Machine Customers: When AI Agents Buy Products and Services for Humans

Machine customers are non-human buyers: devices, software and AI agents that choose and pay for products or services on behalf of a person or business. Gartner analysts popularised the term. In 2026 they range from printers that reorder their own ink to shopping agents from Amazon, Google and others that buy within rules people set.

For brands, the shift is practical. When the buyer is software, the product data, price accuracy and policies an agent can read matter more than the emotional pitch. This guide explains Gartner's three stages, what machine customers can already do, what the data shows about adoption and trust, and how retailers, B2B suppliers and real estate businesses — including in the UAE — should prepare.

Key takeaways

• Machine customers already exist. Printers and appliances have reordered supplies for years; in 2025–26 Amazon and Google shipped AI agents that choose and buy within rules, and Shopify and Walmart opened sales channels inside AI chats.
• AI influences far more buying than it executes. Salesforce attributes 20% of 2025 holiday retail sales to AI and agents, but only 23% of US consumers trust AI to handle payments for them.
• The customer to persuade is the agent's evaluation. Complete product data, accurate price and stock, readable policies and linkable loyalty decide the shortlist — brand-supplied attributes appeared in half of relevant recommendations in a Google test.
• People still want the final say. 60% of US adults in a Visa survey wouldn't let AI spend any amount without their approval.
• Business buying may move first. Gartner forecasts that 90% of B2B buying will be intermediated by AI agents by 2028.

What are machine customers?

Machine customers are devices, software and AI agents that buy on someone's behalf — the buyer's decision is made, or at least prepared, by a machine working within rules its owner set.

Definition

Machine customer — a non-human economic actor that obtains goods or services in exchange for payment. Gartner, whose analysts Don Scheibenreif and Mark Raskino wrote the 2023 book When Machines Become Customers, also calls them "custobots".

They come in three broad forms: connected products (printers, cars, appliances) that replenish or book services; AI assistants that shop for consumers (Amazon's Alexa for Shopping, Google's Gemini and AI Mode, ChatGPT); and business agents that source and buy for companies. Gartner counted more than 9.7 billion installed IoT devices in 2023 — each one, in its analysts' words, a potential buyer.

Machine customers vs agentic commerce: what's the difference?

"Machine customer" describes the buyer; "agentic commerce" describes the market in which agents discover, compare, negotiate and pay. The plumbing — checkout protocols, agent payment tokens, proof of what a human approved — is covered in the guide to agent-to-agent transactions. This article looks at the other side: what changes for the businesses selling to them.

What are Gartner's three stages of machine customers?

Gartner describes three stages: bound customers that execute rules people set, adaptable customers that use AI to choose and act with minimal intervention, and autonomous customers that act independently with wide discretion.

Stage	Who decides	What the machine does	Examples	Status, September 2026
Bound	People set the rules	Executes them inside a closed ecosystem	HP Instant Ink printers ordering cartridges; Amazon Dash Replenishment devices	Mature
Adaptable	People set the goals and limits	AI chooses and acts for selected tasks, usually with a confirmation step	Google's agentic checkout; Amazon Buy for Me; Rufus auto-buy at a set price; Alexa for Shopping's scheduled restocks	Live, mostly in the US
Autonomous	The machine, within broad discretion	Handles most steps of a transaction itself	Agent payment standards for purchases with no human present; procurement agents	Emerging

McKinsey draws a finer version of the same picture: a six-level "automation curve" running from subscriptions and scheduled refills at Level 0 to agent-to-agent "networked autonomy" at Level 5. Its key point is that consumers delegate up to a ceiling that depends on the category — price, emotional weight and the cost of regret — so higher levels aren't always the goal.

Why are machine customers arriving now?

Because in 2025 and 2026 the biggest retail and technology platforms turned AI shopping assistants into buyers, and discovery itself moved into AI.

• Buying agents shipped. Amazon's Buy for Me (beta, April 2025) completes purchases on other brands' websites for the customer; its Rufus assistant added auto-buy at a set price in November 2025, and in May 2026 Amazon merged Rufus and Alexa+ into Alexa for Shopping, which can schedule routine restocks. Google's agentic checkout (November 2025) buys a tracked item when its price falls within your budget — after you confirm.
• Sellers got agent channels. Shopify's Agentic Storefronts put merchants' catalogues into ChatGPT, Microsoft Copilot and Perplexity; from March 2026 Shopify said millions of merchants could sell in AI chats while remaining merchant of record. Google's Universal Commerce Protocol, launched in January 2026, now powers direct checkout for hundreds of thousands of brands and retailers, Google says.
• Discovery moved. Adobe measured a 693.4% jump in traffic from generative AI tools to US retail sites in the 2025 holiday season, and Salesforce reported in July 2026 that agentic search as a shopping starting point had grown 200% in a year.

Gartner saw the scale coming: in 2023 it predicted 15 billion connected products with the potential to behave as customers by 2028, generating trillions of dollars in revenue by 2030. Treat that as a forecast, not a measurement.

How do machine customers decide what to buy?

They compare structured data against the rules their owner set — price, specification, availability, delivery, returns, ratings and loyalty benefits — rather than responding to imagery, urgency or persuasion.

Gartner's analysts put it bluntly: machines don't need to be delighted, and you can't win their loyalty over dinner. A machine commits to the supplier whose sales and fulfilment process works reliably. Amazon describes its shopping assistant using reviews, price, availability, delivery speed, return rates and the customer's history — all data, no mood. And McKinsey's advice to retailers for the first level of automation is simply that verifiable data beats marketing gloss.

A human shopper responds to	A machine customer checks
Brand story and imagery	Structured attributes and specifications
Promotions and urgency	Actual total price, including delivery
The store or site experience	Stock, delivery date and returns policy
A few reviews, read selectively	Aggregated ratings and return rates
How loyal they feel	Loyalty benefits it can actually apply

Common misconception

"Machines are perfectly rational buyers." They're data-driven, not flawless. In Microsoft's Magentic Marketplace simulation, every AI model tested favoured the first acceptable offer it received, and some could be manipulated into paying a dishonest seller. Speed of response and clean, verifiable data both win business — and platforms, not only sellers, can exploit agent biases. The evidence is summarised in the agent-to-agent transactions guide.

What does the latest data show?

The data shows AI shaping a large and fast-growing share of shopping decisions, while the share of purchases machines complete on their own is still small and trust is the brake.

What the data shows

• Traffic (Adobe Analytics, 2025 US holiday season): traffic from generative AI tools to retail sites rose 693.4% year on year; online spending hit a record $257.8 billion. — Adobe
• Influence (Salesforce platform data, 2025 holiday season): AI and agents drove 20% of retail sales, or $262 billion, through recommendations and engagement; retailers running their own agents grew sales 59% faster. — Salesforce
• Discovery (July 2026): agentic search as a first step in shopping grew 200% year on year; only 28% of commerce organisations use agentic AI, with another 44% planning to within six months. — Salesforce
• Reach (vendor figures): Amazon says its Rufus assistant helped more than 300 million customers in 2025. — Amazon
• Comfort (survey, 2,000 US adults, fielded January–February 2026): 58% are comfortable with AI comparing prices, 38% with AI completing a purchase, 27% with AI spending without limits — and 60% wouldn't allow any AI spending without approval. — Visa
• Trust (survey, 2,065 US adults, May 2026): 72% have used an AI assistant; 23% trust generative AI to handle payments on their behalf. — Visa
• Europe (survey, 749 respondents in France, Germany and the UK, December 2025): 84% use AI tools; 38% use them to research or decide purchases. — McKinsey

What this means

The machine customer is already choosing your shortlist, even when a person still presses "buy". McKinsey's European research puts it as decision influence arriving before execution. For brands, that means the battle for AI-assisted discovery matters now, while the fight over autonomous checkout plays out more slowly.

How are machine customers changing retail and brand marketing?

They move the first impression from your website to a machine's evaluation of your data, which changes where brands must compete.

Discovery happens in someone else's interface. When a shopper asks ChatGPT, Gemini or Alexa for a recommendation, your product page may never load. OpenAI pulled back from in-chat checkout in March 2026 but expanded its Agentic Commerce Protocol so merchants can send product feeds and promotions into ChatGPT. Shopify, Google and Amazon all now run agent channels. The AI search guide covers how discovery is shifting across these platforms.

Your data is your shelf position. Google says merchants that adopt its core Merchant Center feed practices see about 5% more conversions the following month. In a test with lululemon, brand-supplied "conversational attributes" were used in 50% of relevant product recommendations in AI Mode.

Loyalty becomes something an agent can apply. Google now lets merchants connect loyalty data so member prices and perks show up across its surfaces, and Walmart's ChatGPT app supports account linking and loyalty. A loyalty programme a machine can't see is invisible at the moment of choice.

Attribution changes. Traffic arrives from AI referrals — or doesn't arrive at all when the agent buys through a protocol. Measure AI referral traffic, share of voice in AI answers and conversion from AI-sourced visits; the generative engine optimization guide explains how to earn citations in AI answers.

Common misconception

"AI agents bought $262 billion of holiday goods in 2025." That's Salesforce's figure for sales influenced by AI and agents, including retailers' own recommendation engines and service bots — not purchases machines completed on their own. Adobe, which measures traffic rather than sales, notes the AI-referral user base is still modest.

Will machine customers take over business procurement?

Gartner thinks business buying will move fastest: it forecasts that by 2028, 90% of B2B buying will be intermediated by AI agents, putting more than $15 trillion of spend through AI agent exchanges.

The tools are arriving. SAP's next-generation Ariba, available since March 2026, embeds its Joule assistant and a Bid Analysis Agent that evaluates complex bids, including total cost, with further capabilities rolling out through 2027. As far back as 2022, Gartner reported that executives believed at least 25% of consumer purchases and business replenishment requests would be substantially delegated to machines by 2030 — an expectation, not a measurement.

McKinsey argues that autonomy advances more slowly in business buying, because authority flows through procurement policies, budget owners and legal teams — but once those policies can be encoded, it scales further than in consumer shopping. Competition then shifts from unit price to predictability, transparency and compliance with the buyer's rules. For suppliers, being "easy to buy from by software" — catalogue APIs, e-procurement integrations, clear service levels — becomes a sales capability.

What are the risks of selling to machine customers?

The main risks are errors at scale, manipulation, platform gatekeeping and consumer-law gaps — plus the danger of optimising for a customer that still isn't buying much.

• Errors multiply. A wrong size or an out-of-date price reaches every agent at once. Clear confirmation steps and easy returns protect both sides — and in US law, a consumer can undo a mistake made with a seller's automated system that offered no chance to correct it.
• Agents can be gamed. First-offer bias and prompt injection mean a less scrupulous competitor, or platform, can tilt an agent's choice. Brands need verified data and monitoring of how agents present them.
• Gatekeepers decide who's visible. Amazon runs its own buying agents while blocking many outside AI agents in its robots.txt file. Each platform sets the terms of its agent channel.
• Consumer law assumes a human. A July 2026 paper from Brussels think tank CERRE argued EU consumer law's disclosure and consent rules weren't built for agents.
• Forecasts outrun outcomes. In 2023 Gartner predicted that machine customers would generate 20% of inbound customer-service contacts by 2026; no published data yet shows whether that happened.

What do machine customers mean for the UAE?

UAE businesses are meeting machine customers first through global AI platforms and a handful of local payment pilots, so the priority is making products and listings readable by those agents.

• Payment pilots. Mastercard's first Agent Pay transaction outside the US was a VOX cinema ticket in Dubai in November 2025, and Aldar customers pay service charges through an AI agent on Visa's rails. Nine banks and fintechs — including Emirates NBD, ADCB, Mashreq, Wio and Tabby — joined Visa's Agentic Ready testing programme in May 2026. The details are in the agent-to-agent transactions guide.
• Business interest. Research by Fast Company Middle East with Visa found that nearly 60% of UAE businesses have strong interest in, or are actively exploring, agentic commerce; the announcement didn't give a sample size.
• Local platforms in AI chats. Bayut became the first UAE-based real estate platform with an app on ChatGPT in June 2026, letting users search listings by describing what they want.

For UAE retailers and service brands the checklist is familiar, with local twists: bilingual Arabic and English product data, prices shown the way UAE shoppers pay them, realistic delivery promises and clear returns policies — the details an agent compares first.

How do machine customers affect real estate?

In property, machine customers act first as discovery agents and payment agents — shortlisting listings and paying recurring charges — while the purchase decision stays human.

• AI search shortlists listings. Zillow launched as an app inside ChatGPT in October 2025, surfacing listings on a map when users discuss buying a home; Bayut followed in the UAE in June 2026. The agent compares what's in the data — price, size, service charges, handover date, payment plan — so incomplete listings drop out. The AI real estate marketing guide covers listing quality and Dubai's advertising permit rules.
• Speed still wins. If buyer agents favour the first acceptable response, a brokerage's response time matters even more; AI lead routing is how brokerages get it under a few minutes.
• Recurring payments go first. Aldar's service-charge agent shows the pattern: frequent, rule-based, reversible payments are what property owners delegate first.

How should brands prepare for machine customers?

Prepare by making your offer complete, accurate and readable by machines, joining the agent channels your customers use, and measuring AI-driven demand separately.

1. Audit product data. Complete attributes, identifiers, sizes, compatibility and specifications — consistent across every channel.
2. Keep price, stock and delivery accurate in real time. Agents punish mismatches by dropping you from the shortlist.
3. Publish policies agents can read. Returns, warranty, shipping and cancellation terms, in structured form.
4. Join the agent channels. Google Merchant Center and the Universal Commerce Protocol, ChatGPT product feeds through the Agentic Commerce Protocol, Shopify's Agentic Storefronts or marketplace APIs.
5. Make loyalty linkable. Connect loyalty data so agents can apply member prices and perks.
6. Decide your agent policy. Which agents you allow, how you verify them and how agent-placed orders are confirmed — see agent-to-agent transactions.
7. Design for confirmation and returns. Agents make mistakes; easy correction keeps customers and keeps you on the right side of consumer law.
8. Measure the AI channel. Track AI referral traffic, share of voice in AI answers and conversion from AI-sourced visits, alongside classic search.

Where to start depends on what you sell:

Business	Machine customers you'll meet first	First move
Online retailer	Google AI Mode and Gemini, ChatGPT, Alexa for Shopping, Perplexity	Clean feeds, agent checkout channels, clear returns
Consumables brand	Subscriptions, scheduled restocks, connected devices	Replenishment programmes and stable product listings
B2B supplier	Procurement agents inside e-procurement suites	Catalogue APIs, punchout integrations, service levels
Services (travel, dining, clinics)	Booking agents	Real-time availability and clear cancellation rules
Real estate	AI search apps and payment agents	Complete listing data, fast responses, payment integrations

What is likely to happen next?

Expect "agent proposes, human confirms" to remain the norm for consumers through 2027, business procurement to automate faster, and platform agent channels to multiply.

• The 2026 holiday season is a test. Visa predicted that millions of consumers would use AI agents to complete purchases by the 2026 holidays, and Google has pitched agentic checkout and its loyalty tools to retailers for the season.
• Trust determines the pace. Visa's research found confidence rises when a known payment brand is involved — payment networks will push agent tokens and verified-agent programmes accordingly.
• Data quality becomes a marketing discipline. Feeds, attributes and policies move from operations into the marketing plan, because they now decide visibility.
• Forecasts will be tested. Gartner's 2028 figures on connected products and B2B buying are the benchmarks to watch against real data.

For how AI agents create value inside a business, see agentic AI for business.

Final takeaway

Machine customers are here in their early forms — auto-replenishing devices, shopping assistants that buy after a tap, procurement agents that evaluate bids — and they already shape a fifth of holiday retail through recommendations. They don't respond to charm; they respond to complete, accurate, verifiable data and reliable fulfilment. Brands that make their products easy for software to find, compare, trust and buy will win the shortlist long before autonomous checkout becomes normal.