Corpay: How Far Should the Agent Go?
Download case PDF (5 pages)Request teacher’s pack
Decision date and information cutoff: October 3, 2026
Decision scope: The next phase of AI-enabled spend management
Case convention: This is a constructed classroom decision based on public information. The hypothetical role and alternatives below are not Corpay's disclosed internal plan. No private meeting, budget, quotation, customer result, or pending management choice is represented.
The decision
By early October 2026, Corpay had announced several AI capabilities in Corpay Complete. Its latest release described agents for spend insights, expense creation by voice, and accounting coding. Earlier in the year, it had announced an AI Virtual Assistant and Agent Card, a capability intended to let trusted AI agents create and use virtual cards within payment controls. The company said the expanded agent team was rolling out to eligible customers in September. It had not publicly reported agent adoption, accuracy, customer savings, or payment-related error rates. [S1] [S2] [S3]
Corpay had thus introduced products spanning different kinds of action. An assistant that answers a question informs a user. An agent that codes an expense changes a workflow record. An agent using a virtual card may initiate a payment under delegated authority. Treating these activities as one undifferentiated category of “AI” obscures the choice in front of a payments provider: how much action to delegate, to whom, under what limits, and with what recourse if an action is wrong.
For this exercise, assume you are advising a hypothetical Corpay Complete product committee at its next planning review. Your assignment is to recommend how Corpay should sequence its next phase of AI-enabled spend management. The committee's choice is constructed; Corpay has not disclosed such a committee or a pending decision.
Choose and justify a sequence among three paths:
- Deepen workflow support. Concentrate on insights, expense handling and coding. Keep payment execution under direct user action while improving the reliability and usefulness of the announced agents.
- Extend bounded payment execution. Test a small number of well-defined Agent Card workflows in which an authorized business user sets the purpose and transaction boundaries. Recommend which workflows could qualify and what approval, exception and recovery rules should apply. The specific test, safeguards and release gate are classroom proposals; no such pilot or controls are asserted to exist.
- Prioritize interoperability. Invest product effort in connecting Corpay Complete and its payment controls to external agent, procurement and business-software ecosystems. Specify how Corpay should define identity, intent, permissions and responsibility across those connections. This path builds on the ecosystem direction present in competitor announcements; it is not a claim about Corpay's current integrations.
You may combine or sequence these paths. Identify the work Corpay should defer. Do not assume that any path has a known customer return or that the company has a fixed investment budget.
A company built around controlling spend
Corpay sells payment and spend-management services to businesses. Its disclosed solution groups include Spend Management, Cross-Border and Vehicle Solutions. In Corporate Payments, the company describes products such as corporate and virtual card programs and procure-to-pay automation. These businesses put Corpay near both the administrative work around a transaction and the payment itself. [S1] [S4]
The company's AI releases followed that existing product footprint. In April, Corpay introduced the AI Virtual Assistant, which it described as helping users see spend and answer questions. In July, it announced Agent Card, describing virtual cards as a way for trusted agents to create and use payment credentials within existing controls. On September 24, Corpay announced additional agents inside Corpay Complete: an Embedded Insights Agent that flags anomalies, a Voice Assistant that creates and completes expenses and answers card/account questions, and an Accounting Coding Agent that assigns general-ledger, tax and cost-center codes. Corpay said this group was rolling out to eligible customers that month. [S1] [S2] [S3]
The announcements establish what Corpay said it was building and rolling out. They do not disclose how many customers had enabled a feature by October 3, whether an agent had completed a payment, how often an expense or code required correction, or what the features cost to develop and operate. Nor do they explain the authorization details for each announced use case. The product label “agent” alone does not answer which actions remain suggestions, which update records, and which can cause money to move.
Growth and investment context
Corporate Payments was growing quickly. In Q2 2026, Corpay reported Corporate Payments revenue of $548.7 million, up 41.7% year over year. The filing attributed 16% organic growth to 43% growth in spend volume, new payables and cross-border sales, acquisitions and favorable foreign-exchange and fuel effects. Acquisitions contributed about $78 million of revenue during the quarter. Revenue per spend dollar declined, which Corpay attributed to the mix of new payables and cross-border enterprise customers. Segment operating income rose 27.2% to $199.6 million. [S4, pp. 36–37]
| Q2 2026 measure | Reported result | Scope |
|---|---|---|
| Corporate Payments revenue | $548.7 million; +41.7% year over year | Segment; includes acquisition and favorable macro effects |
| Corporate Payments organic revenue growth | +16% | Company-defined organic measure |
| Corporate Payments spend-volume growth | +43% | Segment KPI |
| Corporate Payments operating income | $199.6 million; +27.2% year over year | Segment |
| Consolidated revenue | $1,338.8 million; +21.5% year over year | All reported businesses |
| Net income attributable to Corpay | $248.3 million; −12.6% year over year | Consolidated GAAP result |
Source: Corpay Q2 2026 Form 10-Q, filed August 10, 2026, and earnings release, August 5, 2026. Reported Corporate Payments growth includes acquisitions and favorable foreign-exchange and fuel effects; the filing says revenue per spend dollar declined with the new-customer mix. Consolidated results include all Corpay businesses and a $100 million charge for a preliminary FTC settlement. None of these figures measures AI adoption, performance or returns. [S4, pp. 35–37] [S5]
The results offer context, not a ready-made investment case. Organic spend growth could create an opportunity to add useful services around payment volume. At the same time, the mix shift reduced revenue per spend dollar, and operating income reflected integration synergies as well as sales investment and integration expense. Corpay management had also described portfolio rotation, commercial growth and AI products/internal process redesign as priorities. Public disclosures do not say how much capital or staff the company should dedicate to agentic products relative to those other workstreams. [S4] [S6]
Trust and the payment boundary
Corpay's public descriptions emphasize controls and audit visibility, but they provide few details on how a customer's authority would translate into an agent's permitted action. A business might want an agent to find an approved supplier, prepare an expense, select a payment method or execute the payment. Those tasks carry different permissions and consequences.
In a workflow-support path, an agent can identify an anomaly or suggest a code while a person decides what to do. In a payment-execution path, a user may delegate limited authority in advance; an agent's action can create an external payment before a person sees the individual transaction. A virtual card may offer a transaction-specific instrument, but students must assess what that instrument can and cannot control. The public announcement does not specify Corpay's exact limits, authentication process, reversal process, or exception rights. Do not assume a virtual card eliminates mistaken intent, compromised credentials, supplier disputes or duplicate invoices. [S2]
Competitors and payment networks were also describing agent-initiated payment infrastructure. Mastercard announced Agent Pay for Machines in June 2026, highlighting credentials, programmatic permissioning, spending rules and settlement. Visa's commercial-solutions materials described procurement agents sourcing from approved suppliers and paying suppliers through network infrastructure. These are company announcements and positioning, not comparable measures of live deployment or customer results. Visa and Mastercard's network roles also differ from Corpay's spend-management platform role. The announcements nevertheless raise a strategic question: should Corpay build a distinctive in-platform experience, connect to others' agents and standards, or do both in sequence? [S7] [S8]
What Corpay does not yet disclose
As of October 3, Corpay's public materials did not provide:
- the number or percentage of eligible customers who used each AI feature;
- customer-level time savings, retention, spend, willingness-to-pay or realized savings;
- accuracy, false-positive, correction, override or exception rates by task;
- details sufficient to map the human user, agent identity, card credential and payment authorization for every use case;
- fraud, duplicate-payment, dispute, recovery or customer-support outcomes;
- incremental revenue, contribution margin or build/operating costs attributable to AI; or
- the integration commitments and dependencies required for external agent interoperability.
This absence does not show that the products lack value. It means a recommendation should define what the company must learn, how it should measure it and what evidence would justify expanding delegated authority.
Assignment
Prepare a decision memo for the hypothetical product committee. Recommend a next-phase sequence, and explain:
- Which tasks should remain advisory or workflow support, and which—if any—should be allowed to initiate payment?
- Which users, suppliers, transaction types or payment amounts should be in scope first? What permissions and exception paths would you require?
- Should Corpay prioritize its own integrated experience, interoperability with external agents, or both? What must be standardized or controlled at the boundary?
- Which customer, reliability, risk and unit-economic measures should govern expansion? What result would cause you to pause or reverse course?
- What is the strongest argument against your recommendation, and what evidence would change your view?
Distinguish the announced facts from the assumptions in your proposal. Do not assign a numeric return without the missing inputs. The information cutoff is October 3, 2026; later events and outcomes are outside the case.
Sources and conventions
S1. Corpay, “Corpay Launches AI Capabilities to Modernize Spend Management”, April 28, 2026; release body.
S2. Corpay, “Corpay Introduces Agent Card Capability to Power Secure Agentic Commerce”, July 28, 2026; one-page announcement, p. 1.
S3. Corpay, “Corpay Expands Its AI Agent Team to Give Businesses More Control Over Spend”, September 24, 2026; release paragraphs describing launch sequence, agent functions and rollout.
S4. Corpay, Q2 2026 Form 10-Q, accession 0001175454-26-000050, filed August 10, 2026, quarter ended June 30, 2026; MD&A pp. 35–37, 39–40.
S5. Corpay, Q2 2026 financial-results release, August 5, 2026; GAAP/non-GAAP results and settlement charge.
S6. Corpay, Q1 2026 earnings remarks, company-hosted PDF, pp. 4–7; management's stated strategic priorities. These remarks predate the public agent launches.
S7. Mastercard, “Mastercard launches Agent Pay for Machines”, June 10, 2026; release sections on credentialing, permissioning and settlement.
S8. Visa, commercial-solutions agentic AI hub, accessed by cutoff; B2B procurement and supplier-payment discussion. This is Visa's positioning, not independent market evidence.
All figures are reported in U.S. dollars as presented by Corpay. Organic growth is company-defined. The classroom committee, decision, alternatives, boundaries and performance gates are hypothetical. The case does not include a private teaching answer or later outcome.