Company Spotlight: Delta Network
June 30, 2026
In my recent posts on Agentic Data Battle: Intent and Agentic – Intent and the New Data Games, I’ve emphasized that the trust challenge in agentic commerce goes far beyond authenticating the consumer and the agent. We must verify the action itself (the fourth pillar of any transaction). But no one is willing to budge. Platforms don’t want to give out intent to banks or networks (even with explicity consumer consent), they don’t want to be measured. While networks are the right neutral party, network VAS means loss of control. Today’s blog outlines the hard data on agent intent failure (28%) and best in class example of how to fix it.
The Problem: Self-Attestation Doesn’t Work
Mastercard’s “verifiable intent” framework sounds reasonable in principle: the agentic platform attests that it executed the consumer’s intent. The problem? The platform is grading its own homework.
Agentic intent is not a simple boolean. A consumer doesn’t say “buy shoes: yes/no.” They say “buy cotton joggers, navy blue, size medium, under $60, machine washable.” That’s five constraints. An agent can satisfy four and fail one—and still claim success. The consumer gets the wrong product, the merchant eats the return, and the network sees a completed transaction.
Self-attestation by the platform (which benefits from the transaction) is structurally compromised. We need a neutral third party to verify that the agent actually executed what the consumer wanted—before payment is released, not after.
Why This Matters: Consumer Trust Is Fragile
Here’s what keeps me up at night: if agents fail early, consumers may refuse to delegate for years.
The first wave of agentic commerce failures won’t be obvious fraud. They’ll be subtle: the agent bought organic when you said conventional, booked the aisle seat when you wanted window, selected the wrong shade of blue. Death by a thousand paper cuts. Each failure erodes the trust needed for delegated commerce to scale.
Consumers who get burned will revert to manual shopping. And unlike a bad restaurant experience they’ll forget, a rogue agent spending their money will create lasting aversion. We have one shot to get this right. To be clear, I think most the agentic hype is in full swing. Consumers love to shop.. but for the sake of discussing the problems…
Enter Delta Network: Real Data, Real Solution
I finally have real data on intent enforcement failures, courtesy of Myles O’Neil at Delta Network. The results are sobering—and the solution is promising.
The Methodology
Delta ran 100 single-product purchase intents through two agents, spanning easy lookups (1–2 constraints) to hard multi-constraint combinations that may not even exist. Both agents used Shopify’s UCP CLI for discovery. One agent additionally used Delta Mandate for intent enforcement.
Each intent carried explicit constraints: size, color, material, price, product requirements. Both agents were instructed to pass if they couldn’t find a suitable item rather than picking something that didn’t fit.
The Results
| Outcome | Without Delta | With Delta |
|---|---|---|
| Correct purchase | 71.2% (47/66) | 100% (56/56) |
| False positives: wrong purchase released | 28.8% (19/66) | 0% (0/56) |
| False negatives: correct purchase blocked | 32.4% (11/34) | 0% (0/44) |
| Passed: correct no purchase | 67.6% (23/34) | 100% (44/44) |
Rows 1–2 are % of purchases attempted. Rows 3–4 are % of cases where the agent passed.
The false positive rate—wrong purchases that were released—is the metric that matters for delegated spend. The control agent failed this test 28.8% of the time. With Delta Mandate: zero.
Error Rate by Difficulty
| Difficulty | Control Agent Error Rate |
|---|---|
| Easy (1–2 constraints) | 22.2% |
| Medium (3–4 constraints) | 25.0% |
| Hard (5+ constraints) | 71.4% |
The error rate scales with constraint count. More complex intents mean more opportunities for the agent to assert a constraint is met without verifying it. Every failure was the agent claiming a constraint was satisfied without evidence—buying fleece joggers and calling them cotton, or passing “dishwasher safe” on a product whose catalog data doesn’t mention dishwasher safety.
There was no intent the control agent satisfied that the agent equipped with Delta Mandate couldn’t. The agent with Delta found more valid products AND got every one of them right.
The Delta Mandate Approach
Delta’s solution addresses the gap I’ve been writing about for months. Here it is, in their words:
“Existing payment controls can decide whether a transaction may move, but they cannot verify whether the agent bought the right thing, from the right counterparty, on the right terms. Delta closes that mandate verification gap by checking the substance of the agent’s proposed action before payment release: the user intent becomes a policy, the proposed purchase is turned into evidence, and the action is allowed only if the evidence satisfies the mandate.
Agentic commerce cannot scale without chargeback-like guarantees for delegated spend. Users, wallets, fintechs, and orchestrators need to know that if an agent buys outside the user’s mandate, someone is commercially accountable. To our knowledge, Delta is the only solution in market pairing pre-release mandate enforcement with a financial backstop. Under the Delta Mandate Guarantee, if Delta clears a covered transaction that our mandate check should have blocked, Delta covers the loss up to the contracted cap. That gives partners a way to offer protected delegated spend now, before the issuer and network frameworks fully catch up.”
This is the architecture I’ve been calling for. Not self-attestation. Not post-hoc dispute resolution. Pre-release mandate enforcement with accountability.
What This Means for the Industry
The data from Delta confirms what I suspected: agent error rates are unacceptably high for complex intents, and the industry’s current approach—letting platforms attest to their own compliance—won’t scale.
The path forward requires:
- Neutral verification — A third party that checks evidence, not attestations
- Pre-release enforcement — Catch errors before payment, not after
- Financial accountability — Someone takes the loss when mandate verification fails
Delta is the first solution I’ve seen that delivers all three. The issuer and network frameworks will eventually catch up, but merchants, wallets, and orchestrators need protection now.
The agentic apocalypse isn’t inevitable. But avoiding it requires infrastructure that doesn’t exist in today’s payment rails. Companies like Delta are building what the networks should have built years ago.
For more on the intent challenge in agentic commerce, see my previous posts: Agentic Data Battle: Intent, Agentic – Intent and the New Data Games, and Agentic Commerce Economics and Governance.
Data and methodology attributed to Delta Network. More on the Delta Mandate approach: delta-mandate-product.repyhlabs.dev