The Automation Curve in Agentic Commerce

Not every purchase will automate at the same speed. That statement sounds obvious, but it is routinely ignored in conversations about agentic commerce. Leadership teams tend to react in one of two ways: either they treat AI-mediated buying as something that will transform everything simultaneously, or they dismiss it as irrelevant to their particular market. Both responses miss what is actually happening.

The reality is more nuanced. Different purchasing categories are moving through stages of AI automation at different speeds, shaped by a set of predictable factors. Some categories are already well into the shift. Others will not meaningfully automate for years. The pattern of which categories move first and which move last follows a logic that, once you see it, becomes a useful tool for strategic planning.

That pattern is what I call the automation curve.

What the automation curve describes

The automation curve is a way of mapping how quickly different types of purchases will move from human-led processes to AI-mediated or AI-executed processes. It is not a single timeline. It is a set of trajectories, each shaped by the characteristics of the purchase category itself.

At one end of the curve sit purchases that are already automating: low-value, high-frequency, well-structured transactions where the information needed to make a good decision is readily available and the cost of a bad decision is low. Office supplies. Standard software licences. Commodity raw materials. For these categories, AI agents can handle research, comparison, selection, and even execution with minimal human oversight.

At the other end sit purchases that will resist automation for much longer: high-value, complex, relationship-dependent transactions where information is fragmented, risk is significant, and regulatory or compliance requirements add layers of scrutiny. Enterprise software platforms. Major construction contracts. Strategic consulting engagements. For these categories, AI agents will play a supporting role for the foreseeable future, but the decision itself will remain firmly with humans.

Between those two ends lies a wide middle ground where the pace of automation depends on specific conditions. Understanding those conditions is what makes the curve useful.

The five factors that shape the curve

Five factors determine where a given purchase category sits on the automation curve and how quickly it will move along it. These are not theoretical constructs. They reflect the practical realities that make some purchases easy for AI agents to handle and others genuinely difficult.

1. Transaction value

Lower-value purchases automate faster because the consequences of a suboptimal decision are bounded. If an AI agent selects a slightly more expensive stationery supplier, the financial impact is negligible. If it recommends the wrong enterprise resource planning system, the impact could run into millions. Transaction value directly affects how much human oversight organisations require, which in turn determines how much of the process an agent can own.

2. Decision complexity

Simple, well-defined decisions automate more readily than multi-dimensional ones. Purchasing printer toner involves a handful of variables: compatibility, price, delivery time. Selecting a logistics partner for a multi-country distribution network involves dozens of interdependent variables, many of which require contextual judgement. AI agents handle the first type well today. The second type still requires human synthesis, particularly where trade-offs are subjective or organisation-specific.

3. Information structure

Purchases where the relevant information is publicly available, well-structured, and machine-readable automate faster than those where information is fragmented, gated, or ambiguous. If an AI agent can access clear pricing, documented specifications, verified credentials, and independent reviews, it can do its job effectively. If the information it needs is buried in PDFs, locked behind sales conversations, or described only in vague marketing language, the agent's ability to evaluate is severely constrained. This factor is one that businesses can directly influence, which is why it matters so much for commercial strategy. It also connects to the broader challenge of AI-mediated discovery.

4. Risk tolerance

Every organisation has a threshold for how much decision-making authority it will delegate to automated systems. That threshold varies by purchase category. Most organisations are comfortable with AI agents managing routine supply orders autonomously. Far fewer are comfortable with an agent selecting a cybersecurity vendor or a legal services provider without significant human involvement. Risk tolerance is partly rational, based on actual exposure, and partly cultural, based on an organisation's comfort with technology-mediated decisions. Both dimensions affect the pace of automation.

5. Regulatory constraints

Some purchase categories operate within regulatory frameworks that require specific human approvals, documented decision rationales, or particular procurement processes. Government procurement, healthcare purchasing, and financial services all have regulatory requirements that constrain how much can be delegated to AI agents, regardless of the technology's capability. These constraints will evolve over time, but they currently act as a meaningful brake on automation in certain sectors.

The automation curve framework

Bringing these factors together, purchase categories can be grouped into four zones on the automation curve. This framework is designed to be a practical tool for leadership teams assessing where to focus their attention and investment.

The zones are not rigid. A purchase category can shift from one zone to another as information becomes more structured, as organisations build trust in AI decision-making, or as regulations adapt. The framework's value is in giving leadership teams a common reference point for discussing where their products and services sit from their customer's perspective.

What most leaders get wrong

Having discussed this framework with B2B leaders across several industries, I see the same mistakes repeated. They are worth naming directly.

Treating all purchases as equivalent. The most common error is failing to differentiate. A business that sells both commodity products and bespoke consulting services needs two different strategies for agentic commerce, because those offerings sit in different zones on the curve. Applying a single approach across the board leads to either over-investment in areas that do not need it yet or under-investment in areas that are already shifting.

Waiting for certainty. Some leaders want to see clear evidence that their specific market is being affected before they act. By the time that evidence is unmistakable, the early movers have already established their position in agent-assembled shortlists. The businesses that show up well in AI-mediated evaluation are the ones that prepared before it became urgent. This is not about speculative investment. It is about practical steps, like improving structured data and machine-readability, that deliver value regardless of how quickly automation arrives.

Confusing the curve with a cliff. The automation curve is gradual, not sudden. There is no date on which buyers in your market will switch from human-led to AI-led purchasing. The transition happens incrementally, category by category, task by task. Leaders who frame this as a binary event, either it has happened or it has not, miss the incremental shifts that are already underway.

Ignoring the supply side. Most discussions about the automation curve focus on buyer behaviour. But the curve is also shaped by what suppliers do. If your competitors make their information more accessible and machine-readable than yours, they shift higher on the curve in terms of agent visibility, regardless of the overall pace of automation in your market. Your position on the curve is partly determined by your own actions.

Delegating to IT. Agentic commerce is a commercial shift, not a technology project. The decisions it requires, about how you present your offerings, how you structure your information, and how you position for AI-mediated evaluation, are strategic decisions that belong with leadership. IT has an implementation role, but the direction needs to come from the people responsible for commercial outcomes.

What this means for your business

The automation curve gives you a practical way to prioritise. Here is how to use it.

Map your offerings to the framework. Look at what you sell from your customer's perspective. Which of your products or services sit in Zone 1 or Zone 2? Those are the areas where AI-mediated buying is already happening or will be soon. Prioritise your investment in structured data, machine-readable information, and trust signals for those categories first.

Assess your information readiness. For each zone, ask whether an AI agent could effectively evaluate your offering using the information that is currently available. Can it find your pricing, specifications, compliance credentials, and differentiation without human help? Where the answer is no, you have a gap that will cost you as automation progresses. Assessing your broader organisational readiness is a natural next step.

Plan for the middle zones. Zones 2 and 3 are where the most strategic opportunity sits for most B2B businesses. These are categories where AI agents are beginning to shape shortlists and recommendations, but where human decisions still matter. Businesses that invest now in being visible and credible to AI agents in these zones will have a significant advantage over the next two to three years.

Do not neglect Zone 4. Even in high-value, relationship-dependent categories, AI agents will increasingly handle the research phase. Your customer's agent may not select your consulting firm autonomously, but it may decide whether your firm appears on the long list that a human then reviews. Being findable and well-represented in AI-accessible information matters even for categories where full automation is years away.

Revisit the curve regularly. The factors that shape the curve are not static. As information becomes more structured, as trust in AI decision-making grows, and as regulations adapt, categories will shift along the curve. What sits in Zone 3 today may move to Zone 2 within eighteen months. Building a regular review of where your categories sit into your strategic planning process keeps you ahead of the shift rather than reacting to it.

The bottom line

The automation curve is not a prediction about when everything will change. It is a framework for understanding that change is already happening, unevenly and at different speeds, across different purchasing categories. The leaders who use that unevenness to their advantage, by investing strategically in the areas that are shifting fastest while preparing for the areas that will shift next, will be the ones who maintain control over their commercial outcomes.

The question is not whether your customers will use AI agents to buy. It is which of your offerings will be affected first, and whether you are ready when they are.

---

For a plain-language introduction to the broader shift, read What Agentic Commerce Means in Plain English. To explore how AI is changing the way buyers discover suppliers, read about AI-mediated discovery. And to assess your organisation's preparedness, start with Andrew's work on readiness.

If you want to map the automation curve for your specific business and prioritise your investment, explore how we can work together or subscribe to the newsletter for ongoing insights.

---