Machine Customer

Custobot · Non-human customer · Machine buyer · Autonomous customer

Definition

A machine customer is a non-human market participant — a device, program or AI agent — that selects and pays for goods or services itself, either under rules a person has set or autonomously.

What a machine customer is

Gartner defines a machine customer as a non-human economic actor that takes the place of a human customer and obtains goods or services in exchange for payment. The term was established by analysts Don Scheibenreif and Mark Raskino in the book “When Machines Become Customers” (2023), and in October 2023 Gartner put machine customers — also known as custobots — on its list of the top 10 strategic technology trends for 2024.

Early machine customers were not chatbots but devices: a printer on HP Instant Ink reorders its own ink, and with Amazon Dash Replenishment appliances reorder their own consumables — both are examples Gartner cites. Since 2025 they have been joined by LLM-based AI agents, including those built into agentic browsers: they take a task in plain words and compare offers from different stores themselves.

Gartner’s three stages

Stage	Who sets the rules	What the machine does	Example
Bound	The owner, strictly	Executes a fixed set of actions	A printer orders cartridges from a single supplier
Adaptable	A person sets the limits	Chooses an option and acts with minimal intervention	An agent compares offers and builds a basket within budget
Autonomous	The machine itself	Acts independently and runs the whole deal	Gartner places this stage further out in the future

For concrete scenarios, a more granular scale is handier — see agent autonomy levels.

What changes for retail

Banners stop working. A machine does not respond to promo badges or mood photography — it reads attributes, price and terms.
Decisions are made on data. Total price including delivery, availability, delivery time, ratings and reviews, return terms — everything that can be compared across stores.
Loyalty becomes a setting. Unless the person has specified a preferred brand or store, the agent goes wherever the terms are better.
Authority has to be verified. A store needs to know the agent is buying with a real person’s consent — hence payment mandates and agentic payment protocols.

Important: a machine customer does not cancel marketing, it moves it. The person decides which agent and which brands to trust; the agent decides where and what to buy. More on selling to agents in the B2A entry.

How to prepare

Audit product pages: attributes, sizes, compatibility and composition must exist as text, not only inside images.
Keep price, stock and delivery times in sync between the site and the feed — an agent may read a mismatch as unreliable data (see agent-ready product feed).
Move delivery, return and warranty terms into dedicated fields and structured markup instead of leaving them only in the terms of sale.
Give agents a machine-readable path to checkout — an API or an agentic protocol; the checklist is in the agent-readiness entry.
Track agent orders as a separate channel in your analytics so you can see their share and conversion.

Frequently asked questions

Is a machine customer the same thing as an AI shopping agent?

Not quite. In Gartner's terms a machine customer is a broader idea: it covers devices that reorder consumables, industrial equipment and procurement software. An LLM-based AI shopping agent is the most visible special case as of September 2026 — it understands a task in natural language and compares offers from different stores by itself.

Which stages of machine customer evolution does Gartner describe?

Three. A bound customer executes a strict set of actions within the rules its owner has set. An adaptable customer acts on a person's behalf and chooses an option with minimal intervention — but for selected tasks and under rules people define. An autonomous customer acts independently of people and runs the entire transaction process itself.

How much will machines actually buy?

There are no measurements yet, only analyst estimates. In 2023 Gartner forecast that by 2030 machine customers would be directly involved in, or have influence over, trillions of dollars in purchases. That is a forecast, so base decisions on your own data: what share of orders already comes from agents and how fast it grows.

What should a store do right now?

Clean up the data a machine decides on: complete product attributes, current prices and stock, delivery times and return terms in machine-readable form. The next step is to give agents a reliable way to place an order through an API or an agentic protocol, rather than making them click through the page layout.