The Agents Are Coming For Your Breakage

by Ory Adler, published October 30 2026

A shopper tells her agent: "Reorder my usual grocery basket with the cheapest total after points and offers." The agent checks three retailers in a few seconds, nets out every earn rate and personalized offer it can see, and places the order for pickup tomorrow morning.

Your 1-point-per-dollar program just became a 1% discount in a comparison table. The "you're 200 points away" email, the bonus event, the satisfying climb of a growing balance, none of it changes the agent’s calculation. In a 2026 Commerce and PayPal study of 3,000 shoppers across the US, UK and Australia, about two-thirds said they want to try agentic shopping tools, and the job they most want done is finding lower prices, discounts and promotions.¹ Checkout.com puts agent involvement at ~3% of transactions today, yet a third of consumers expect AI to drive at least 10% of their purchases within the next year.²

Yes, most shoppers still want the final say before an agent buys anything³, but that doesn't buy you much time. Clicking the "buy" button or visiting a store doesn’t change which retailers make the shortlist, but the visibility of offers and points balance does.

You Can't Defend What You Can't Measure

Few loyalty leaders can cleanly prove their program's ROI today. The program is open to everyone, so there's no natural control group. The data feeds pricing, assortment, promotions, real estate and labour, so its value runs through across a dozen P&L lines. Every budget cycle, finance asks, “What are we getting for these investments?”, and marketing answers it with a mix of conviction and slideware. Agents will put those economics under pressure and if you don't have a strong and up-to-date understanding of the baseline (true cost net of breakage, incremental margin contribution, the value of the data), every budget review gets harder to win.

Agents Turn Every Member Into Your Cherry-Picker

In some of the programs we designed, we focused on finding the customers who drove the most value for the business and treating them differently from the rest of the customer base. Every program also has the opposite segment, the rational deal-seekers who never miss a bonus and never let a point expire. You accept them because they're often a small part of the base and realistically, you cannot simply turn them away. Agents could make far more members behave like that segment, tracking the balances and offers shoppers forget, breaking your historical breakage curve.

For example, a program that gives back 1.5% of spend with 25% breakage really costs ~1.1%. Push breakage to 5% and it costs ~1.4%. That's a ~27% increase in reward cost with no change to the program, the earn rate or the customer base. Then there's the balance sheet. Under IFRS 15 and ASC 606, revenue tied to points is deferred based on expected redemption. Lower expected breakage means more revenue deferred and a bigger liability. Your CFO will want that model before signing off on your next agentic investment case.

The harder problem is differentiation. Points have always carried a perceived value above their cash value. Mental accounting, the goal gradient and the pleasure of a growing balance all do work that a straight discount can't. An agent buying on price gives those effects little weight as it converts your points into their cash equivalent and drops them into a single cost-and-convenience equation. The knee-jerk response will be to pour money into points or price to win the agent's cost math, and that's the race to the bottom. And the other implication is that points will lose their role as the headline feature of the program as agents simply treat them as a discount.

An Open or Closed System is a Choice Until it is Not

All of this assumes third-party agents can see your member balances and personalized offers. The two biggest players have already chosen different approaches. Amazon went closed: it sued Perplexity in November 2025 to stop the Comet browser agent from shopping its store, alleging among other things that Comet misappropriated Prime membership benefits, and won a preliminary injunction this March. An appeals court vacated it in August, but the case continues. ⁴ Walmart went open, launching API gateways and partnering with OpenAI so ChatGPT users can buy from its catalogue through Instant Checkout. ⁵ For everyone else, competitive pressure may make that choice harder to sustain.

When an agent ranks "cheapest total after points and offers," an offer it can't see counts as zero. The day that one competitor in your category exposes its program, its member offers will count in comparisons while yours will not. Even if your underlying prices are competitive and your member offers are richer, they will do you no good if they are invisible. So the first grocer, pharmacy or pet retailer to expose its member pricing to agents puts the rest of the category on the clock: follow or lose the comparison by default, it’s an agentic commerce version of the prisoner’s dilemma. The category might be better off if everyone stayed closed, but staying closed gets harder once a competitor opens the door. Amazon can afford to stay closed (at least for now) because it owns the agent (Alexa for Shopping and Buy for Me), the membership and the traffic⁶ but most retailers own at most one and many none of those.

What can Survive the Agent?

Anything that sits outside the price-and-convenience math.

  • Membership fees create a sunk cost the customer wants to earn back, and an agent told "use my memberships first" will respect it. That helps explain Amazon's lawsuit: the asset it went to court to protect was its Retail Media income and Prime Member Base.

  • Status and recognition of you as a customer that shows up at the service desk, or in fulfilment priority, or in how a store associate greets you is hard to quantify, which makes it hard to commoditize.

  • Household needs, dietary preferences, the dog's breed and age. The retailer that feeds the agent the richest context has a better chance of becoming its default.

  • “Money-can't-buy Experiences” are also highly effective, so long as they are specific and relevant to your brand and genuinely limited

Where to Start?

  1. Understand the program’s true worth to the business. Start with the direct impact: incremental sales and margin from changed member behaviour, measured against a credible control. Then add everything the program feeds: the pricing, assortment and promotion decisions built on its data, and adjacent income like the retail media revenue your audiences make possible. That full number is your baseline, and it’s what you’ll be defending as agents reduce breakage.

  2. Re-run your economics at agent-level redemption. Model breakage at different levels and look hard at the P&L and the deferred revenue line.

  3. Decide your agent-access posture before someone decides it for you. Who in your category opens first, and what's your response the week they do? How will you differentiate your program and brand in a world where points are just another form of discount?

Amazon and Walmart are the biggest threat to most retailers, and in an agentic world their moats look stronger. Amazon owns the membership, the agent and the convenience-enabling logistics, and it's in court to keep outsiders out. Walmart has the convenient locations reach, traffic, buying power and a paid membership, and it chose to be inside the agent from day one.

The exposed retailers are the ones in the middle: a free points program, no agent of their own, and a loyalty currency that is about to become a line item in an agent’s comparison. Before making more investments, I’d want to know how much of the program’s ROI depends on customers not realizing its full potential. If the economics weaken when customers optimize everything they’re entitled to, you have a loyalty problem to solve now, before the AI problem exacerbates it.

Sources

  1. Commerce & PayPal, Agentic AI Shopping Research, press release (June 23, 2026): https://investors.bigcommerce.com/node/12861/pdf

  2. Checkout.com, Agentic Commerce 2026: The State of Consumer Demand and Merchant Readiness (June 9, 2026): https://www.checkout.com/newsroom/consumer-demand-for-ai-shopping-is-forming-fast-but-trust-for-agentic-commerce-is-still-catching-up

  3. Gartner, Consumers Want AI Shopping Help, But Not AI Purchase Decisions (May 27, 2026): https://www.gartner.com/en/newsroom/press-releases/2026-05-27-gartner-survey-finds-consumers-want-ai-shopping-help-but-not-ai-purchase-decisions

  4. Amazon v. Perplexity. Lawsuit, Payments Dive (Nov 2025): https://www.paymentsdive.com/news/amazon-sues-perplexity-ai-shopping-agents/804923/; preliminary injunction, GeekWire (Mar 10, 2026): https://www.geekwire.com/2026/03/10/judge-blocks-perplexitys-ai-bot-from-shopping-on-amazon-in-early-test-of-agentic-commerce/; injunction vacated by the Ninth Circuit, Courthouse News (Aug 2026): https://courthousenews.com/ninth-circuit-lifts-block-on-ai-powered-shopping-assistant/

  5. Walmart. OpenAI partnership and ChatGPT Instant Checkout, Retail Dive (Oct 2025): https://www.retaildive.com/news/walmart-openai-chatgpt-shopping-instant-checkout/802708/; Walmart Agent Gateway, TechCrunch (Oct 21, 2026): https://techcrunch.com/2026/10/21/while-amazon-blocks-ai-agents-walmart-rolls-out-the-welcome-mat/

  6. Retail Systems, Amazon sues Perplexity over agent-led shopping (Rufus, Buy for Me): https://retail-systems.com/rs/Amazon_Sues_Perplexity_Over_Agent_Led_Shopping_On_Its_Marketplace.php; Rufus renamed Alexa for Shopping, About Amazon (May 2026): https://aboutamazon.com/news/retail/amazon-rufus-ai-assistant-personalized-shopping-features