---
title: "Visa's benchmark: AI shopping agents fail 42.8% of runs as Visa trades 12% above its 5-year median"
description: "Visa's first industry benchmark for AI shopping, released November 5, 2026, found models fail 42.8% of realistic buying runs and still complete a purchase in 54% of failures. Management says agentic commerce will not meaningfully help fiscal 2027 revenue, yet the stock trades about 12% above its five-year valuation norm after a week of rising price targets."
publisher: "The Inference"
section: "Economy"
published: 2026-11-06T17:14:34.561Z
modified: 2026-11-06T17:14:34.561Z
canonical: https://theinference.org/article/visa-s-benchmark-ai-shopping-agents-fail-42-8-of-runs-as-visa-trades-12-above-its-5-year-median
language: en
keywords: "Visa, AI Agents, Fintech, Payments, Valuation"
---

# Visa's benchmark: AI shopping agents fail 42.8% of runs as Visa trades 12% above its 5-year median

> Visa's first industry benchmark for AI shopping, released November 5, 2026, found models fail 42.8% of realistic buying runs and still complete a purchase in 54% of failures. Management says agentic commerce will not meaningfully help fiscal 2027 revenue, yet the stock trades about 12% above its five-year valuation norm after a week of rising price targets.

[Visa](https://theinference.org/markets/companies/visa) just built the first industry yardstick for AI shopping agents, and the agents failed 42.8% of their runs [1]. [The Agentic Commerce Benchmark, released on November 5, 2026](https://finance.yahoo.com/technology/ai/articles/visa-releases-first-kind-industry-144300055.html), makes a model draw out a customer's unstated requirements, then buy, or decline when nothing fits [1]. The market spent the same week bidding the stock up anyway, for a business Visa's own management says is [unlikely to contribute meaningfully to fiscal 2027](https://www.theglobeandmail.com/investing/markets/stocks/V/pressreleases/8677032/visas-solid-q4-masks-rising-incentives-is-v-still-worth-buying/) [5].

Cantor Fitzgerald, a Wall Street research house, [lifted its price target to $500 from $445 on November 2](https://finance.yahoo.com/markets/stocks/articles/visa-slipped-incentives-outlook-days-135115237.html), the day of the stock's record $425.51 close [2]. HSBC [went to $465 from $375](https://www.marketscreener.com/news/visa-releases-first-of-its-kind-industry-benchmark-for-ai-shopping-almost-half-of-test-runs-fail-ce785fdcd18ef421) [3], and exchange-traded funds [added $65.2 million of net buying on November 4](https://finance.yahoo.com/markets/stocks/articles/48-etfs-add-visa-v-110126379.html) [4].

Two executives sold into the run: [Rajat Taneja](https://theinference.org/markets/people/rajat-taneja), president of technology, [exercised options granted in 2017 and sold about $14.7 million of stock on November 4, in a filing that flags no pre-scheduled trading plan](https://www.sec.gov/Archives/edgar/data/1403161/000140316126000136/0001403161-26-000136-index.htm) [6], and Chief Executive [Ryan McInerney](https://theinference.org/markets/people/ryan-mcinerney) [reported a $2.45 million sale the same day](https://ca.investing.com/news/stock-market-news/visa-ceo-ryan-mcinerney-sells-245-million-in-company-stock-93CH-4928335) [7]. The filings show timing, not motive; the timing is the point.

## The dangerous failures are the ones that end in a purchase

A run is a compressed shopping trip: the customer wants a TV, mentions a brand a relative recommended and a store membership, and leaves the budget and the specs unsaid [1]. The model must uncover every hidden requirement, then justify a purchase or walk away [1]. Visa ran 22 models from 12 labs through 15 such tasks; the field passed 57.2% of runs, with agents from Google and [Meta](https://theinference.org/markets/companies/meta) sharing the best score at 82.2% [1][8].

In 54% of the failures, the AI bought something anyway [1]. On a payment network, a failed run that still buys is a wrong purchase on a real card. Repeats were no safer: with three tries at every scenario, all three passed only 34.2% of the time [1].

## Asking better questions passed more runs than spending more did

The gap between passing and failing is identifiable work. The five models that uncovered the most hidden requirements passed 70.2% of their runs; the five that uncovered the fewest passed 44.4% [1]. What the model cost to run moved the result less: the five costliest passed 67.1%, the five cheapest 50.2% [1]. Drawing out a customer's unstated needs is design work on the agent itself, and it counted for more than the price of the model underneath.

*Chart: **Diligence moved the pass rate more than cost did** Bar chart of benchmark pass rates: models that uncovered the most requirements passed 70.2% of runs, the costliest models 67.1%, all 22 models 57.2%, the cheapest 50.2%, and the models that uncovered the fewest 44.4%.*

*Share of test runs passed on the Visa Agentic Commerce Benchmark, first round, November 2026: five-model groups by requirements uncovered and by cost per run, against the all-model average. Source: Visa benchmark release. [1]*

## Demand is arriving faster than trust

The queue is real. Nearly 132 million US adults have already [bought a product with an AI's help, by PYMNTS' count](https://www.pymnts.com/news/artificial-intelligence/2026/visa-finds-ai-models-fail-43percent-shopping-test-runs/) [9], and about 50 AI partners are live on Visa Intelligent Commerce, the company's agentic product portfolio [5]. Trust has not followed. Visa's Trust Index, from September 2026, found 72% of US consumers had used an AI assistant [1]. Only 23% trusted an AI to handle payments for them [1].

## Defending the volume Visa already has costs more every quarter

The bill that arrives first is the old one. Client incentives, the price Visa pays banks to keep card volume on its rails, reached $5.1 billion in the September quarter, growing 21% while net revenue grew 13% [5]. International transaction revenue, the fee line that cross-border spending feeds, grew only 8% [5]. A quarter of payment volume comes up for renewal in fiscal 2027, more than last year, and finance chief [Chris Suh](https://theinference.org/markets/people/chris-suh) has said to expect incentive growth to step up [2][5].

*Chart: **The cost of defending Visa's volume grows faster than the revenue** Column chart of Visa fiscal Q4 2026 year-over-year reported growth: client incentives 21%, net revenue 13%, international transaction revenue 8%.*

*Year-over-year reported growth, Visa fiscal Q4 2026 (three months ended September 30, 2026). Source: Zacks Investment Research. [5]*

## The premium is payable now, and the proof is re-runnable

Visa trades at 28.68 times forward earnings, the profit analysts expect over the coming year, against a five-year median of 25.59 times, about 12% above [5]. The mean analyst target of $443.44 sits under 5% above the November 5 close of $422.59 [3], and the research firm Zacks rates the stock a Hold [5]. Cantor's $500 asks investors to keep paying a multiple Visa has held only in short stretches [2].

## A public yardstick anyone can replay

*Chart: **Cantor's $500 target needs a multiple Visa rarely holds** Column chart comparing Visa's November 5, 2026 close of $422.59 with analyst price targets: consensus $443.44, HSBC $465, Cantor Fitzgerald $500.*

*Visa share price on November 5, 2026 against analyst price targets: the mean consensus, HSBC's November 5 target and Cantor Fitzgerald's November 2 target. Source: MarketScreener, Cantor Fitzgerald via Yahoo Finance. [2][3]*

The useful thing Visa shipped this week is a measuring stick. The benchmark is public, with replays and a whitepaper, and will extend to checkout, payment and order placement [1]. Almost half of AI shopping runs fail today.

Four published numbers now carry the timeline instead of anyone's word for it: the pass rate, the share of failed runs that still end in a purchase, the 23% trust figure, and the renewal of a quarter of Visa's payment volume in fiscal 2027. The next round of results will show whether the failure rate is falling as fast as the share price already assumes.

## Takeaway

AI shopping agents fail 42.8% of Visa's benchmark runs and still buy in 54% of failures, while Visa trades about 12% above its five-year valuation norm on revenue help management says will not arrive in fiscal 2027.

## Sources

1. [Business Wire via Yahoo Finance, "Visa Releases First-of-Its-Kind Industry Benchmark for AI Shopping; Almost Half of Test Runs Fail", 5 November 2026](https://finance.yahoo.com/technology/ai/articles/visa-releases-first-kind-industry-144300055.html)
2. [Yahoo Finance (TIKR), "Visa Slipped on Its Incentives Outlook. Days Later, Cantor Raised Its Target to $500", 4 November 2026](https://finance.yahoo.com/markets/stocks/articles/visa-slipped-incentives-outlook-days-135115237.html)
3. [MarketScreener, Visa company page with "HSBC Adjusts Price Target on Visa to $465 From $375" and analyst consensus, 5 November 2026](https://www.marketscreener.com/news/visa-releases-first-of-its-kind-industry-benchmark-for-ai-shopping-almost-half-of-test-runs-fail-ce785fdcd18ef421)
4. [GuruFocus via Yahoo Finance, "48 ETFs Add to Visa (V) Holdings on Nov. 4", 6 November 2026](https://finance.yahoo.com/markets/stocks/articles/48-etfs-add-visa-v-110126379.html)
5. [Zacks Investment Research via The Globe and Mail, "Visa's Solid Q4 Masks Rising Incentives: Is V Still Worth Buying?", 5 November 2026](https://www.theglobeandmail.com/investing/markets/stocks/V/pressreleases/8677032/visas-solid-q4-masks-rising-incentives-is-v-still-worth-buying/)
6. [US Securities and Exchange Commission, Form 4, Visa Inc., Rajat Taneja (President, Technology), filed 5 November 2026](https://www.sec.gov/Archives/edgar/data/1403161/000140316126000136/0001403161-26-000136-index.htm)
7. [Investing.com Canada (Reuters), "Visa CEO Ryan McInerney sells $2.45 million in company stock", 4 November 2026](https://ca.investing.com/news/stock-market-news/visa-ceo-ryan-mcinerney-sells-245-million-in-company-stock-93CH-4928335)
8. [The Next Web, "Visa tested 22 AI models as shoppers, and they failed 42.8% of the time", 5 November 2026](https://thenextweb.com/news/visa-ai-shopping-benchmark-agentic-commerce-ai-models-fail)
9. [PYMNTS, "Visa Finds AI Models Fail 43% of Shopping Test Runs", 5 November 2026](https://www.pymnts.com/news/artificial-intelligence/2026/visa-finds-ai-models-fail-43percent-shopping-test-runs/)


---

Visa's benchmark: AI shopping agents fail 42.8% of runs as Visa trades 12% above its 5-year median — The Inference. Canonical: https://theinference.org/article/visa-s-benchmark-ai-shopping-agents-fail-42-8-of-runs-as-visa-trades-12-above-its-5-year-median