«  View More Posts

AI In Pricing Software: Why They’re Not All The Same

June 13th, 2022 (Updated 03/10/2023) | 13 min. read

By Guillaume Dupont

Artificial Intelligence (AI) has become such a buzzword in the price optimization world recently, that’s it hard to differentiate between offerings. Surely if all the pricing solutions do AI, then they’re all delivering the same kind of capabilities, right? 

Actually, no.  

If a language app claimed their AI could teach languages, it wouldn’t be untrue, there is some vocabulary-learning going on, but by no means are you enjoying the promised “get a new soul” experience. 

As providers of AI in our price optimization solutions, we at Pricefx are well positioned to talk about the different types of AI used in pricing software, what they each can and can’t do, and the benefits of each. 

So, let’s jump in to learn more about why not all AI pricing software is equal. 

 What is AI in Price Optimization Software?

In the context of price optimization, AI helps you set more appealing prices, create better segment groups, put better offers together, and predict market and customer reaction to a price change.  

Using this invaluable data, you can drill down into the data even further to optimize your operations and portfolio to better reflect what your customers want and are willing to pay.  

For the most part, the subset of AI used in price optimization software is machine learning (ML), which uses algorithms to enable accurate prediction of outcomes.  

It is great for helping you understand which products/prices have worked in the past and, based on that, offers a prediction of what should work in the future. It provides a view of trends in customer behavior and supports the development of new products and pricing strategies. 

Three Generations of Artificial Intelligence in Price Optimization

So, let’s look at the various levels of price optimization AI and ML available on the market to explore the key differences between gen-one, gen-two, and next-gen price optimization software. 

1. Gen-One Machine Learning

A powerful tool for segmentation and price elasticity calculations for B2B (and sometimes B2C), machine learning is the most common type of AI used in pricing software.  

Let’s say you want to predict how volume could be impacted by a price change. To arrive at an accurate prediction, you’ll need to use as many different data sets as you can, (transaction, customer, product, cost, competitor, region, etc.). 

Just as a new over-skilled employee still requires training on company-specific intelligence, processes, and rules… so too does your machine learning technology. You need to feed its brain with all the data sets you want it to consider. The machine will identify the patterns. The more up-to-date the data is, the more accurate the machine will be in its predictions.  

Once you’ve got that model all set up, you can dive in; asking your questions and getting highly accurate recommendations  Results are often displayed in the form of a decision tree – where you follow Yes/No branches to find the exact prediction for each set of attributes you are interested in. 

So, you can ask: “How will a 10% increase in price impact volume… > for large customers > in Switzerland > in the Food and Beverage industry > serving B2C > …” and get super specific with your results. 


2. Gen-Two Market simulation

An invaluable B2C tool, market simulation also looks at historical data and builds a predictive model for elasticity from it (the way machine learning does), but it can do so at a much more complex level. 

With market simulation, you can create a population of your various customer segments, assign attributes and behavioral traits, and set up the desired function (e.g. willingness to pay [WTP]). It will then model these simulated customers using transactions in historical sales.

You can use it to explore how customer behavior changes when you alter your price and clearly predict which products your simulated customers would choose over the original in case of an increase. You can test all sorts of what-if scenarios, like:What if my competitor decreases price?”,What if I offer a promotion?”,What if I bundle these products?”, “How will volume rebates impact profit?”, etc 


With the right technology, your simulator can use real-time data to create a realistic sandbox for you to safely test your hypotheses in before going to market with them. 

3. Next-Gen Distributed AI

Taking things up a few notches, is Multi-agent Artificial Intelligence (MAAI), valuable to both B2C and B2B companies. 

MAAI mimics the way a complex society of individuals behaves. There is no actual AI learning in this approach, rather you’re creating an AI paradigm that imitates your customer, your processes, and the marketplace. 

Imagine a beehive and its inhabitants. Taken alone, each busy bee spends the day doing its one task (gathering /nursing/ attending/ guarding). It is a simple element in the scheme of things. One cog in the wheel. But when taken as a colony, something bigger emerges. You cannot reduce the complexity of a system like that; the whole is greater than the sum of its parts.  

You have many bees (agents), working on their own objectives (gathering) in order to support a much bigger goal (survival of the hive). And it’s not a static structure. A change will impact behavior. 

This autonomy of each agent is what makes MAAI so powerful. A swarm of small interacting agents, each with their own job to do (e.g. protect margins), can choose how they go about their tasks with free will, constrained only by rules around behavior and interaction.  

When you send in a new overarching objective (e.g. increase sales), each agent will work as part of the greater army of agents to achieve this objective within the context of your entire portfolio and waterfall while also working on its own objective (protect margins). And it will do so within the parameters you’ve set, like: “Must give specific discounts to certain customer groups,” or “No increases of more than 5% per annum.” 


MAAI optimization allows you to deal with multiple constraints, multiple objectives, and thousands of parameters, and put them all together into one single optimization process, such as full waterfall optimization or a full price list, for a much more nuanced and holistic approach to price optimization. 

Things To Consider When Choosing AI Price Optimization Software

Here are some key considerations when choosing an AI-powered price optimization solution. 

1. Can It Solve The Problems You’re Facing?  

Your pricing solution should be able to cope with the scope and complexity of the problems you need to solve.  

Most solutions claiming AI capabilities are referring to their ML-driven price optimization tools. If you’re looking for insights like: “The price that has historically worked for this customer segment in this region is $X”, then these pricing solutions will give you exactly that.  

If you want anything more nuanced or complex, then you’ll need to upgrade to one that at least offers market simulation.  

2. Does It Take Your Entire Portfolio And Waterfall In Account?


With ML, it’s important to recognize that the recommendations it makes exist in a vacuum.  

It won’t have considered how the price it suggests could impact the rest of your portfolio. Could the decrease see customers upgrading? Might the discount result in cannibalization?  

ML is very good at providing insight into price elasticity, with smart suggestions for target, floor and stretch prices. However, while these prices make sense for the segment and by considering the distribution of transactions within the segment, there is no consideration for how you can actually achieve this price. It hasn’t included your waterfall in the equation, nor did it understand any of your business constraints. 

How useful will “optimal” be if it isn’t applicable to real life, and if it needs to be altered in post-process to better reflect the reality and your numerous business constraints? If you do want to be able to optimize your entire portfolio at once, then you need MAAI. 

3. Can It Optimize More Than One Thing At A Time?

Imagine your lungs suddenly stopped prioritizing breathing to focus on helping your heart pump blood around your body. You’ve suddenly lost an essential player in the healthy functioning of you and your overall goal (staying alive) has failed.  

The same goes for your business organism. Focusing on one objective at a time could lead to an imbalance somewhere else. A unique objective applied throughout your portfolio will not be sufficient to enforce all your business objectives including channel-specific, market-specific or product-specific strategies. All essential elements must be considered and given appropriate weight in your optimization efforts. With MAAI, each agent is working towards its own goal in the context of the overall goal, so you achieve optimization across the board while carefully satisfying the many objectives that come with it. No other type of price optimization AI can do this. 

4. Can It Factor In All Constraints?

Companies often have pricing strategy rules that stipulate things like: “We can’t price higher than the competitor,” or “Our basic product must remain cheaper than our premium product at all times.” 

With most price optimization solutions, these constraints are applied in post-process, so the machine learning knows nothing about them. It is going to solve your problem as if there were no constraints. So the “optimal” results it provides may not even be usable without many adjustments or overrides that will ultimately make most of its outputs stale. 

More sophisticated solutions will be able to take these constraints into consideration before solving the problem so that the optimal solution it offers is actually optimal and actionable. 

If you want to optimize multiple elements taking multiple constraints into account simultaneously, then MAAI technology is the only way to go. 

5. How Much Work Does It Require From You?

When using a model to solve a problem, you want it to be as close to reality as possible, so you need as much real-life, real-time data as you can get. 

A lot of machine learning solutions require very structured data and detailed feature engineering; meaning you must teach the technology what it’s likely to find in the data so that it can understand it. You’ll also need to set a list of parameters and rules to ensure your results meet your actual needs. If you haven’t taught it to spot a pattern, it probably won’t find it when left to its own devices. It will also miss more complex patterns, or patterns that shift as market conditions change. 

When you have more sophisticated AI, you just give it the data. Of course, you’ll probably have to make small adjustments, but, it requires much less structure and engineering because the algorithm is smarter and can do a lot of the work and find solutions by itself. More advanced AI will be able to understand some extraordinarily complex data patterns and handle very complicated problems. 

If you’re able to build more sophisticated models, then your model will closer reflect reality, which means you’re understanding things better… and optimizing things better. And this without having to constantly override the model’s outputs. 

6. Does It Go the Extra Mile?  

Unlike ML, AI can take you beyond price optimization towards broader commercial optimization. 

With sophisticated AI you’re able to upsell & cross-sell, remind customers of shopping carts, recommend related products, and suggest similar alternatives – all incredible tools for improving the digital customer experience, something touted to be more important than price to some customers in the B2C world.  

Another way of ensuring customer satisfaction is through pricing timeliness. If your pricing solution is able to adjust the price of your offering automatically and dynamically in line with real-time raw material or transportation costs, then you’re in a great position to be competitive in the space and pass savings on to your customers. (Plus, when prices go up, you can pass along those increases quickly too, so you’re not losing margin to latent price changes.) 

A pricing solution that integrates cleanly with and provides support to your ecommerce application is going to not just help you achieve smarter pricing but will also help you better understand your customers and deliver better brand experiences to them.  

7. Does It Share Its Secrets?

Do you remember how in math class you had to show your workings to prove that you didn’t just pluck a number from thin air to present as your answer? Well, it turns out, we expect the same from robots.  

Typing a business-critical equation into a machine and getting a magical answer doesn’t instill trust. This is a black box approach to optimization and is what you get with typical AI-enabled solutions. You want to understand the workings and rationale behind a recommendation, but you can’t even reverse engineer the solution to check that all essential elements were considered. 

If you want to have some idea of how a price was arrived at or to be able to verify a suggestion, then you’ll need your algorithms to support a clear-box approach, where you can see into the workings behind a result. Being involved at every step gives you clarity, decision support, and a way to step in and amend if necessary. It gives you confidence in your pricing next-steps and gives sales conviction (and negotiation power) when dealing with customers. 


Some solutions even enable you to drill down to the individual agent level to see what constraints it was impacted by and by how much! 

Time To Choose Next-Gen Capabilities?

The above list goes a long way a to explain why not all AI in pricing software is equal. Is it all AI? Technically, yes. Is it delivering the same level of price optimization? Not by a long shot. 

ML is a powerful tool that can help you optimize price based on historical trends such as margin levels or customer willingness to pay. (We’re not knocking it – we include it in our own offering!). But it really only enables simple solutions to one-dimensional problems. This makes it the perfect choice for companies whose digital maturity in pricing is still young or those who know they won’t be dealing with complex pricing or discounts. 

Many of our customers have extremely complicated pricing, with complex negotiations, customer-specific prices, and layered rebate systems in place. That’s why we’re so invested in delivering market simulation and MAAI capabilities.  

Not only do they deliver right from the get-go, taking a smarter approach to your data and requiring less hand-holding from you, but these advanced AI technologies keep you involved with the process, providing in-depth (agent-specific) accounts of what they’re “thinking” at every step, and why the recommendations support all your optimization goals simultaneously while also taking your entire portfolio, waterfall, and all your constraints into account. 

Not all AI in pricing software is equal. Make sure you’re choosing a robust, sophisticated solution that integrates with your current ERP or SAP and meets your precise price optimization needs—today and in the future. Get a full and in-depth explanation of Machine Learning vs AI in Pricing Software in the article link below to learn more.



Guillaume Dupont

Solution Architect & Data Scientist - Presales in Solution Strategy , Pricefx

Guillaume is a Senior IT architect with a passion for commercial, digital and management problems. He has a proven track record of leading cutover-critical project deliveries and complex mission-critical multi-million presales and enterprise architecture engagements for Fortune 500 customers. He has been a trusted driver of internal change with experience initiating or overseeing transverse initiatives leading to tremendous process, product or portfolio changes.