AI is moving faster than your pricing organization

Artificial intelligence is developing at a speed that most organizations have rarely experienced with previous technologies. A recent McKinsey Global Institute discussion paper, The AI economy: Interconnected forces, feedback loops, and speeds of change, looks at AI not simply as a technology but as a system. Its central argument is that AI capabilities, infrastructure, investment, regulation, organizational change and human adoption are interconnected but they are moving at very different speeds. That difference in speed matters. By one measure cited by McKinsey, the complexity of tasks that frontier AI can reliably handle (measured by how long the same tasks would take a human expert) has been doubling roughly every four months since 2023. At the same time, changing organizational workflows can take years.

This creates an important question for retail pricing. What happens when the intelligence available to a pricing organization develops faster than the organization itself? The answer may define the next stage of AI-powered pricing.

AI adoption is moving faster than organizational change

The adoption numbers initially look impressive. McKinsey cites a 2026 survey of 1,719 respondents across 97 countries in which nearly nine out of ten said their organizations regularly use AI in at least one business function.

But there is another number that may be much more important. Outside a small group of AI high performers, representing only 6% of the survey, just around one-quarter of respondents said their organizations were redesigning workflows around AI rather than inserting AI into existing processes.

This distinction between using AI and changing how the organization works because of AI is critical. A company can deploy AI widely without fundamentally changing its operating model. Employees can use copilots. Analysts can generate reports faster. Managers can summarize information in seconds. Teams can automate individual tasks. All of these improvements can create value. But they do not necessarily change how decisions are made. And in pricing, that distinction becomes particularly visible.

Consider how a pricing decision is made today

A typical retail pricing process involves many steps. Sales, cost, competitor and market data are collected. Analysts identify products that require attention. Pricing opportunities are investigated. Different scenarios are considered. Business rules and margin requirements are checked. Recommendations are prepared. Managers review them. Approved prices are passed to another system for execution.

AI can make almost every one of those steps faster. It can analyse thousands of SKUs. It can identify anomalies. It can monitor competitors. It can estimate the potential effect of a price change. It can prepare recommendations and explain why a particular action makes sense. This is already valuable.

But imagine that AI reduces an analysis that previously took a pricing manager two hours to ten minutes. The process is now significantly faster, but it is still essentially the same process. The system produces analysis, the manager reviews it, the recommendation moves through an approval process and eventually someone or something executes it.

Now imagine something different. An AI pricing system continuously monitors the market and detects an important competitor price change. It evaluates whether the competitor actually has the product available, checks the SKU's strategic importance, considers current demand, inventory and margin, evaluates several possible responses and tests them against the retailer's commercial rules. If the recommended action falls within clearly defined guardrails, it can proceed automatically. If the situation is unusual, strategically important or potentially risky, it is escalated to a pricing manager together with the relevant context and recommended actions. AI is no longer simply helping someone perform the existing process faster. The pricing process itself has changed.

When one bottleneck disappears, another appears

This leads to one of the most useful ideas in McKinsey's analysis. In a complex system, removing one constraint does not necessarily remove the constraint on the system as a whole. Instead, the bottleneck can simply move somewhere else. McKinsey illustrates this at the level of the AI economy. Early constraints around advanced chips and packaging were followed by pressure on electricity, grid connections and data-center capacity. As physical constraints are addressed, McKinsey suggests that future constraints are increasingly likely to involve applications, workforce skills and organizational workflows.

The same phenomenon can happen inside a pricing organization. Suppose AI allows a retailer to analyse 50,000 SKUs in minutes rather than days. The analytical bottleneck disappears. But if every recommendation still requires manual approval, the bottleneck has simply moved from analysis to approval. Now suppose the retailer allows routine recommendations within established guardrails to be approved automatically. The approval bottleneck begins to disappear. But perhaps prices still have to move through several disconnected systems before reaching stores and digital channels. The bottleneck has moved again, this time to execution.

Solve that problem and another may appear: poor product data, unclear commercial rules, fragmented ownership or insufficient governance over which decisions AI is allowed to make.

The important lesson is that deploying more powerful AI does not automatically make the entire pricing organization faster. The speed of a pricing decision is ultimately constrained by the slowest important part of the decision system.

Pricing therefore needs a systems view

McKinsey uses an interesting systems-thinking analogy. Imagine selecting the best engine, transmission, brakes and other components from different cars and assembling them into a single vehicle. The result would not necessarily be the world's best car. It might not work at all, because those individually excellent components were never designed to operate together.

Pricing technology increasingly faces a similar problem. A retailer might have an excellent competitor-monitoring solution, a separate optimization engine, a promotion tool, an analytics platform, business rules stored elsewhere and another system responsible for executing prices. AI can now be added to each of them. Individually, every component may become smarter. But that does not necessarily create an intelligent pricing system. Pricing decisions are inherently interconnected.

A competitor price movement may affect positioning. Positioning needs to be considered alongside elasticity. Elasticity interacts with volume. Volume affects inventory. Inventory influences markdown decisions. Promotions change customer behaviour. Margin objectives constrain which responses are acceptable. Optimizing one piece without understanding the others can simply move the problem somewhere else.

The opportunity created by AI is therefore larger than making each pricing tool individually more intelligent. It is to connect the different stages of pricing into a coherent decision loop.

From pricing workflow to pricing decision loop

Traditionally, pricing processes have often been designed as a sequence of human activities supported by software: Data → analysis → recommendation → approval → execution.

Agentic AI opens the possibility of something more dynamic: Question → Decision → Execution → Improvement.

Market signals can be monitored continuously rather than periodically. AI agents can detect situations requiring attention instead of waiting for analysts to search for them. Different agents can contribute specialized capabilities to the same decision -  competitor analysis, guardrail checking, simulation, optimization or operational execution. Routine decisions can move through the system with limited human intervention when they remain inside established boundaries. Higher-risk decisions can be escalated. And the results of previous decisions can feed back into future ones.

This is an important change because it shifts AI from being primarily an analytical layer to becoming part of the pricing operating model itself.

Human expertise becomes more important but differently

This does not imply removing pricing managers from pricing. Quite the opposite. As AI becomes capable of handling more repetitive analytical and operational work, the value of human expertise increasingly moves toward the areas where judgment matters most. Someone still has to decide what the business is trying to achieve. Should the priority be margin, market share, traffic, inventory reduction or customer price perception? Someone needs to determine how aggressively the retailer should respond to competitors. Someone needs to define acceptable margin boundaries. Someone needs to decide which products are strategically important. And someone needs to intervene when market conditions fall outside what the system has been designed to handle.

The role therefore changes from reviewing every decision toward designing and supervising the decision environment. Instead of asking a pricing manager to manually inspect thousands of routine recommendations, the organization can ask that manager to define where AI may operate autonomously, what rules it must respect and which situations require human judgment. That is a much better use of scarce pricing expertise.

Governance has to move at the same speed

Greater autonomy, however, makes governance more important rather than less. McKinsey identifies governance as another area that is developing more slowly than AI capabilities. In a separate survey cited in the paper, only around 30% of organizations had reached level three or higher on a four-level AI trust and governance maturity scale. The report also points to a basic problem that appears as AI becomes more autonomous: organizations need clarity about which actions AI may take, which decisions require human approval and who is accountable when something goes wrong.

These questions are highly relevant to pricing. An AI system should not simply be given the instruction to “maximize margin.” It needs commercial boundaries. How large can a price change be? How frequently can a price change? Which SKUs require approval? Which margin levels cannot be crossed? Which competitors matter? What happens to KVIs? When should an unusual recommendation automatically be escalated? These guardrails are not obstacles to AI autonomy. They are what make responsible autonomy possible. The more pricing decisions AI can make, the more important it becomes to clearly define the environment in which those decisions are allowed to happen.

From individual AI tools to AI Pricing Agents

This systems perspective is central to how we think about Pricerium. The future of pricing is unlikely to be one enormous AI model making every commercial decision independently. Nor is it likely to be dozens of disconnected AI tools, each optimizing one small piece of the pricing process.

A more practical architecture is a network of specialized AI Pricing Agents, each responsible for particular decisions or tasks while operating within the same broader pricing environment. A Competitor Agent can monitor market movements. A Guardrail Agent can evaluate proposed actions against business constraints. A Price Analyst can investigate performance and anomalies. An Optimization Agent can evaluate possible price points. A Strategic Advisor can bring a broader commercial perspective. Operational agents can help move decisions toward execution. And human experts can remain involved wherever commercial significance, uncertainty or risk requires judgment.

The important innovation is not simply that each agent uses AI. It is that these capabilities can participate in the same decision process. This creates the possibility of moving from fragmented pricing software toward an intelligent pricing operating system in which information, models, AI agents, business rules and human expertise work together.

The target will keep moving

There is another reason organizations should avoid designing AI transformation around today's capabilities alone.

AI itself is becoming part of the feedback loop accelerating AI development.

McKinsey notes that current models can already contribute code, data generation and evaluations used to develop subsequent generations of models. Better models can drive greater adoption; adoption attracts investment; investment expands infrastructure and R&D; and these investments can contribute to further improvements in models.

For retailers, the implication is important. A pricing architecture designed around the assumption that today's AI capabilities are fixed may become outdated surprisingly quickly. The objective should therefore not be to predict exactly what AI will be able to do three years from now. It should be to build a pricing environment capable of absorbing new capabilities as they emerge. That means modular architecture. Clear decision rights. Connected data. Explicit guardrails. Human escalation paths. And workflows in which the degree of autonomy can increase as technology, confidence and governance mature.

The next competitive advantage may be organizational

For the past several years, much of the AI discussion has focused on models. Which model is smartest? Which is fastest? Which can process the most context? Which produces the best output? Those questions matter, but they may gradually become less important for businesses as powerful AI capabilities become widely accessible.

A different competitive question then emerges: How quickly can an organization turn improving AI capabilities into better business decisions?

In retail pricing, the answer will depend on much more than the model. It will depend on data quality, workflows, decision rights, integration, guardrails, execution capability and the relationship between AI agents and human experts.

McKinsey's broader systems argument is therefore particularly relevant to pricing: technological capability alone does not determine economic impact. Value appears when the surrounding system evolves with it.

Retailers do not necessarily need to transform everything at once. Improving an existing pricing process with AI can be a sensible starting point and can create immediate value. But organizations should be careful not to confuse the first productivity improvement with the final destination. Because the most important question is changing. It is no longer simply: “What can AI do for our pricing team?” Increasingly, it is: “How would we design pricing if we started with what AI can do now and what it may be able to do next?” That is a much bigger question. And answering it may ultimately create much more value than simply making today's pricing process faster.

‍Source: McKinsey Global Institute. The AI Economy: Interconnected Forces, Feedback Loops, and Speeds of Change.McKinsey & Company, 2026.