Why ecommerce brands need AI customers

Ecommerce brands need AI customers, the AI agents for ecommerce that shop as customers, because much of their revenue runs through automated emails, texts, codes and reminders that can stop working without raising an error, and a customer sees the failure first. An AI customer becomes a real customer of a store and its rivals, asks AI assistants what shoppers ask, and keeps what it receives as evidence, every week and after every change. Most other AI agents for ecommerce work for the store, answering shoppers or helping run the admin.

Founders, heads of ecommerce, and growth and CRM leads at Shopify and direct to consumer brands can use AI customers to see their store as its customers do, from the sign up to what AI assistants say about it.

Most AI agents for ecommerce work for the store. Gorgias’s AI Agent sells, supports and engages customers 24/7, and Shopify’s Sidekick is an AI commerce assistant that works with a store’s own data. An AI customer works for your company from the customer’s side, and what AI customers are defines the term in full. The case for every company is in why AI customers matter.

Black Friday is on Friday 27 November 2026. Adobe forecasts $12.9 billion of US online spending that day, about 2.9 times the daily average of its $275.1 billion forecast for 1 November to 31 December. For a store whose sales follow the market, a broken step costs about 3 times as much on Black Friday as on an average day of the season.

How an ecommerce brand earns

An online brand earns from the first order and from the orders after it. The first order usually comes from ads, search and social media. The orders after it come back through email and text, which the brand owns, and many start with an automated message: the welcome series, the browse and basket reminders, the post purchase emails and the win back.

Klaviyo’s own 2026 benchmarks show that automated flows were 5.3% of email sends and nearly 41% of email revenue. Over Black Friday to Cyber Monday 2025, Klaviyo reported that across its customers, email and text drove 42% of total revenue, and 43% on peak days.

What 1 flow is worth to your company is in your email platform’s flow report: its attributed revenue for the last 30 days, divided by 30, is its revenue a day.

The week runs on a handful of tools: Shopify, Klaviyo or Omnisend for email, Postscript or Attentive for texts, Meta and Google for ads, Gorgias or Shopify Inbox for support, and GA4. Monday starts with last week’s numbers, and the rest goes on the next campaign and the next theme or app change. From October the week fills with Black Friday: the offer, early access, the codes and a freeze on code, apps and flows around 20 November.

Each step of the shopper’s path belongs to a different tool and often a different person: the ad to the performance lead, the pop up and flows to the email lead, the code and checkout to the ecommerce lead, the theme to a developer or an agency. Each tool reports its own part.

What a store can’t see

A store’s reports describe what the store did, such as sessions, orders, emails sent and chats closed, and they are accurate about it. The failures that raise no error happen afterwards, in front of 3 groups no report follows end to end: shoppers on the site, subscribers waiting for a message, and AI agents reading the store for someone else.

Shoppers on the site

A broken step on the site looks like a slow week. A store owner on the Shopify Community described it in September 2026: “Orders slow down, add-to-carts drop a little, and it just feels like a quiet week until someone finds the actual fault.” In the same thread, another owner said an app update broke a product button and took a couple of days to notice. Without a test from the customer’s side, a store learns of a break from a customer or from its numbers.

Subscribers waiting for an email or text

A flow that stops sends nothing, so nothing reports an error. A brand testing its basket flow wrote on the Klaviyo Community: “If I go to view our analytics for the flow, everything is 0.” A test of 362 ecommerce brands by Joy Joya, using new email addresses and phone numbers, found that 158 sent no welcome email, 53 never sent the text they promised and 20 sent an offer that didn’t match the pop up.

Basket reminders fail the same way. Baymard Institute puts the average documented basket abandonment rate at 70.22%, and Rejoiner’s analysis of 1,000 direct to consumer brands found that 68% of abandoned carts triggered a reminder email, so about 1 in 3 got none. The causes to check in Klaviyo and Shopify are in why a welcome or basket email never arrives.

AI assistants and shopping agents

AI assistants now answer shoppers’ questions about a store’s products and prices, and OpenAI’s help centre says ChatGPT may offer Instant Checkout for some eligible products and merchants, so the purchase happens in the chat. In a study Product.ai ran and Business Insider reported, 85% of 913 checkable price answers from ChatGPT, Claude, Gemini and Perplexity matched the current or listed price, and the wrong ones were off by a median of $300. A store sees an AI answer only when a shopper clicks through from it.

Rivals’ offers sit outside every report as well. A rival’s welcome code, early access price and Black Friday teaser go only to its subscribers, so they never appear on its product pages (a competitor’s real price is the price after the welcome code).

What a silent step costs

The cost of a silent step is its revenue a day multiplied by the days it stays silent, and the days depend on how often someone checks. On average a break runs for half the gap between checks before it is found: about 15 days with a monthly review, 3.5 with a weekly test and half a day with a daily one.

• Revenue a day: the flow’s attributed revenue for the last 30 days ÷ 30. The low case halves it, because attribution also counts shoppers who would have bought anyway.
• Days saved per break: the days a break runs with today’s checks, minus the days it would run with the new check.
• Breaks a year: the theme publishes, app installs, flow edits and expired codes in your last 12 months that stopped a step.
• Saved = revenue a day × days saved per break × breaks.

For Black Friday week, use last year’s Black Friday week from the same report. Lebesgue’s analysis of $22.6 million of Meta spend found that the median brand paid 59% more per 1,000 impressions in Black Friday week 2025 than in September, while its conversion rate rose 48%, so a broken step that week wastes dearer clicks and loses more buyers.

Set the saving against what checking costs. Checking those 4 flows daily by hand, at 15 minutes each, takes 30 hours a month, which is $13,633 a year at $37.87 an hour, the 2025 US median pay for market research analysts. On the example, it breaks even at 1.2 to 2.4 breaks a year. What AI customers cost to run, line by line at published prices, is in AI customers at scale.

Shopify website monitoring by hand

By hand, the hours a week are checks × stores or rivals × minutes each × runs a week, divided by 60. A November week at 1 store, with example inputs to replace with your own:

• Your own journeys: the 12 tests in the Black Friday checklist at about 10 minutes each, run weekly and after 2 changes, take 6 hours.
• 5 rivals: 15 minutes each to read what arrived and check prices, every weekday, take 6.25 hours.
• AI answers: 15 questions, asked 3 times on 3 assistants, take an afternoon of about 4 hours.

That is about 16 hours a week, or 70 hours and about $2,650 a month at the same hourly cost. The hours grow faster than the store:

• Every market adds a store’s worth of checks, with its own pop up, codes, currency and delivery promises.
• Every run needs a new identity, because a welcome email comes once per address.
• The waits stay long. A basket reminder set for 1 hour still needs watching for a day.
• The evidence is whatever someone saved, usually a screenshot. The email itself, kept as it arrived, shows when it came and who sent it.

A store that runs these checks as a routine is practising customer-side observability. Customer-side observability is seeing what a company does to its customers by being one of them: every email, text, price, page, checkout and AI answer it sends, kept over time with evidence anyone can check. For a store, it rests on 3 things:

• Pass marks written in advance. The welcome arrives within 5 minutes with the pop up’s code, and the basket reminder within the flow’s delay plus 1 hour, with the product and price. On Shopify, a checkout counts as abandoned after more than 10 minutes incomplete once the shopper gives an email, so the checkout reminder’s deadline starts there.
• A cadence. The same day after every theme, app, flow or discount change, every morning for prices and the pages behind live ads, and every week for each journey end to end.
• An owner for each fail. The email lead owns the flows, the ecommerce lead the codes and checkout, the performance lead the ad pages, and the owner of the product feed the AI answers.

AI agents for ecommerce: what AI customers do

An AI customer is an AI agent that becomes a real customer of a company, with its own inbox, phone number and browser, and keeps what the company sends and shows it as evidence. It receives what any new customer receives, at the store and at every rival. AI customers working for your company do 5 jobs for a store, each with a check, a pass mark and an action for that week.

Obsession makes the first AI customers that work for you: they sign up to any company you name and stay its customer for months, keeping every email, offer and price it sends, with proof. AI customers say they are AI whenever they talk to a person, never use fake payment details, and place a real order only on a card your company funds.

What changes in 2026 and 2027

• Live: AI channels on by default. Shopify’s agentic storefronts, which list a store’s products in ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta, are active by default for eligible stores.
• Live: a cart that watches prices. Google’s Universal Cart, launched in the US on 15 October 2026, saves products from several retailers, tracks price and stock changes and shows promo codes.
• Live: discount logic has moved. Shopify Scripts stopped running on 30 June 2026, replaced by Shopify Functions, so every custom discount once built on Scripts needs testing again.
• Announced: what a stop to a text covers. An FCC order released on 1 October 2026 lets a business treat a stop to an informational text, such as a shipping update, as covering only that kind of message, while a stop to a marketing text still ends all of its marketing texts, as Latham & Watkins explains. The order takes effect 30 days after it is published in the Federal Register. This is not legal advice.
• Forecast: more shoppers arrive from AI. Adobe reported that traffic from generative AI tools to US retail sites rose 693.4% in the 2025 holiday season, from a small base, and forecasts a further 130% rise year over year this season.
• Forecast: agents as buyers. McKinsey forecasts that by 2030 AI agents could orchestrate $900 billion to $1 trillion of US consumer retail revenue.

Each change adds a reader or a rule that a store’s own reports don’t follow, and each is checked the way a welcome email is: write down what the store’s pages say, act as a customer would, and compare. AI customers repeat that every week and after every change, so a wrong price in an AI answer or a dead code after a discount change is found the week it starts. The Shopify settings to review and a readiness test are in agentic commerce for merchants.

Shopify store monitoring options, compared

Obsession makes AI customers, so every other option is described from its provider’s own pages as of 24 October 2026.

When a store doesn’t need AI customers

• A store that rarely changes, runs no paid traffic and has 1 or 2 flows. A new address, a phone and a sheet, once a month and after every change, cover it.
• Uptime at every hour. A scripted monitor alerts within minutes when a page or the cart goes down.
• A brand that sells mainly through marketplaces or wholesale and sends no flows of its own. Its customers’ messages come from someone else’s systems.
• Last year’s rival campaigns. An AI customer holds what arrives from the day it joins, so earlier emails need an archive that already covered the brand.

Start this week

1. Open your email platform’s flow report and note the revenue a day of your welcome, browse, basket and checkout flows.
2. Write a pass mark for each: what arrives, by when and with what.
3. Run each journey once from a new email address and phone number, on your phone, and record what arrives against the pass marks.
4. Ask 3 AI assistants what your customers ask about price, delivery and returns, and compare the answers with your pages (what ChatGPT says about your store sets out the check).
5. Use the cost sum to set how often each check runs before your freeze, around 20 November.

Questions about AI agents for ecommerce