by Dr. Jan Wieneke
Price optimization is the discipline of determining the price or price guidance that best meets a stated business objective, subject to real-world constraints, using data and analytical models rather than intuition or historical habit.
Does that sound too technical? It has to be. Price optimization that is not grounded in the mathematics of constrained decision-making is not optimization. It is repricing with a confident name attached.
What price optimization is not is simpler to state.
It is not a single formula that finds the perfect price. In enterprise B2B, where deals are negotiated, customers have specific relationships and histories, and the same product can carry twenty different prices across twenty different accounts, the idea of one optimal price is a simplification that breaks almost immediately in practice.
What price optimization actually provides is a governed system: demand-aware, segmentation-informed guidance that helps pricing teams set better list prices, and helps sales teams quote deals that balance margin and win probability. In the most mature implementations, it does both simultaneously and continuously.
This article explains how that system works, what the data and modeling requirements are, why B2B price optimization looks fundamentally different from retail price optimization, and what the organizational conditions are that determine whether it delivers.
What is price optimization?
Price optimization is the process of finding the price or price range that best achieves a specific business objective, given what is known about customer demand, cost structure, and commercial constraints.
The formal version of that definition comes from operations research and economics. An optimization problem has three components: an objective function that defines what you are trying to maximize or minimize, a set of decision variables that you can control, in this case prices, and a set of constraints that limit the feasible solution space, in this case margin floors, competitor ceilings, approval policies, and change caps.
In pricing, the objective function might be maximizing contribution margin, maximizing revenue, maximizing win rate, or some weighted combination of all three. The constraints are the guardrails that prevent the model from recommending prices that are technically optimal in the math but commercially or operationally impossible in practice.
What price optimization is not
It is not a single formula that finds the perfect price.
It is not dynamic pricing, though dynamic pricing can be one application of an optimization model. For a full explanation of how dynamic pricing works in enterprise B2B, see our guide to dynamic pricing.
It is not price management, which is the governance and execution layer that sits around the prices optimization recommends.
And it is not the same problem in B2B as it is in retail or e-commerce, where high transaction volumes, observable demand responses, and public price points make the classical demand curve a practical starting tool. In enterprise B2B, those conditions rarely hold.
How optimal prices are actually determined
The mechanics of price optimization separate two distinct activities that are often conflated: forecasting and optimization.
Forecasting estimates what will happen to demand if prices change. Optimization uses those forecasts, along with costs and constraints, to choose the best feasible price. They are connected but they are not the same step, and confusing them is one of the most common reasons pricing science projects underdeliver.
In practice, the inputs to a price optimization model are:
Demand response or price elasticity by segment. Elasticity measures how sensitive customer demand is to price changes. A price elasticity of minus two means a one percent price increase is associated with approximately a two percent decrease in demand. But enterprise B2B pricing does not run on one companywide elasticity number. Different customers, products, regions, channels, and deal types have meaningfully different demand responses, and an optimization model that ignores that variation will recommend prices that are wrong for most of the situations it is applied to.
Willingness to pay (WTP). WTP is the maximum price a customer or segment would accept for a given product or service. In consumer research, WTP is often estimated through survey methods including conjoint analysis, Gabor-Granger pricing tests, and Van Westendorp price sensitivity meters. In enterprise B2B, where transaction data exists at the account and deal level, WTP signals can often be inferred directly from historical deal outcomes: what prices were accepted, at what discount depth, under what competitive conditions.
Segmentation. Pricing lives in differentiated contexts. The same product sold to a large strategic account through a direct sales relationship, a mid-market account through a distributor, and a spot buyer through an online portal warrants different price logic for each context. Segmentation is the scaffolding that lets an optimization model apply the right logic to the right situation rather than producing one answer that is suboptimal for every situation it is applied to.
Constraints. Real optimization models operate inside guardrails: margin floors that protect the business from pricing below an acceptable threshold, price ladders that preserve logical relationships between product tiers, change caps that prevent recommendations that move too far too fast and create customer friction, and approval policies that route high-risk exceptions to the right decision-maker.
Business objectives. Different objectives produce different optimal prices. A business prioritizing margin protection will optimize differently than one prioritizing volume growth or competitive win rate. The objective function is a business decision, not a modeling decision, and it needs to be made explicitly before the model is configured.
A worked example
A simplified single-product case illustrates the basic logic, using a standard profit-maximizing pricing rule. If a product's marginal cost is $6 and its price elasticity is minus four, the rule gives an optimal price of roughly $8. Optimal price equals marginal cost divided by one plus the reciprocal of elasticity. At elasticity minus four, that is $6 divided by 0.75, which gives $8.
That calculation is a useful starting point and a reasonable answer to how do I calculate optimal price in Excel. You can build a simple demand curve, estimate elasticity from historical transaction data, apply the formula, and arrive at a candidate price.
What it cannot do is handle the enterprise B2B reality. Negotiated deals, discount waterfalls, customer-specific agreements, rebate obligations, portfolio interactions across hundreds of SKUs, and the difference between what a price list says and what actually lands in pocket margin all mean that the single-product textbook case breaks quickly at scale.
That is why serious B2B price optimization moves from one-price-point math to segment-level and deal-level optimization under constraints, with machine learning (ML) doing the demand estimation work and a governed optimization engine doing the decision work.
Most descriptions of price optimization treat it as a single activity. In enterprise B2B, it is two distinct layers operating at different points in the pricing workflow, with different data requirements, different modeling approaches, and different organizational ownership.
Understanding the distinction is not just analytically useful. It is the difference between deploying the right capability at the right moment and building an optimization system that answers a question nobody on the commercial team is actually asking.
Layer one: list price optimization
List price optimization operates upstream of the commercial transaction. Its job is to set forward-looking, segment-aware base prices that reflect demand response, competitive positioning, cost structure, and business objectives across the product portfolio.
The inputs are primarily historical transaction data at the segment level, market intelligence, cost information, and the elasticity estimates that emerge from modeling those inputs together. The output is a recommended price or price range for each segment, product, and channel combination in scope, along with the scenario modeling that shows what happens to margin and volume if the recommended prices are accepted, partially adopted, or rejected.
The governance requirement is simulation before deployment. A manufacturer managing tens of thousands of SKUs across multiple regions cannot test list price changes in the market before rolling them out. The optimization system has to do that testing analytically, letting the pricing team compare the margin impact of alternative strategies before any price goes live.
List price optimization is also the layer where index-linked and formula-based pricing logic lives. In chemicals, distribution, and process manufacturing, list prices are often tied to raw material indices, commodity benchmarks, or cost-escalation formulas. Optimization here means building the logic that automatically recalculates base prices when input costs move, within governed constraints that prevent changes from moving faster than the market or the customer relationship can absorb.
Layer two: deal optimization
Deal optimization operates downstream, at the point where a sales team is building a quote, responding to a request for proposal, or negotiating contract renewal terms with a specific customer.
The modeling problem at this layer is fundamentally different from list price optimization. It is not primarily a demand curve question. It is a win probability question.
As Wipro's research on machine learning for B2B pricing describes it, in negotiated environments the better modeling question is often bid price versus win probability: at what price does this deal have an acceptable likelihood of closing, given everything we know about this customer, this competitive situation, this product mix, and this deal structure? The answer to that question is a price corridor, not a point.
That corridor is what deal optimization produces. A floor that protects margin. A target that reflects what the business wants to realize. A ceiling that reflects the maximum the market and the relationship will bear. All three are visible to the sales team building the quote, with the reasoning attached so they can use the guidance in a customer conversation rather than around it.
The data that powers deal optimization is different from list price optimization. It draws on deal-level transaction history, win and loss outcomes at different price points, customer-specific relationship data, competitive intelligence, and the sales team's own contextual judgment encoded over time through feedback loops.
As we frame it, salespeople want to price their deals to maximize both their chance to win and their chance to profit. Deal optimization is the system that learns from past deals and combines those learnings with an assessment of all the relevant variables to generate that guidance. The corridor is not a constraint on the salesperson. It is the answer to the question they are already asking.
Why both layers are necessary
List price optimization without deal optimization produces well-designed base prices that erode in execution. Sales teams discount away from the list price in every negotiation, and without deal-level guidance calibrated to win probability, the erosion is driven by intuition rather than data.
Deal optimization without list price optimization produces sophisticated guidance on top of a poorly calibrated base. The corridor is only as useful as the list price it is anchored to. If the list price is wrong, the floor, target, and ceiling derived from it are all wrong by the same amount.
The two layers are not sequential. They are complementary. In a mature pricing capability, list price optimization continuously recalibrates the base, and deal optimization continuously learns from deal outcomes and feeds that signal back into the next round of list price modeling. The feedback loop between them is what makes the system improve over time rather than degrade.
Most of what has been written about price optimization was written for retail and e-commerce. That is not a criticism. It is a statement about where the analytical frameworks were developed and where they work well.
In retail, prices are public. Transactions are anonymous and high-volume. Demand responses are observable across thousands of daily transactions. Competitors' prices are visible in real time. Price changes can be tested, measured, and reversed quickly. The classical demand curve, where demand is a predictable function of price, is a reasonable working model under those conditions.
In enterprise B2B, almost none of those conditions hold.
Prices are negotiated, not published
In most B2B selling environments, the price a customer pays is the outcome of a negotiation, not a list price read off a public board. That negotiation involves relationship history, competitive context, deal structure, volume commitments, rebate expectations, and the individual judgment of the sales team member making the call.
That means the demand curve in B2B is not a clean function of price. It is a messy, context-dependent relationship between price and win probability, mediated by all of those other variables. Estimating it requires deal-level transaction data, win-loss records, and modeling approaches that account for the negotiated nature of the transaction.
Transaction data is sparse at the segment level
In retail, a large category might generate millions of transactions per year, giving the optimization model a rich statistical foundation for estimating elasticity. In B2B, a pricing team might have dozens or hundreds of transactions per product-segment combination per year, and some combinations might have very few data points at all.
Sparse data is not just a modeling inconvenience. It means that standard elasticity estimation methods produce wide confidence intervals and unstable estimates. The practical response is to use hierarchical models that borrow strength across segments, to incorporate external data sources where available, and to be honest about the uncertainty in the estimates rather than presenting point estimates as if they were precise.
This is one reason why the worked example in the previous section, while mathematically correct, requires careful handling in B2B. The elasticity input that drives the calculation is itself an estimate with uncertainty attached. An optimization model that does not propagate that uncertainty into its output is overconfident in ways that can create real commercial problems.
Win probability replaces demand elasticity as the primary modeling target
In negotiated selling environments, Wipro's research on machine learning for B2B pricing makes the case clearly: the more useful modeling question is not what is the elasticity of demand for this product? It is at this price, what is the probability of winning this deal, given everything we know about the customer, the competitive situation, and the deal structure?
That reframe changes the data requirements, the modeling approach, and the output format entirely. The model is trained on historical deal outcomes rather than demand responses. The output is a probability distribution over win outcomes at different price points rather than a demand forecast. And the guidance it produces is a price corridor calibrated to acceptable win probability rather than a point estimate of the profit-maximizing price.
Pricefx win rate optimization allows sales teams to close more profitable deals by improving pricing precision, discount discipline, and decision confidence.
Customer relationships create price memory
In B2B, customers remember what they paid last time. They have contracts that establish reference points. They talk to other customers. They compare renewal offers to acquisition offers. And they have procurement teams whose job is to exploit any inconsistency in the seller's pricing behavior.
Research on reference prices and loss aversion shows that prices above a customer's reference point generate a disproportionately negative reaction compared to the positive reaction generated by an equivalent discount below it. In B2B, where the reference price is anchored to the last contract, the last renewal, or the last comparable deal the procurement team is aware of, price increases carry asymmetric commercial risk that a retail-style optimization model simply does not account for.
The practical implication is that B2B price optimization has to model the relationship, not just the transaction. A price that maximizes margin on an individual deal can damage the relationship in ways that cost far more over the lifetime of the account.
The most sophisticated price optimization model in the world delivers zero commercial value if the sales team does not trust it enough to use it.
That is not a soft observation about change management. It is a structural constraint on what price optimization can deliver in B2B. The last mile of every pricing decision is a human conversation between a salesperson and a customer. If the salesperson cannot explain the price, they will either discount away from it or ignore the guidance entirely. Both outcomes erase the value the model was designed to create.
The trust problem in B2B pricing AI
Trust in a pricing recommendation has two dimensions.
The first is outcome trust: does the guidance lead to better commercial results over time? That is established through track record, through win rate data, and through the visible connection between following the guidance and the margin and volume outcomes that result.
The second is process trust: can the salesperson understand why the recommendation is what it is? In practice that means asking:
-
Can they explain to a customer why the price is at this level and not lower?
-
Can they defend it to a procurement team that is pushing back?
-
Can a pricing manager explain it to a finance director who wants to understand why margin improved or declined?
-
Can the recommendation be examined, challenged, or overridden when the business context requires it?
Process trust is harder to build and faster to lose. A black-box model that produces a number without context does not earn process trust. It creates anxiety: if the salesperson cannot understand the recommendation, they cannot own it. And a price they cannot own is a price they will negotiate around rather than defend.
The IDC MarketScape: Worldwide B2B Revenue and Profit Optimization Platforms 2025-2026 frames the direction of travel clearly: "pricing must evolve from manual management or back-office administration to proactive orchestration of data, analytics, and automation across channels." Explainability is not a concession to organizational resistance. It is the condition that makes that orchestration trustworthy at scale.
McKinsey's 2026 research on the next phase of AI in B2B pricing reinforces the same point: agentic pricing systems require explainable, resilient models with human guardrails. The emphasis on explainability is not a design preference. It is a recognition that the commercial stakes in B2B pricing are too high for decisions that cannot be examined, challenged, or overridden when the context requires it.
What explainability requires in practice
An explainable price optimization system does four things.
It shows the reasoning alongside the recommendation. Not just the floor, target, and ceiling, but the factors that drove them: the segment elasticity estimate, the win probability at each price point, the margin floor dictated by cost and policy, the competitive context that informed the ceiling.
It surfaces confidence alongside the estimate. A recommendation derived from thousands of similar historical transactions carries more weight than one derived from a handful of data points. The sales team should know which situation they are in.
It allows override with accountability. When a salesperson has context the model does not, they need to be able to act on that context. An explainable system does not prevent overrides. It logs them, routes them through appropriate approval where warranted, and uses the pattern of overrides over time to improve the model.
It connects recommendations to outcomes. When a deal closes above the target price, the model should know. When it closes below, the model should know that too. The feedback loop that connects recommendation to outcome is what makes the model trustworthy over time rather than trustworthy on day one and increasingly stale thereafter.
The governance layer that makes optimization safe to deploy
Explainability is the user-facing requirement. Governance is the organizational requirement.
Enterprise pricing decisions affect customer relationships, contract obligations, competitive positioning, and in some industries, regulatory compliance. The pricing model is one input into those decisions, not the final word. The governance structure around it, the approval workflows, the exception handling, the audit trails, determines whether the organization can deploy AI-driven pricing recommendations at scale without creating commercial, legal, or reputational risk.
That governance layer is not a limitation on what AI can do in pricing. It is the condition that makes it safe to do more. Organizations that build explainable, governed optimization systems are the ones that can expand AI's role in pricing over time. Organizations that deploy black-box models without governance are the ones that pull back after the first high-profile pricing error.
As we see it at Pricefx, the goal is not optimization that operates autonomously. It is optimization that operates transparently, with humans retained for the decisions where context, relationship, and judgment matter most, and AI accelerating the analytical work that makes those human decisions better.
How machine learning, optimization, and agents work together
The current market conversation about AI in pricing collapses three distinct capabilities into one label. That collapse causes real problems: buyers invest in the wrong layer, vendors overclaim what their models do, and organizations end up with sophisticated analytics that do not connect to commercial decisions, or automated recommendations that nobody trusts because nobody understands them.
The three layers are distinct. They work together. And they are not interchangeable.
Layer one: machine learning estimates
Machine learning (ML) is the layer that does the statistical estimation work. It learns from historical data to produce the inputs that the optimization layer needs: demand response by segment, win probability at different price points, customer elasticity patterns, churn risk signals, and the clustering logic that groups products and customers into meaningful pricing segments.
ML does not set prices. It builds the picture of the commercial environment that the optimization engine works within. A machine learning model that estimates elasticity poorly will feed bad inputs to a well-designed optimization engine and produce bad recommendations. The quality of the ML layer determines the ceiling on what the optimization layer can achieve.
Layer two: optimization solves the constrained decision
The optimization engine takes the ML outputs and solves the constrained pricing decision: given what we know about demand response, given our cost structure and margin requirements, given the policy constraints the business has defined, what price or price range best achieves the stated objective?
This is the layer that produces the floor, the target, and the ceiling. It is also the layer that runs the scenario modeling: what happens to margin and volume if we accept the recommended price, if we price ten percent above it, if we price at the floor? Those scenarios are what give pricing teams the confidence to make changes at scale rather than one SKU at a time.
The optimization layer is deterministic within its constraints. Given the same inputs and the same objective function, it will produce the same output. That determinism is a feature, not a limitation. It is what makes the recommendations auditable, explainable, and defensible.
Layer three: agents monitor, prioritize, and route action
Agents are the operational layer that makes optimization commercially useful in real time. An optimization model that runs monthly and produces a report is useful. An agent that monitors your transaction data continuously, identifies the accounts where pricing is drifting below floor, flags the products where cost increases have not been reflected in price, surfaces the deals where win probability suggests a higher target is achievable, and routes those findings to the right person with the right context, is a commercial capability.
Pricefx Agents are built around this principle. Always-on monitoring across products, customers, quotes, and transactions. Event-driven detection of margin risks and commercial opportunities. Explainable recommendations with reasoning attached. Human review before action, with no autonomous execution of decisions that affect customer relationships, contracts, or realized margin.
The distinction matters because agents without a governed optimization foundation produce fast findings on top of poorly calibrated pricing logic. The agent identifies that a customer is paying below average for a product category. But if the optimization layer has not correctly estimated what the right price for that customer and product actually is, the agent's finding is directionally correct but actionably wrong.
The National Instruments proof point
The commercial case for the three-layer architecture is not theoretical. National Instruments extended Pricefx AI optimization with a bring-your-own-science capability, combining the platform's optimization infrastructure with their own pricing models. The result was a 6.7 percent margin increase in 2022.
Mukesh Sampath, who led the implementation, described it this way: the approach "really tripled the value capture in the marketplace." The bring-your-own-science model matters here because it illustrates the architecture in practice. The optimization infrastructure provided the governed deployment layer, the scenario simulation, the approval workflows, and the execution pipeline. The customer's own pricing science provided the demand estimation layer. The two worked together rather than requiring the customer to choose between platform optimization and their own models.
That is what a mature three-layer architecture looks like: machine learning estimating the commercial environment, optimization solving the constrained pricing decision, and agents connecting the output to the moment in the workflow where a human needs to act on it.
Price optimization is frequently used as a catch-all term for several distinct pricing disciplines. The distinctions matter commercially because each one addresses a different problem, and misidentifying which problem you have leads to deploying the wrong capability.
Price optimization vs. price management
Price management is the governance and execution layer: list prices, customer agreements, discount structures, rebates, channel terms, approval workflows, and performance measurement. It governs how pricing decisions are executed consistently across the business. For a full explanation of what price management covers and how the waterfall works, see our guide to price management.
Price optimization is the intelligence layer that sits inside that governance system. It recommends better prices and better deal guidance. Price management governs how those recommendations are deployed, enforced, and measured.
Optimization without governance produces better recommendations that are inconsistently applied. Governance without optimization produces a well-managed system that is optimizing fo the wrong targets.
Price optimization vs. dynamic pricing
Dynamic pricing is the practice of adjusting prices frequently in response to changing market conditions: demand signals, inventory levels, competitive moves, and timing. Price optimization can power dynamic pricing by continuously recalculating recommended prices as inputs change.
But price optimization is broader. It also applies to annual list price reviews, negotiation guidance for deals that close over weeks or months, and contract renewal pricing where the cadence is slow and the relationship context matters more than real-time market signals.
Dynamic pricing is one application of optimization logic. Optimization is the discipline that makes dynamic pricing analytically defensible rather than just reactive. For a full explanation, see our guide to dynamic pricing.
Price optimization vs. cost optimization
Cost optimization reduces the cost base or cost-to-serve. Price optimization determines the best price given customer demand response, cost structure, and commercial constraints.
The two interact: a cost reduction changes the margin floor, which changes the feasible solution space for the optimization model, which may change the recommended price. But they are different levers operated by different teams with different analytical methods.
Conflating them typically leads to one of two errors. Either the pricing team treats cost reduction as the pricing solution, which produces lower costs but not necessarily better prices. Or the pricing team ignores cost dynamics entirely, which produces better-looking price recommendations that do not account for the margin reality they are operating within.
Price optimization vs. revenue management
Revenue management is a distinct discipline built around maximizing revenue from perishable inventory under capacity constraints. Airlines, hotels, and freight carriers use it to optimize yield across finite, time-sensitive supply.
It shares conceptual ground with price optimization but is built for a fundamentally different operating context. If your business sells physical products or professional services on negotiated terms rather than capacity-constrained inventory that expires if unsold, revenue management is not the right frame.
The confusion is worth naming directly because revenue management software vendors sometimes market into B2B manufacturing and distribution accounts where the operating model does not fit. The constraint structure, the data requirements, and the decision cadence are all different.
Price optimization in enterprise B2B is not a single model or a single moment. It is a system of three connected layers: machine learning estimating the commercial environment, optimization solving the constrained pricing decision, and agents connecting that output to the humans who need to act on it.
Organizations that build that system with explainability and governance at the center are the ones that can expand AI's role in pricing over time. The ones that skip the foundation find themselves with fast recommendations that nobody trusts and a commercial capability that stalls before it compounds.
The IDC MarketScape evaluation below shows how leading pricing platforms are being assessed against exactly these criteria.
Ready to implement pricing software? Talk to an expert.
Industry Advisor, Pricefx
Dr. Jan Wieneke is an Industry Advisor at Pricefx, focused on the manufacturing sector. He brings a rare mix of academic depth and hands-on pricing leadership, having held senior global pricing roles at Envalior and LANXESS and worked as a strategic pricing consultant across B2B, B2C, and consulting. He holds a doctorate (Dr. rer. pol.) in experimental economics from Bielefeld University, where his research explored decision-making and behavioral strategy. Based near Düsseldorf, Germany, he writes on pricing in manufacturing and the practical side of pricing strategy.
