The Algorithm Knows the Optimal Price. But Not the Right Pricing System.

Few levers have as direct an impact on a company’s margins and profitability as price – and few are so often underused. Artificial intelligence adds new momentum to the topic, as algorithms can derive price recommendations at the customer, product, and channel level and reveal margin erosion early on. So should companies simply hand the field over to AI entirely? Not a good idea, says enomyc pricing expert Jan Ulrik Holsten, because the algorithm only optimizes within a predefined system. Whether that system makes economic sense remains a management decision.

Pricing is one of the most effective profit levers available to a company. McKinsey quantified the effect some years ago based on the income statements of S&P 1500 companies: if the average price rises by one percent while volume remains constant, operating profit increases by around eight percent on average. Conversely, after a five percent price cut, a company would have to increase its volume by 18.7 percent to offset the effect on earnings (McKinsey, 2003; confirmed 2014).

Artificial intelligence significantly expands what is possible in pricing: it can analyze large volumes of transaction data, determine price elasticities with greater precision, and steer pricing decisions at a more granular level than would be possible manually. BCG puts the potential of AI-driven pricing approaches at two to five percentage points of EBITDA improvement; in a revenue growth management transformation at a beverage manufacturer, BCG reports €150 million in additional gross profit (BCG, 2021; BCG X, 2026).

The decisive question, therefore, is not whether AI should be used in pricing, but which pricing system it should optimize.

In the Mittelstand, Margins Are Determined by the Architecture of Terms and Conditions, Not the List Price

In many Mittelstand companies, “the price” effectively does not exist. Between the list price and the actual transaction price lie base discounts, volume tiers, rebates, early-payment discounts, freight charges, advertising allowances, and customer-specific special terms. The reasons are manifold, ranging from a lack of transparency that has built up over time to discretionary latitude successfully defended by sales to a lack of knowledge about elasticities.

In other words, adjusting list prices regularly is not enough. What matters is making the price waterfall transparent all the way down to the pocket price actually realized, and understanding where margin is created and where it is lost.

Three questions are particularly relevant here:

1. Which terms buy which consideration in return? A discount is economically justifiable only if it is offset by measurable value – for example, volume, planning certainty, lower sales costs, or defined services provided by the trading partner.

2. Where should sales be allowed to decide? Pricing autonomy without defined ranges creates inconsistency; fully centralized pricing decisions, on the other hand, can ignore market and customer knowledge. The design must combine both.

3. Which pricing logic fits the business model? An industrial B2B supplier with individual quotes requires a different architecture than a wholesaler or an omnichannel provider. In B2B, quotation costing, customer segmentation, and discount management take center stage; in retail, dynamic pricing, promotions, and channel-specific pricing decisions are gaining importance.

Even the most technically sophisticated pricing system can therefore be economically counterproductive if it automates the wrong pricing logic.

AI Increases the Precision of Pricing but Does Not Replace Its Architecture

Many companies worldwide already use AI to set prices dynamically: according to Simon-Kucher’s Global Pricing Study 2025, more than 70 percent do so. At 41 percent, the most common use case to date is gathering market intelligence. Among companies that do not use AI, 54 percent cite a lack of internal expertise or resources as the barrier (Simon-Kucher, 2025). This makes one thing clear: technical adoption is running well ahead of organizational and strategic development.

Current McKinsey data point in the same direction. Between 65 and 85 percent of the companies surveyed expect to use GenAI or agentic AI in pricing within the next one to three years; today, usage stands at just 10 to 30 percent, depending on the application. At the same time, McKinsey emphasizes that the economic impact does not stem from technology alone, but from redesigned processes, robust data, suitable models, and change management (McKinsey, 2026).

AI can create an advantage particularly where many recurring decisions must be made on a sufficiently broad data foundation. These include transaction data analysis, customer segmentation, demand forecasting, price recommendations, and the identification of margin leakage. What it cannot decide on its own is the economic architecture within which this optimization takes place.

Five Management Decisions the Algorithm Will Not Make for You

Every AI-driven pricing approach requires parameters that should not be derived from historical transaction data.

• Pricing strategy: Should an offering be positioned on the basis of cost, competition, or perceived customer value?

• Segmentation logic: Which differences between customers, products, regions, and channels are economically relevant?

• Terms and conditions architecture: Which discounts and rebates exist – and what specific consideration justifies them?

• Decision rights: Which price ranges may a system use autonomously, when does sales decide, and when is approval required?

• Price enforcement: How are price recommendations realized in the actual customer conversation?

The last point is frequently underestimated. An analytically optimal price that the sales force can neither understand nor defend to the customer generates no contribution to earnings. Pricing does not end with a recommendation in the CRM. It ends with the realized price in the P&L.

Value Is Created by the Right Design, Not by More Technology

At a manufacturer with a global spare parts business, the problem was not a lack of pricing software, but a historically evolved system of terms and conditions. A flat base discount of 35 percent was supplemented by manually maintained special net prices. The redesign combined a recalibrated base discount with clearly defined considerations, volume tiers, a bonus-malus system for loyalty and performance, and country factors. The result was not just a different price, but a system of terms and conditions that can be actively managed.

A second case illustrates the limits of data-driven optimization. At a premium manufacturer of charcoal grills, an additional ash bucket added €1.68 in production costs. For customers, however, this detail increased perceived value so significantly that a €75 higher selling price became achievable. The lesson: value drivers often lie where one least expects them – and no AI in the world will find them in transaction data that does not yet exist.

The two cases represent different pricing tasks: in the first, an existing system of terms and conditions must be structured. In the second, the company must first understand what customers are actually willing to pay for. Only then does algorithmic optimization begin.

Pricing Projects Should Start with an Economic Diagnosis, Not with Tool Selection

This logic dictates a clear sequence. It starts with a quantitative baseline assessment: price realization, discount structure, profitability by customer and product, existing decision rules, and the quality and granularity of the data. Only then is it possible to determine where earnings are actually being lost and which pricing lever is relevant.

In the second step, the target system is developed: pricing strategy, segmentation, terms and conditions architecture, governance, and – where economically justified – the requirements for analytics and AI. Only at this point can one assess which technology is needed.

In the third step, the pricing logic must be embedded in ERP, CRM, and sales processes. This is precisely where it is decided whether an analytical business case translates into realized earnings.

Simon-Kucher’s Industrials and Business Services Value Creation Study 2025 underscores the importance of this approach: among the value creation levers examined, pricing delivered the fastest impact, at an average of 7.8 months, and had the lowest failure rate, at four percent. At the same time, around two-thirds of failed value creation initiatives were attributable to controllable factors such as poor implementation, unrealistic business cases, or resistance within portfolio companies (Simon-Kucher, 2025).

AI Does Not Replace Pricing Expertise. It Increases Its Scalability.

The decisive question, then, is not how intelligent the pricing algorithm is, but whether the company in question has an economically coherent system of prices and terms. If it does, AI can help leverage elasticities more effectively, manage transactions in a more differentiated way, and detect margin erosion earlier. Those who digitize legacy special terms, flawed segmentation, or unresolved decision rights merely automate their weaknesses. In the end, the advantage will lie not with the companies that have the best access to the algorithm – but with those that know which decisions it should make.