With advances in tools and technologies to help mine larger and larger available datasets, algorithms are increasingly used across many business functions. However, if not managed appropriately, too much of a good thing can backfire. This was the message of a recent Wall Street Journal report by Deloitte, On the Board’s Agenda: Board Oversight of Algorithmic Risk. This article, the first of a two-part series by Lee Nyari, examines the dangers that Deloitte outlines in its report and offers distributors some solutions for reducing algorithmic risk in price management.
Algorithmic pricing refers to the approach of using sophisticated algorithms for invoice price management. This approach seeks to minimize human involvement, such as manual price overrides or one-off exception records in pricing systems. This contrasts with more traditional pricing approaches, where default system prices reflect high-level business strategies, while invoice prices are routinely managed by sales professionals. Algorithmic pricing in distribution is often viewed with a healthy level of skepticism. This is not surprising, given the mixed track record of algorithmic pricing in distribution.
While many success stories exist, industry insiders are also keenly aware of algorithmic pricing approaches that failed to deliver targeted results – or worse, that caused harm and major disruption to business operations. As Deloitte observed in its Wall Street Journal article, “algorithms can, and do, go wrong and can have serious adverse effects when they do.”
Pricing algorithm failures range across distributors of all sizes and of all industries. The pricing algorithms involved are sometimes developed internally by the distributors themselves, or they may be brought to the table by consultants/pricing software vendors.
Our discussion relates Deloitte’s observations about algorithmic risk to our experience with algorithmic risk in distribution pricing projects. In part one of this two-part series, we identify five key considerations regarding algorithmic risk in distribution pricing. In part two, we will review the stories of two distributor pricing projects, to demonstrate how different distributors practically deal with algorithmic risks in real life pricing projects.
Throughout the entire discussion, we offer some words of advice on managing algorithmic risk, as distributors pursue their own pricing journeys.
1. Pricing is a useful and powerful business lever in distribution
A key question Deloitte asks is: “what are the potential impacts if the algorithms go wrong.” Pricing is a high-stakes game. A successful pricing project can have major favorable impact on a distributor’s business performance. Ineffective pricing practices can also create a lot of pain. They can cause strategic damage, such as weakened strategic customer relationships, market share losses, low morale, etc.
Weak pricing practices are more frequently being exposed due in large part to the growing popularity of e-commerce, and the resulting increased pricing transparency in the distribution industry. Combine pricing’s importance in driving performance, the industry trends toward price transparency, and the emphasis often put on the mathematical sophistication of toolsets to analyze large and complex distributor pricing datasets – and you can quickly see why algorithms (and algorithmic risks) represent a major consideration in distribution pricing
2. Excessive price complaints may signal algorithmic risk
Deloitte advises us to look for signs that algorithms may not be functioning effectively, such as “complaints about them from constituencies such as customers, suppliers, employees.” They suggest that when signs of problem algorithms are spotted, the business should conduct an independent review of the relevant algorithms.
Distributors vary in how they respond to price complaints from sales reps and customers. Some simply allow for habitual manual overrides of the outputs of their pricing algorithms. These distributors might have “given up” trying to fix their pricing algorithms, or their management may believe that the ongoing flow of price complaints is somehow normal.
Seasoned executives understand: “some push-back on pricing is healthy – otherwise you are not pricing to resistance, and you are leaving money on the table.” It is true that some noise around pricing is to be expected. Still, if pricing is a constant source of headaches in certain parts of the business, then this may be a sign that the relevant pricing algorithms could be misfiring.
Several other types of corrective actions are also frequently prioritized before reviews of pricing algorithms. They include stepping up controls to force the use of the pricing recommendations produced by the algorithms, and/or investing in pricing-related training programs for sales reps. While some of these types of corrective actions can be helpful, let’s explore some situations where reviewing the pricing algorithms themselves might be a more effective response.
3. Pricing algorithms may be insufficiently market-informed
As any distribution sales rep will tell you, effective pricing is about more than just mathematics. It requires, among other things, an understanding of markets and customers, as well as the particulars of the offerings themselves. This cuts to the way pricing algorithms are developed. Pricing algorithms do not replace industry expertise. The algorithms are most effective when they are deeply informed by industry experts’ judgment.
Algorithmic risk increases when knowledgeable market experts do not provide extensive, direct input into model design decisions. Too often, however, their input in pricing initiatives is limited to providing data, being interviewed and validating results. When it comes to pricingalgorithms, the devil is often in details (assumptions, weights, etc.). Ideally, these experts should have direct input on these important details. If there are frequent price complaints in specific business areas, algorithmic risk may be reduced by following Deloitte’s advice: conduct a review of the relevant pricing algorithms, including the minutiae about weights, assumptions, etc., and do this in a meaningful way, with the involvement of relevant market experts.
Unfortunately, the pricing algorithms are kept in a “black box” in some popular pricing software packages. This can make it impossible for distributors to independently review these algorithms.
When distributors raise potential issues with their algorithms, some pricing software vendors reportedly push back on change requests – particularly if the type of change would go beyond simpler tweaks or adjustment. These kinds of issues have led several large distributors to discontinue pricing software contracts with “black box vendors,” at times before prices ever hit the market.
4. Popular mathematical approaches for developing pricing algorithms could stand some updating
Other reasons some pricing algorithms underperform expectations relate to the mathematical approaches themselves. Some popular distribution pricing solutions were first developed decades ago. Certain mathematical approaches embedded in them could stand to be updated, so they better reflect recent changes in marketplace dynamics and more strongly leverage recent advances in pricing science.
Such investments to improve the market-alignment of pricing algorithms not only help decrease algorithmic risk but can also reduce the risk of the pricing algorithms being rejected, ignored or overridden. Here we discuss two important aspects of popular approaches to developing pricing algorithms, which could stand some updating
Approach 1:
Over the past decade, distributors have seen major shifts in customer behaviors. Price transparency is the new norm: customers research prices on distributor websites and make use of electronic means of comparing prices. The algorithms in certain distribution pricing solutions have arguably not kept pace. These algorithms continue putting most weight on factors other than such key external data/competitive prices.
They continue to be anchored in analysis of internal distributor datasets, such as customer size and industry, item velocities and cost ranges, and distributions of margins/prices invoiced in past transactions. Undoubtedly, there is a lot of value in analyzing internal distributor data. However, the world has changed. Competitive pricing data is increasingly available, and it is now often a primary driver of customer purchasing behavior.
These external competitive data points should have a commensurately powerful impact in pricing algorithms. Just ask Amazon if its pricing algorithms treat competitive web prices as an afterthought/adjustment. For distributors that compete more heavily in or with e-commerce, this can be a big miss. Many distributors’ algorithmic risk is elevated because their old-fashioned pricing algorithms do not properly align with current purchasing behaviors. In fact, their old-fashioned pricing algorithms may be putting them at a competitive disadvantage, relative to e-commerce competitors with pricing algorithms that better align with new customer purchasing behaviors – like Amazon, for example.
Approach 2:
Cube/matrix-based segmentation approaches are quite popular in distribution. There are valid reasons for their popularity, but science has evolved, and pricing cubes may no longer suffice to provide a competitive advantage. First, let’s briefly review the advantages of pricing cubes.
Pricing cubes may be formulated by way of mapping available distributor pricing data into file schemas. The professional judgment of pricing and market experts can be used to deploy scoring methods and assign weights to available raw data/attributes.
This data mapping process can result in a manageable number of “strategic pricing dimensions” (examples of these dimensions may include strategic versus less strategic customers, sensitive versus less sensitive products, etc.). These cubes/dimensions may even be developed with or without using statistical tools (higher, but less frequently deployed versions of these cube-based solutions include statistical analyses to help identify or validate drivers of price sensitivity using available data). The process of building/configuring cubes can be streamlined, making for efficient pricing project implementations.
Particularly the more advanced versions of the cube-based pricing approaches can be helpful: they are rather sophisticated compared to segmentation schemes developed internally by distributors.
But pricing science has evolved. To generate market-aligned pricing algorithms, top-ranked B2B price optimization solutions (Zilliant, Vendavo, PROS, etc.) have, for well over a decade, stopped using cube/matrix approaches for segmentation. Segmentation trees are developed in these tools, instead of cubes or matrices. This reflects widespread agreement among pricing scientists: segmentation trees allow for more precise, more flexible, and ultimately more market-aligned segmentation schemes. Segmentation trees are built using specialized multivariate regression tools (for more discussion on cube- versus tree-based segmentation, see page 5 of the data sheet available here.
As these vendors suggest, decision trees are best developed in conjunction with discussions with market experts about how drivers of price sensitivity vary across different segments of the business. Accordingly, segmentation trees look very different across different distribution businesses.
In brief, price structures anchored in cubes or matrixes may be less effective in how strongly they reflect the true complexities of relevant markets. In such cases, algorithmic risk can be reduced, and the precision/market-alignment of prices can be improved, by switching analytics approaches from cubes to segmentation trees.
Numbers don’t lie – but this doesn’t mean that distributors should blindly trust mathematical price optimization toolsets. Distributors should avoid being taken in by statistical jargon, scatterplots and examples of past success stories – even if they may only have limited backgrounds in statistics and only a high-level understanding of how the vendor’s mathematical models for deriving the pricing algorithms. Numbers don’t lie, but models, averages etc. can. If the signs are present, it is healthy to question the validity of the mathematical models and associated pricing algorithms.
5. Distributors erroneously assume their pricing algorithms include price optimization technology
Finally, some distributors make erroneous assumptions about the pricing solutions they are investing in. With some exceptions (again, think Zilliant, Vendavo, PROS, etc.), vendors of distribution pricing solutions do not talk about estimating elasticities or demand curves in their solutions – which are arguably hallmarks of true and complete statistical price optimization toolsets.
Still, given the marketing and black-box nature of the pricing package under consideration, some buyers erroneously assume that they are getting just such a thorough, bulletproof statistical price optimization toolset.
Investing in a “Cadillac” statistical price optimization tool may be overkill for most distributors. Substantial improvement may be achieved, and has been achieved, using somewhat simpler but still quite sophisticated approaches.
Some popular distribution pricing software may be described as a sophisticated toolset to implement a set of pricing strategies which have worked at other distributors, which are (at least at the highest levels of the “strategic dimensions”) pre-defined in the solution, and which are supported by an advanced backend analytical platform (these backend analytics tend to use basic statistical measures, such as distribution means and standard deviations to identify outliers).
These solutions are potentially quite powerful. Still, distributors should not view them as “bulletproof” price optimization tools, and their use does not mean that algorithmic risk is minimized in the businesses.
Part two of this series reviews the stories of two distributor pricing projects to demonstrate how different distributors deal with algorithmic risks in real life pricing projects.
Lee Nyari is managing partner of The Innovative Pricing Group, a consultancy offering strategic price management solutions for distributors. Reach him at lnyari@pricinginnovation.com.
