine third party agents' model dependencies sit outside any published governance.
The agent platform runs natively across the company's own core, payments, issuer processing and servicing, so a client deploys agents into systems of record without new integration. Support for all cores and no forced migrations, adopted after forced conversions damaged client relationships, open it to institutions not on its core.
Hosting regions, residency pinning and tenancy separation between institutions on shared infrastructure are unpublished, though a hyperscaler collaboration signals public cloud. Cross platform agent actions touch systems of record for thousands of institutions, which makes placement a sharper question than for a standalone application.
The marketplace raises questions nothing answers: whether the agent platform carries a fee, whether agents bill per agent or per workflow, and who bills for third party agents. Quarterly filings reveal the size and profitability of the selling business unit, more financial disclosure than private vendors give, but no product price.
More than 6,000 institution clients include 41 of the top 50 United States banks and reach down through community banks and credit unions. Coverage spans account processing for the institution, issuer processing for cards and merchant acquiring, both sides of a transaction.
Alternatives to Fiserv
Vendors in the same categories as Fiserv whose profiles look most alike, each with a line on what the product does. Similar is not the same: check that each one does the job you need. No vendor pays for placement.
Jack Henry is the third of the three companies whose core systems run most United States banking, and it is the one aimed squarely at the small end of the market.
Azentio Software is a Singapore headquartered technology provider for banking, financial services and insurance, assembled from 2020 by Apax Partners through the combination of several…
Finastra is a London headquartered financial services software group selling across lending and corporate banking, payments and universal banking.
SAP Fioneer is a financial services software company headquartered in Walldorf, Germany, created in 2021 as a carve out from SAP, which remains a shareholder and strategic partner, with…
Zeta sells Tachyon, a cloud native single vendor stack bringing card issuance, processing, lending, core banking, fraud and risk, loyalty and digital banking together for banks and fintechs.
Bottomline is payments infrastructure with fraud analytics layered on top.
The index publishes no overall score or ranking. See what we assess and the methodology.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | Data Protection Terms | Implementation | Source |
|---|---|---|---|---|
|
Not published. No price, unit of billing, tier or contract term appears on any vendor surface for the agentic platform or the underlying processing lines
|
Not published on any vendor surface. The estate spans account processing, digital banking, card issuer processing, network services, payments, electronic commerce, merchant acquiring and a point of sale platform, and nothing published indicates the commercial basis for any of them or how the agentic platform is charged. The marketplace structure leaves the central question open: whether agentOS carries a platform fee, whether first party agents are licensed individually, by workflow or by volume, whether third party agents are billed by the platform or by their builders, and whether any revenue share applies between them. As a listed company Fiserv does publish segment level revenue and client counts under securities law, which gives a buyer scale and unit economics at the business level without giving a price. | No tiered data protection terms are published. Data handling commitments appear only at the agent layer, where identity bound access with role based permissions and data masking at the infrastructure level are named as platform controls governing what an agent may see. Nothing published describes the company's own data processing terms, retention, subprocessors or residency for the underlying core, payments, issuing or acquiring platforms. Independent commentary on the agentic platform advises buyers to verify client data ownership statements and masking rules for agent actions that cross platforms, which indicates those terms are negotiated rather than published. | No implementation, conversion or professional services fee is published. Implementation cost is nonetheless the subject this company has addressed most directly in public, for reasons that were not favorable to it: executives have acknowledged that forced core conversions and missed product deadlines drove client attrition, and the corrective commitments are stated as support for all cores rather than a favored one, no forced core migrations, and modular core agnostic capabilities clients adopt on their own timeline. Those commitments are about avoided cost and avoided disruption rather than about price, and they are concrete because they were made in response to a named failure. For the agentic platform specifically, the integration burden is designed to be near zero for existing clients, since it runs natively across core, payments, issuer processing and servicing, and the marketplace is presented as the route to deployment. Nothing published describes onboarding effort, professional services requirements or timelines for either path. | Vendor Published |
Neither the company's site and investor materials nor its product announcements and the trade coverage give a price, unit or tier for any line, including the agentic platform. Two things set this apart from ordinary enterprise silence. First, securities law supplies a partial substitute. Quarterly and annual filings disclose segment revenue, client counts, growth and risk factors, so a buyer can size the business unit selling to them and read management's own account of its difficulties.
Private vendors rarely offer that. Second, the marketplace makes the pricing silence structurally more awkward. A marketplace has at least three commercial relationships in it: between the institution and the platform, the institution and the agent builder, and the platform and the agent builder. None of them is described. A buyer evaluating whether to build an agent, buy a first party one or take a partner's cannot compare those paths on cost using anything published.