Do You Really Know Your Customer?

The four things an agent must carry to checkout.

A real customer, declined

A customer asks an agent to buy a laptop for no more than $1,200. The agent reaches checkout from a cloud browser. The customer is real and the card is valid. The merchant's bot screen sees an automated session and declines the purchase.

So the merchant loosens the screen. The next agent adds a $249 warranty to a $1,099 laptop and goes over budget. Another hits a timeout after the payment went through, cannot tell whether it did, and buys a second laptop. Each order looks legitimate at the gateway, and each one is a likely dispute.

Built for a person at the counter

Know your customer grew up around a simple assumption: the person in front of you is the person who will pay. Card checks followed the same line. Address verification, the security code and 3-D Secure all ask whether this is the cardholder. When the answer is yes, the merchant has done its part.

That question still matters, but it no longer settles the purchase. A cardholder can be fully authenticated and still never have agreed to the warranty, the second laptop or the price. Authentication tells the merchant who is there. It says nothing about what that person allowed.

The agent at checkout

At an agent's checkout the customer is absent, and the agent acts on instructions the merchant never saw. Screening the session for automation catches the wrong thing, because the automation is the whole point. A well-behaved agent and a hijacked one look the same to a bot detector.

The harder part is what no single merchant can see. A customer who allowed $1,200 for a laptop may have an agent that already spent $900 at another store this morning. The merchant sees one order. The limit spans every store the agent touched, and only a party in the path of all of them can count against it.

The carry test

Before accepting an agent's purchase, a merchant should be able to check four things the agent carries with it. An agent that cannot produce them is asking the merchant to accept the order on trust.

  1. Who sent it. A principal the merchant can verify, and the chain of delegation from that person to this agent. An agent that hands work to a sub-agent should not gain authority along the way.
  2. What it was told to buy. The customer's instruction in a form a machine can check: the merchant or category, the item, a price ceiling, an expiry, and whether a substitute is acceptable. A laptop under $1,200 rules out the warranty on its face.
  3. Whether that authority still stands right now. The customer may have revoked it an hour ago, or the budget may already be spent elsewhere. This answer has to come from someone who sees the agent's spending across every store, at the moment of purchase.
  4. One identity for this purchase. A stable reference that survives a retry, so a timeout after payment leads to a status check and never a second order. The merchant can then tell a retry from a new purchase.

The answers to all four should come back as a signed record that a third party can verify later, without trusting the agent or the merchant. That record is what a merchant points to when a dispute arrives.

What to ask for

Merchants should stop treating automation as the signal. Ask the agent for the four, and decline the purchases that cannot show them. A legitimate agent that carries them deserves the approval a bot screen would have denied.

Agent operators hold the other side. They are left with the bill when their agent buys twice or overspends, so they have the strongest reason to produce the four on every purchase. FederatedIQ is built for that decision: made before the purchase runs, counted across every store the agent touches, and signed.

Knowing your customer now means knowing what your customer allowed.

Cross-link: second Vol 2 essay, following N° 040 (The Specific Answer). Direct extension of N° 037 (A Credential Is Not a Mandate) and N° 038 (A Wallet Is Not a Decision): the arc's credential/mandate and tap/decision distinctions applied to the merchant-side KYC surface as an operational checklist. The "cannot be checked by one merchant alone" structural claim about aggregate spending across stores is the practitioner version of the cross-system authorization problem specified in N° 032 (The Authorization Gap Has a Literature Now) and the foundational N° 020 (The Authorization Gap). Vol 2 charter for naming FederatedIQ in editorial voice established in N° 039 (After the Argument) and exercised in N° 040. The "signed record a third party can verify" requirement is from N° 030 (A Signature Is Not a Premise). 3-D Secure, AVS, and CVV behavior described at the industry-standard level, no specific vendor cited. The laptop/warranty/retry scene is illustrative rather than reported; no specific merchant or incident is named. The one-page product pitch for FederatedIQ is at fixgovernance.ai/fiq. The product itself is at federatediq.ai.

End N° 041 · Vol 2 · 02