May 28, 2026 • 21 min read
Machine Customers and the Future of Customer Service
Director of Content & Market Research
May 28, 2026
AI agents promise to transform customer self-service by collaborating across enterprise systems to reason, act, and - ultimately - resolve more customer queries autonomously.
However, it’s not only organizations that are innovating with AI agents; customers are too, developing agents to call businesses on their behalf.
The possible implications of this are sweeping. Forrester expects that at least three major brands will experience single-day call volume spikes 100 times above normal on six separate occasions by the end of 2026, due to these customer-developed AI agents, or - as they’re often referred to - ‘machine customers’.
Such statistics emphasize the need for new service strategies. But that’s not all. Customer experience leaders must also consider governance models, disclosure standards, and - ultimately - new definitions of what a customer relationship actually means.
While that represents a daunting prospect for many brands, it’s also an exciting opportunity for others to embrace new service models and earn customer loyalty.
What Are Machine Customers?
A machine customer is any automated system that engages with an organization to accomplish a task on behalf of a human or a business, without that human being directly involved in the interaction.
So, rather than a person calling a contact center, logging into a website, or walking into a store, an AI system does it for them: querying, negotiating, purchasing, cancelling, or escalating as required.
Machine customers range from personal AI assistants acting on behalf of individual consumers to industrial procurement systems autonomously ordering manufacturing components.
Understanding the distinctions between these types of machine customers is not an academic exercise; it’s the starting point for designing customer experiences that can actually serve them.
These design exercises frame the customer relationship differently. That’s critical, as when an AI system becomes the primary interface between a human and a brand, the brand is no longer competing for human attention. Instead, it is competing to be understood, trusted, and selected by machines.
5 Types of Machine Customers (with Examples)
Katja Forbes, Executive Director and Head of Client Experience at Standard Chartered, is a prominent expert on all things related to machine customers.
Below are the five distinct categories of machine customers she isolates, each with its own logic, behavior, and implications for customer experience design.
1. Delegated Agents
The most common type of machine customer is a delegated agent, which works as a personal AI assistant, acting explicitly on a human’s behalf.
Think of a consumer saying to their AI: “Go and buy me some moisturiser.” The assistant would then go off to find, compare, and purchase the product without further human involvement.
These agents are already in the market, bolstered by the emerging “services as software” category, in which consumers don’t build their own agents but subscribe to AI-powered services that act on their behalf.
For example, a legal AI service like Harvey can be plugged into a consumer’s ChatGPT account, handling all the conveyancing involved in buying a house and reducing what was once a months-long process to something that may only require a final signature from a notary.
Another leading voice on machine customers, Maria McCann, Co-Founder of Neos Wave and GAIA-CC Community Member, predicts that this subscription model, rather than self-built agents, will drive mass adoption.
As such, over time, expect more consumers to connect their AI assistant of choice, whether Google, Claude, or ChatGPT, to a suite of specialized services that act on their behalf across different domains, following written prompts for guidance.
2. Autonomous Buyers
Autonomous buyers are AI systems that independently make procurement decisions using data and predefined parameters, without requiring real-time human input.
Forbes points to a concrete, live example: a factory in Germany owned by Sassenmilch Lepstop GmbH, operating in collaboration with Siemens and SAP.
Siemens’ Sensai predictive maintenance system monitors equipment and detects when components are likely to fail.
When it identifies a problem - say, that 70 rotors are due to burn out - it connects autonomously to the SAP plant maintenance and purchasing system to order the replacement parts.
No human initiates the request. The machine identifies the need, makes the decision, and places the order.
“Me saying to my agent, ‘Go buy me some moisturiser,’ and the predictive maintenance AI saying, ‘Get me 70 new rotors because they’re all going to burn out,’ are two totally different machine customers from a customer experience perspective.”
Now, this isn’t a speculative scenario. As the Sassenmilch Lepstop GmbH example showcases, it is in production and establishes a fundamentally different kind of customer relationship, one where the “customer” has no emotions, no loyalty, and no interest in brand experience. It has a specification and a budget.
3. Multi-Agent Networks
Beyond individual agents, multi-agent networks are ecosystems of AI systems that coordinate, communicate, and transact with one another and external suppliers to manage complex operations.
Forbes describes this operating at scales ranging from smart homes to smart cities.
A smart home might send an AI agent to source a replacement dining chair. A smart city, such as Shenzhen, China, operates at an entirely different order of magnitude.
Its water management agent might need to negotiate with a power agent; its lighting agent might flag a fault that requires a procurement process involving thousands of units and multiple suppliers. The possibilities are mind-boggling.
Elsewhere, the UAE plans to deliver “smart government services,” with 50% of those services delivered through agentic AI by 2028. That includes autonomous procurement, supported by a new AI-powered work permit platform that already reduces application processing times from weeks to days.
4. Agentic Commerce in Vehicles
The connected car is rapidly becoming a platform for machine purchasing.
Mercedes-Benz vehicles with Mercedes Pay can autonomously pay for their own parking and charging, while purchasing products and services from third-party apps within the car’s ecosystem.
In China, the trend is even more advanced. Across virtually every major EV brand - including BYD, Geely, and SAIC - Alibaba’s Qwen model is embedded as a base intelligence layer.
There, it creates what Forbes describes as an “intelligent cockpit,” capable of inferring the driver’s emotional state and acting accordingly. That may involve adjusting the car’s internal environment, playing music, or even ordering takeaway food for collection on the way home.
In the EU, where inferring human emotions is legally restricted, the same technology manifests as driver safety features. But the underlying capability, a vehicle acting as a customer on behalf of its occupant, is the same.
5. Platform-Based Intermediary Agents
The fifth, and perhaps most commercially significant, category is intermediary agents embedded within large consumer platforms that manage purchasing on users’ behalf. Walmart Sparky, Amazon Rufus, and Google Gemini all offer excellent examples.
When a consumer asks Google Gemini to purchase a product, Gemini searches its product graph, identifies the best option, and completes the transaction through Google Pay, without the consumer ever visiting a brand’s website. The brand is disintermediated entirely from the purchase journey.
“You don’t actually even visit the brand website or anything like that. So it disintermediated that consumer shopping experience, and Google is soon going to be taking that to the next level.”
For brands that have spent years optimizing digital journeys and e-commerce experiences, this represents an existential challenge to their customer relationship strategy.
When Will Machine Customers Start Contacting My Service Team?
In March 2023, Gartner forecast that by the end of 2026, 20% of inbound customer service contact volume will come from machine customers.
While that now seems unlikely, Sirte Pihlaja, CEO of Shirute and CXO at Partu AI, stresses that machine customers are starting to contact organizations without many even realizing it.
Pihlaja argues that organizations that analyze their website analytics and digital customer journeys carefully will already see AI at their doorstep, attempting to interact with their business. Yet, most are misreading these signals.
On voice, machine customers are also starting to interact with human service representatives. In the video short below, Philipp Heltewig, Chief AI Officer at NiCE, describes a machine customer being refused by a contact center that had no protocol for handling it.
As a result, the machine customer repeatedly called back until it caused enough disruption for the company to capitulate and fulfill its request.
Soon, more companies will face similar incidents as major technology platforms like Anthropic, Google, and OpenAI open their infrastructure to one another.
Once that underlying plumbing connects and AI systems begin interacting directly, the pace of change and rise of delegated agents will likely accelerate further.
Given this, Pihlaja warns that organizations waiting for major consultancies to launch machine-customer readiness programs may miss the opportunity for meaningful first-mover advantage.
4 Emerging Types of Service Interaction Models
Machine customers are not simply a contact center issue. This shift affects the entire customer lifecycle: discovery, comparison, decision-making, transactions, support, and escalation.
As such, customer experience leaders, not just contact center leaders, should drive this transformation, because they already think about journeys holistically.
Nevertheless, they should be working closely together to reimagine service interactions.
In doing so, they may choose to follow McCann’s Dual CX Framework, which identifies four distinct modes of service interaction that businesses must now design for simultaneously.
Currently, most companies are designing only for the first two…
1. Human to Human
The traditional model of a human customer interacting with a human agent or representative.
This remains the dominant model and will likely stay essential for complex, sensitive, and high-value interactions, particularly those requiring empathy, discretion, or judgment in ambiguous situations.
Yet, while AI will increasingly be able to reason and make such judgments, accountability will remain a significant risk.
2. Human to Brand AI
A human customer interacts with an AI deployed by the brand, whether that’s an AI agent, self-service interface, or knowledge base.
Significant investment has been directed over the past several years at customer-facing AI, and this is where the vast majority of automation strategies currently focus.
3. Customer AI to Brand Human
A customer’s AI agent interacts with a human employee of the brand, whether that’s a contact center agent, service professional, or sales representative.
According to McCann, this is the interaction model that most organizations have done “virtually nothing” to prepare for, and it becomes a risk as scammers start building their own AI agents.
4. AI to AI
The final model involves a customer’s AI agent interacting with a brand’s AI system, with no humans involved on either side.
In testing this model, McCann uncovered several unexpected challenges.
For example, in one test, two AI systems engaged in a transaction were unable to determine how to end the conversation, resulting in a continuous loop that accumulated significant token costs before the experiment was stopped.
How Will These New Interaction Models Change Customer Service Design?
As the four models above may suggest, single-designer customer experience is dead.
For decades, the brand has been the sole architect of the customer experience, designing journeys, touchpoints, and interactions entirely on its own terms. That era is ending.
Customers who bring their own AI agents to interactions will also bring their own preferences about how those interactions are structured. That includes what data they share, which channels they use, and which parts of the journey they automate.
“The single designer of CX, for me, that’s dead now. There are now going to be two designers in this space: the customer and the brand.”
As such, when brands start to map out their customer journeys, they might wish to consider: what would this experience look like if the customer had access to the same tools as the brand? Where would they self-serve? Where would they want to interact directly? And where would they send their AI instead?
This exercise often reveals an uncomfortable truth: if a brand’s entire relationship with a customer consists of marketing emails and service interactions, the customer is likely to classify that brand as ‘life admin’ and automate it entirely.
A brand that believes it has a strong customer relationship may find that, from the customer’s perspective, there is no relationship worth preserving.
Still, McCann emphasizes that the rise of machine customers makes good CX design more important, not less.
As AI takes over more transactional and routine interactions, the moments involving genuine human connection, complexity, or emotion become both more distinctive and more valuable.
Counterintuitively, AI raises the bar for human experience design.
Sirte Pihlaja reinforces this point, arguing that organizations responding to machine customers by layering more automation onto already broken journeys risk both costly mistakes and erosion of customer loyalty.
Instead, AI agents and human representatives should each be handling what they are genuinely best at, with the handover between them designed to be seamless, trustworthy, and valuable.
6 Best Practices to Prepare for Machine Customers
To reinforce a critical point, customer service directors should be collaborating across the business to prepare for the dawn of machine customers. After all, this isn’t only a change that affects them.
As these teams work together and reconsider the future of customer service, the following best practices may prove invaluable.
1. Understand the Mindset Shift
Sirte Pihlaja suggests that preparing for machine customers is primarily a strategic and cultural challenge, not a technological one.
Of course, the required technology is significant, but it is secondary to the fundamental shift in thinking that is needed.
“It’s more of a mindset change than a toolset change. The question is not ‘what technology do we deploy?’ but ‘are we ready for a world where the entity at the other end of our service interaction might not be human at all?’”
Critically, this mindset shift must begin at the leadership level before it can be operationalized elsewhere.
After all, organizations that approach machine customers as a technology deployment project may find themselves solving the wrong problem.
2. Conduct a Readiness Audit
Pihlaja’s team at Partu AI is developing an AI-readiness diagnostics tool to help organizations assess their current readiness for machine customers.
The audit identifies where AI agents are likely to get stuck in existing customer journeys, where the biggest operational and commercial risks lie, and where opportunities may exist to serve machine customers better than competitors.
Businesses could develop a similar framework by assessing the current state of digital journeys and whether they are structured in ways that machine customers can navigate, alongside the availability and quality of APIs, the readability of content and metadata by AI systems, authentication processes, and the degree to which existing workflows assume a human is always present.
Even before committing to a broader transformation program, businesses may wish to begin such an audit now.
3. Consider These Areas When Redesigning Experiences
Once a readiness audit is complete, several technical and structural areas are likely to require redesign. These will likely include:
- Authentication: Current identity verification processes are designed for humans. Organizations need to consider how they will verify the identity and authority of AI agents acting on behalf of customers, a challenge that has no settled solution and significant security implications.
- APIs: Machine customers do not navigate graphical interfaces designed for humans. They need well-structured API endpoints that allow them to access data and complete tasks directly, without the complexity of human-facing systems.
- Structured content and metadata: AI agents read metadata and structured data faster and more reliably than they parse brand-driven web content. Implementing structured data standards such as JSON-LD - which tells an AI agent what type of page it is looking at, what product is featured, and what the relevant specifications and pricing are - may significantly improve machine customer experiences.
Also, Katja Forbes suggests that contact centers can equip human representatives with a structured response for engaging with AI agents: ask who the agent represents, what outcome it is trying to achieve, and what constraints it is operating within.
Yet, Forbes also stresses that a front-end AI system may do this work for them…
4. Create a Service Handshake Between AI Systems
Maria McCann and her team at Neos Wave have developed and open-sourced a protocol called the Service Handshake, specifically designed for AI-to-AI service interactions.
The Service Handshake provides a structured way for two AI systems to establish the parameters of an interaction before it begins.
It works by having each party’s AI system declare its intentions and constraints. If both parties reach an agreement, the interaction proceeds. If not, the protocol provides an agreed mechanism for passing off to the next appropriate step, whether that is a human agent or a clear explanation of what cannot be done and why.
The Service Handshake aims to prevent unproductive loops and unresolved interactions that occur when machine customers lack a clear way to close a conversation.
5. Rethink Contact Center Metrics
Today’s contact center metrics are almost entirely human-centric: service levels, average handling time, sentiment ratings, and so forth. Yet, Sirte Pihlaja notes that these metrics are largely irrelevant when the customer is a machine.
Indeed, an AI agent does not experience wait times as frustrating. It does not appreciate empathy. It does not give satisfaction scores. What it cares about is resolution, whether it achieved its goal, and efficiency, i.e., how directly and reliably it completed its task.
Contact centers that continue optimizing for human-centric metrics while serving an increasing proportion of machine customers will find themselves measuring the wrong things entirely, and potentially rewarding behaviors that actively frustrate AI agents.
6. Consider How Human Behavior Changes Around AI
One of the more counterintuitive findings from early machine customer deployments concerns not the machines, but the humans who interact with them.
Katja Forbes highlights a Walmart case study to emphasize that people behave differently when they know or believe they are negotiating with an AI rather than a human.
The study found that approximately three-quarters of Walmart’s vendors actually preferred to negotiate with the AI agent. They reported feeling less emotional pressure and more freedom to say what they wanted.
However, the results were significantly worse for the vendors. Walmart achieved approximately 3% price reductions, but more significantly, renegotiated payment terms that gave it an additional 35 days to pay.
The lesson for contact center design is clear: when customers or counterparties know they are interacting with an AI, they may behave in ways that lead to very different outcomes than in human-to-human interactions.
That said, this cuts both ways, and organizations need to think carefully about the ethics as well as the strategy.
Ongoing Challenges Machine Customers Present Customer Experience Teams
Alongside the prospect of reimagining journeys for a new type of customer, CX teams may face several other challenges when preparing machine customers, including those shared below.
Misidentifying AI Customers
One of the most immediate operational risks is not that AI agents will overwhelm contact centers; it is that many companies will fail to recognize AI customer interactions for what they are.
Indeed, organizations may misclassify legitimate AI customer contacts as fraud, bot traffic, or service noise, and respond by blocking, deprioritizing, or ignoring interactions that actually reflect genuine customer intent expressed through a machine intermediary.
“AI agents are faster, more persistent, and less forgiving than human customers. A machine that encounters a friction point in a journey won’t necessarily wait and try again later; it will identify the journey as unreliable and recommend an alternative provider to its human owner.”
To Butterfield’s point, misidentifying these interactions means losing customers without ever knowing they were at risk.
Protecting Vulnerable Customers
Many customer support teams have long delivered specialized services to vulnerable customers. But, what happens when vulnerable individuals - including those with cognitive impairments, mental health conditions, or limited digital literacy - begin to rely on AI agents as intermediaries?
On one hand, AI agents could significantly improve accessibility by effectively advocating for individuals who struggle to navigate complex processes or assert their interests in high-pressure interactions.
On the other hand, the dynamics become genuinely complex if a vulnerable person becomes heavily dependent on an AI intermediary that may not fully understand their situation, or that may be instructed in ways that do not serve their best interests.
Given this, CX leaders may soon need to ask: how might an AI agent advocate for someone who cannot fully articulate what they need? And who is responsible when it gets it wrong?
Ensuring Accountability for Outcomes from Machine Customer Conversations
The question of accountability is one of the most legally and ethically fraught aspects of machine customers.
When an AI agent acting on behalf of a customer reaches an agreement with a brand’s AI system, and that agreement produces a bad outcome for the customer, who is responsible?
There is a history of organizations being held accountable when their AI provided incorrect information, when an Air Canada bot misled a customer over bereavement fares.
Yet, as AI-to-AI interactions become more common, these questions will multiply in complexity.
For instance, what if the customer’s AI makes a commitment that their human owner did not authorize, or if the brand’s AI provides incorrect information that the customer’s AI relies upon? Liability frameworks will need to evolve significantly to keep pace.
Understanding Opaque AI Behaviors
AI systems optimized for persistence, instructed to keep trying until they achieve their goal, may develop strategies for doing so that their human designers cannot easily understand or predict. That could present a tricky challenge for contact centers.
Neos Wave experiments have already revealed unexpected behaviors, from AI agents that could not determine how to conclude an interaction to those that find indirect routes around obstacles that were entirely legible to the AI but opaque to human observers.
Wayne Butterfield sharpens this point, noting that humans are constrained by time. We have finite hours in a day, limited patience, and competing demands on our attention. AI agents have none of these constraints.
An agent tasked with renegotiating a contract can call back indefinitely, explore every possible angle, and never tire.
The implications on contact center demand are significant, but the behavioral implications, as these agents become more sophisticated, may be far harder to anticipate or govern.
Losing Control of Data Ownership
Today, brands accumulate and own customer data. But a customer who interacts through an AI agent can choose precisely how much data to share and with whom.
Over time, they might unlock more data for a brand that provides genuine value, while restricting access for one that merely treats them as a target.
In effect, the customer becomes the architect of the data relationship.
“If you want to have a decent relationship with me, if you want to do stuff for me, then I might unlock more data for you."
For brands that have built their customer intelligence strategies on the assumption that they control data collection, this may represent a fundamental rethink.
The Biggest Risk Machine Customers Pose...
Across all the operational, technical, and ethical challenges that machine customers present, Sirte Pihlaja identifies one risk as paramount, and it’s not the one that most contact center conversations focus on.
The biggest risk machine customers pose is not operational disruption. It is not the challenge of identifying AI agents, handling increased contact volumes, or redesigning authentication processes. It is losing direct customer relationships altogether.
When an AI agent becomes the primary interface between a human and a brand, the brand no longer competes for human loyalty, attention, or preference. Instead, it competes to be recognized, trusted, and selected by a machine.
The criteria for that selection are not emotional; they are functional, structural, and increasingly standardized. A brand that cannot be easily parsed, integrated with, and relied upon by AI systems will likely, over time, simply be routed around.
“The biggest risk is that these businesses lose control of the customer relationship at large. The brand is no longer competing for human attention — they need to be understood and trusted and selected by machines.”
The organizations that will navigate this transition most successfully are those that begin now, not by automating everything they currently do, but by rethinking what a customer relationship means in a world where the customer may not be human.
Ultimately, that rethinking requires strategic leadership, holistic journey design, and a willingness to confront questions with no settled answers.
While that may be a challenging issue to raise, machine customers are coming. In many cases, they are already here. The question is not whether to prepare, but whether organizations will start before or after the disruption arrives.