5 Things Pricing AI Can’t Do for You

What comes to mind first when you think of artificial intelligence?  

Many of us remember playing against a computer in chess. Fascinated with our opponent’s inborn mastery of the rules of the game, we might be tempted to apply this wide-eyed faith in artificial intelligence’s seemingly all-knowing capabilities to other areas of our lives. But that same AI designed to beat us in chess would be pretty bad at a game of Monopoly. It’s not good for anything else, and that’s exactly the point. 

In the same way, pricing AI is not an entity from science fiction that will take over the reins of your business and solve all its problems without any guidance from you. While data-driven systems are extremely good at accomplishing a defined objective, it’s up to you to define the parameters, guardrails, and assumptions embedded in its framework to carry out that task successfully and in the way you imagined.   

At Pricefx, as a leading cloud-native pricing software company with AI-powered price optimization among our core product offerings, we recognize that artificial intelligence is an incredibly dense subject to wrap your head around. In order to set realistic pricing goals, it’s crucial our customers understand the limitations and misunderstandings around data-driven solutions just as much as the benefits.  

In this article, we address 5 of the most common ways our customers get pricing artificial intelligence wrong, and what they need to understand instead to get the most out of their data-driven solutions.  

What is Pricing Artificial Intelligence?  

Artificial intelligence for pricing refers to using a data-powered mechanism designed to solve a predefined pricing scenario as best as possible relative to a set of goals. Now, what exactly does  “best as possible” mean in a pricing AI setup? That’s defined by you as a user and speaks directly to the specific challenges you’re trying to address. 

 At the end of the day, you will need to know what you are looking to achieve, and implicit in that, will need to guide the AI engine to fulfill your desired task.  

Oddly enough, to turn Artificial Intelligence into Actual Intelligence, the most important part of the puzzle is you. The real intelligence in AI for pricing does not reside in an all-knowing set of microchips driven by some mysterious force, but rather it’s the user ability to properly leverage this process that makes it intelligent.  

To better prepare you to introduce artificial intelligence into your pricing, we’ve responded to some of the most common misunderstandings we’ve heard from our customers about what pricing AI can do for them. 

5 Things Pricing Artificial Intelligence Can’t Do for You 

1. AI can’t figure out your business logic by itself 

Every company has a structure that its members follow, including its pricing team. Company X knows that it can’t or shouldn’t move certain pricing levers in certain cases.  Does it always make sense to reprice your customers every week or day?  

It also knows there are specific actions which shouldn’t be taken due to a company’s unique business goals and constraints, or perhaps even the end customer’s pricing expectations. (For example, an AI pricing engine might recommend that a choice cut of meat should be priced the same as a prime cut, but would this be considered reasonable?).These business rules are often implicitly understood in the organization or industry – but does your AI know these rules just as well as you?  

Even in an artificial intelligence environment, that same structure is needed to support successful outcomes.  

Pricing AI doesn’t know by default it has to incorporate our business logic into the automated processes we’re asking it to do at scale. If you focus on the desired outcome more than the business rules underlying your AI system, there’s a chance you’ll be disappointed with the results. Neglecting the guidelines that shape the pricing structure of the data solution can lead to misalignment between your expectations for the tool’s performance and the actual outcome, causing certain conclusions that are unusable for your business.  

In order to embed your pricing framework into the data solution, your company will need to provide the algorithm with specific guidelines such as product classifications, reference points for analysis, and specific criteria for determining prices within and across products.  

This infographic further breaks down the basic components of the pricing structure needed for data models: 

2. AI can’t set your pricing objectives for you 

Before leaning on artificial intelligence to ramp up your company’s pricing, you should have a general idea of what you are looking to get out of the solution – because the algorithm may change based on what your objectives are. Maybe you’re looking to minimize margin leakage in B2B pricing negotiations through a more targeted data-driven pricing structure, or perhaps you’re looking to simplify your pricing by generating a list price that a majority of target customer based would be willing to pay.  Whatever your unique pricing goals turn out to be, the algorithm and the corresponding overall solution should be designed to directly respond to them. 

As a professional in the pricing space, you recognize that pricing is not a monolith. There are as many pricing solutions as there are business problems, leading to a high degree of variability in  pricing strategies that exist within any given industry. There is no such thing as a universal pricing strategy that will hold true for all companies.  The strategy that you are looking to execute on must be established  – as pricing AI cannot set it for you. 

As much as we’d like it to be, pricing AI is not a machine that runs through your data and magically spits out a single global approach for your business to follow. AI is built on the opposite logic; it gets things done by following our directions. 

Simply put, the right question is not what AI can do for you, but what you need AI to actually do for you specifically.  

3. AI can’t give you great solutions if you don’t give it the right information  

Predictive pricing models, as is the case with any AI solution, are only as good as the data you feed them. The insights that a model will provide are based on the general patterns arising from the data. Contrary to popular belief, the data inputted into a pricing model does not need to be perfect (insert that cliché that is always used in this situation: garbage in, garbage out), but it should be sufficiently representative of the pricing scenario you are looking to address.  

For example, asking a pricing solution to provide you an acceptable price range that may be leveraged during a negotiation will be tough to do if all your historical transactions were sold at one single price point. After all, it’s difficult to predict a reasonable range of prices in the future if the AI algorithm did not see a range of prices in the past.    

How reliable, then, does this render these pricing models once special situations are introduced? This is where we need to tread carefully.  Prescriptive pricing models will in most cases provide a recommendation for any situation, as their job is to provide you the best outcome possible by applying the underlying logic to the historical data. If there is little or no historical data tied to the situation you’re trying to evaluate, what can you reasonably expect pricing AI to do?

Pricing AI only knows what it’s seen in the past. While it can give a general recommendation for how to act, ultimately those situations will likely require the user to nudge those kinds of recommendations in the right direction. The assumption about AI that historical data is by default representative of how a business actually wants to price runs the risk of institutionalizing bad decisions by recognizing them as inherently “good” in the algorithm. To nudge it appropriately, you should have a working knowledge of the underlying logic – and no, you don’t have to be a data scientist to understand the general logic. Be leary of any AI pricing solution that only a data scientist can control, as those type of solution are typically not pricing solutions, but rather mathematical experiments that can be irrelevant to your goals.  

This inability to handle special situations appears to be a limitation of artificial intelligence –  and it is. After all, it’s artificial. Your goal is to make the intelligence Actual by providing a framework that allows the user to appropriately manage these types of situations.  This presents an interesting trade-off to consider: does your company want to develop a system that automatically encompasses all possible challenges under the sun? If yes, keep in mind that there are a lot of space that falls under the sun, and you will be implementing your product for a long time constructing a tarp big enough to cover this large area. Treating AI instead as a support mechanism for your decisions and working alongside AI to evaluate its recommendations negates the need for that giant tarp (note: this will require understanding how results have come up in the first place). 

The more realistic alternative is accepting that a machine can’t always do the work alone. Properly designed influencing and override mechanisms will provide the necessary cover for a data-driven pricing solution’s blind spots, and there will be moments when you need to add a pinch or two of salt to the recommendations generated. After all, while data-driven tools are designed to provide optimal solutions, these solutions are not always what we actually need or want to see.  

4. AI can’t show you how it works – unless it’s designed to  

In any math class, we’re told to show our work, otherwise it becomes impossible to understand how we’ve arrived at a certain conclusion. That kind of transparency is not always guaranteed in the case of artificial intelligence. The reality is that the more complex the algorithm, the more opaque its processes become for us users, which brings us to clear box and black box AI systems in pricing. 

What is clear box and black box artificial intelligence? 

Transparent machine learning, or a “clear box” approach, in price optimization software ensures that any optimized result can be explained and adjusted as needed by the user, while black box features a path from input to output that is undeterminable, or incomprehensible, to users.  

Black box AI systems, such as neural networks, have impressive analytical capabilities in part due to their ability to leverage the interactions between a large number of variables to detect the intrinsic patterns in its datasets — and yes, it’s as complicated as it sounds.  Although these black box approaches can (and have) generated impressive results in non-pricing applications, when it comes to pricing, it’s important to remember that this level of complexity comes at the cost of a company’s more active participation in and understanding of its pricing, which may or may not align with what is needed.  

More importantly – consider the change management of implementing pricing optimization.  If users such as salespeople do not understand where a price recommendation came from, they will ignore it.  If that happens, it does not matter how great the algorithms behind the AI are – the output is wasted and not used. 

Between clear box and black box artificial intelligence, which approach makes the most sense for your pricing? 

The best way to decide this is to answer the simple question – “how much will your users blindly trust output from your AI?”   

Where your company is in its pricing journey determines the level of trust, control, and insight needed into pricing outcomes. Other factors to consider include what frequency and speed prices should be adjusted, and your company’s tolerance to using pricing recommendations that may be perceived as impractical. Some highly mature price optimization industries, such as airlines, can handle a black box model because the factors to control are too numerous to be worth their continuous review – that is, when the ROI of user influence is on the lower end.  

Even an automated system that is 99% right will be wrong 1% of the time, and that might be a bit much for a business where quotes with 100 recommendations or more are common. If your business is looking for a mechanism that allows them to update prices on a consistent basis with minimal human oversight, can trust uncertain output, and can tolerate a certain degree of error, then a dynamic system using a black box approach might be a good fit.

With that said, many businesses today are not at this stage yet, as the type of confidence and “guts” needed for these black box approaches comes only after a long period of experience with data-driven solutions when the proven benefits trump the skepticism of its inconveniences.  

The degree of clarity users have in their pricing process presents another feature of AI your company should keep in mind when navigating the data-driven pricing options in the marketplace – especially in the early stages of pricing maturity when trust in the outcome is core to the successful adoption of new technology. 

5. AI can’t be a “silver bullet” for every pricing problem  

One of your main tasks as a business when firming up your pricing strategy is to consider when it makes sense to use a computationally intense data-driven analysis — and when it doesn’t.   

If you have limited resources for price management, we recommend thinking carefully about using data-driven solutions for tasks which either do not warrant a high degree of analysis or do not address business objectives in a significant way.  

Despite how tempting it can be to default to intelligent software for all your pricing needs, the reality is that often, the simple or low-impact pricing tasks don’t need it. Setting up a simple cost-plus margin rule of 30% may be a perfectly acceptable way to price a product that you sell once every five years. On the other hand, with a product you sell 50+ times a day, something which has a significant impact on your bottom line, you may want to take a different approach. To better support those kinds of decisions, it would then make more sense to use pricing AI to get granular in your analysis and figure out how to optimally charge certain customers in certain situations at a highly segmented level.  

The bottom line is this: artificial intelligence is here to make your life easier, not the other way around. If simpler methods can easily carry out certain pricing functions, there is no reason to overcomplicate a setup that works for you. Save your resources for the decisions that have the largest impact.  

The Function of Pricing Artificial Intelligence Is Simpler Than You Think, and Should Be Treated That Way  

At the end of the day, it’s not necessary to understand the equations driving intelligent software perfectly. But you should understand the limitations and the general logic of pricing AI processes, so you know how to better leverage them.  

Contrary to popular belief, we don’t find artificial intelligence all that sexy. After all, pricing AI is just part of an overall pricing mechanism that allows businesses to pull the pricing levers, understand what’s going on in their business at a granular level, and adjust the levers as necessary. And we find beauty in that simplicity. 

Interested in taking a closer look at some of the terms used most in this article? Read our article below to find out how machine learning, AI, and price optimization are related. Or to revisit the truly cool things AI can do for you, consider checking out this article.