INTRODUCTON

The pricing process in any business involves the most profit-sensitive decisions as it directly influences sale–purchase transactions. Companies can no longer afford to fail in their pricing decisions; all products and services must be priced right, all the time. To make this process more challenging, the right price today may not be the right price tomorrow, as business conditions continuously change.

Distinctive advantages are achieved when prices are always right. Otherwise, unprecedented pressures emerge from competitors and customers. This is an urgent issue as technology continues to rapidly advance, acquiring and using abundant amounts of data, and helping to create more knowledgeable customers and competitors. In response, companies now need to have well-designed pricing machines, just like farmers need tractors.

Pricing machines need to be able to automatically sense the markets and respond to the shifting conditions without delay. Adaptors have been gaining and will continue to gain instantaneous benefits. Those who resist will blame the market economy, government or everything else under the Sun for their pricing failures.

Pricing is a very old idea, but science-based, data-driven pricing is relatively new. The process relies on the law of supply and demand. In fact, scientific pricing¹ is perhaps the most successful, repetitively used, and largest application of micro-economic theory to date.

Numerous forces are driving the pricing field forward, including better data, increased competition, additional customer choice, modern computing power, more products, shorter product lifecycles, more substitutable products, faster new product introductions, and less government and international regulation. In addition, scientific pricing cuts across many disciplines — including mathematics, operations research, statistics, economics, marketing, finance, accounting, computer science, sociology, politics, and psychology. Advances in these disciplines will also have a positive influence on the pricing field.

This paper calls attention to some important pricing issues and makes 'guesstimates' about emerging trends. It summarises lessons learned as we worked to put the art and science of pricing into practice across many industries.

Several warnings are appropriate:

• First, any predictions about a complex topic like pricing will most likely be accurate less than 99 per cent of the time.
• Secondly, although predictions are highly interconnected, they are presented in isolation due to limited space and an interest in clarity. For example, segmentation forms a foundation for any pricing system; but this paper ignores the topic when discussing competition, even though segmentation techniques can identify and price competitive products.
• Thirdly, predictions are loosely ordered with no particular meaning.
• Fourthly, the word 'products' is used to indicate 'products or services.'
• Finally, some predictions may be a stretch, while others may already be happening for some products and markets.

The reader is also referred to numerous futuristic papers presented in the previous issues of this journal (see, eg, Lieberman, 2002; Shoemaker, 2003; van Ryzin, 2005; Ratliff and Vinod, 2005).

ZERO AUDIENCE FOR 'THE PRICE IS RIGHT'

Bob Barker retired after hosting his final episode of 'The Price Is Right,' which was broadcast on 15th June, 2007 after 35 years of successful programming. We do not currently know whether CBS will find a successor or abolish the programme altogether; but as scientific pricing sweeps across many industries, this show has no real future.

The Price is Right would have had zero audience if contestants were asked to estimate the price of an airline ticket, as the price varies based on many factors, including time of departure (season, month, week, day, and hour), time of booking, seating capacity of the aircraft, quality of the travel (nonstop, one-stop, or connections), selling channel, number of seats requested, and the schedules of competing airlines.

Perhaps the show could continue in a session at INFORMS National Meetings, where contestants with PhDs and laptops crunch numbers through sophisticated market response optimisers to guesstimate the price of a product under given market conditions. Still, how many would attend?

THEORY OF PRICING RELATIVITY

Pricing is necessarily relative. Customers make purchasing decisions based on relative prices, not actual or specific prices. Therefore, competition is the most influential factor and will shape the postulated theory of pricing relativity. Competitive actions and reactions will prove greater challenges in the future as real-time competitive data become increasingly available to everyone (sellers, buyers, and competitors).

Most available pricing systems concentrate on internal patterns of historical demand and essentially ignore competition, primarily due to the lack of accurate, timely, and consistent data. The common hypothesis is that, although the pricing system ignores competitive prices, it can still sense the market through its own historical prices and demand, and subsequent optimisations will appropriately respond to potential competitive actions. After all, each competitor does not know every other competitor's prices; and most customers may not currently shop around. As long as competitors and customers behave consistently, pricing systems that ignore competition can produce profit-generating prices.

These views will continue to have some merit as having any pricing system is better than none at all. Ignoring competition, however, often gives rise to unexplained noises in one's own historical data, which leads to higher demand prediction errors. In addition, demand models cannot be made sensitive enough to detect or estimate the effects of competitors' reactions in a timely manner, and a slow response can be very costly.

As real-time competitive data becomes a reality, the need to incorporate competition into pricing systems will be a key issue. We will need a new theory of pricing relativity, which will recommend the best response based on the pricing equivalent of space (eg, market conditions) and time (eg, long-term profits).

The economic literature is filled with papers investigating competitive pricing based on game theory. The new theory of pricing relativity might use game theory for certain products and markets, but only if its stringent assumptions can be relaxed to allow operational-level decision making.

More likely, a new theory of pricing relativity will devise appropriate mechanisms to collect competitive data into a form that can be automatically consumed by pricing systems. It will tell us which products or services must be competitively priced. It will extensively utilise customer loyalty programmes and recommend new ways to reward customers so that they remain loyal.

The new theory will help us to design and choose more accurate demand models and follow-up optimisation procedures to recommend optimal prices under multiple conditions, not just in response to what the competition does. The theory will understand product costs, qualities and positions, estimate competitive responses under different scenarios, initiate price increases or reductions, and ignore, go half way, or match competitive prices in a profit-maximising manner. As pricing ability can be enhanced with lower cost structure (eg, low-cost carriers or Dell's direct marketing), the new theory will explicitly consider cost structures while pricing, and the phrase 'revenue management' will no longer be appropriate to describe our field.

INFORMATION ASYMMETRY

Traditional pricing models necessarily assume that sellers have more information than buyers; however, this is not the case in some situations. In certain business-to-business transactions, buyers have more information than sellers, and are able to request competitive bids and/or negotiate. For commodities, the market sets prices, and neither sellers nor buyers have any pricing power.

Naturally, the degree of information asymmetry that exists has major implications on the type of pricing models to be implemented. If sellers have more information, the traditional pricing models are used to recommend an optimum take-it-or-leave it price. If buyers have significant information, auction or negotiation models could be employed. In some situations, an optimal price band (ceiling, target and floor prices) is recommended. If both sellers and buyers have perfect information, financial models are used. In this case, the system recommends when to buy or sell in the spot market and when to use the futures market to hedge a purchase or sale.

In the future, the degree of information asymmetry about each product may dramatically shift. Scores of products will become commodities traded in very efficient markets. There will be an explosion of ingenious market types like eBay. Future pricing systems will be able to jointly optimise prices of related products with different degrees of information asymmetries and apply the most appropriate pricing models to produce optimal recommendations.

Although airlines have had the most experience in advanced pricing technologies, anyone that experiences on-line travel sites can easily notice questionable prices. Examples are plentiful where the difference between the highest fare and lowest fare in a given origin and destination itinerary is offensively high. Perhaps an airline seat has become a commodity, so that in some markets the continued use of the ancient revenue management models may be more harmful than helpful.

A CHICKEN-AND-EGG PROBLEM: PRICING SEGMENTATION

Segmentation classifies customers, products, orders, and time so that those with similar attributes are mapped to the same category. Effective segmentation is a foundation for any scientific pricing system. In fact, pricing opportunities are virtually impossible to find without segmentation, as blanket price increases or decreases cannot consistently result in higher profits.

Segmentation remains both art and science. The art relies on business judgment to identify a viable set of attributes for defining pricing segments. These attributes may include sales rate, customer size, selling channel or market type, regional sensitivity, competitive intensity, cost-to-serve, product profitability, product lifecycle, bundling tendency, and product substitutability.

Sometimes, segmentation variables that are known to be important from a business perspective are not statistically significant because of data quality and/or scarcity. In these cases, variables are incorporated as pricing rules. For example, it is reasonable to think that a renter would pay a $20 per month premium for a pool view in an apartment. As rent data are sparse, it is difficult to statistically quantify and price apartments with pool views. In this case, a predefined pool view premium could be added in a pricing rule table and incorporated into the final rent asked for each unit with a pool view. The pool view premium could also be derived by other methods, such as a survey or price experimentation.

The science of segmentation uses statistical clustering and/or decision-tree techniques to reveal relationships among postulated attributes and identify appropriate micro-segments. These algorithms, however, significantly over-segment, as splitting cannot utilise price sensitivities as criteria (Biberoglu et al., 2005).

The issue arises because segmentation is a chicken-and-egg problem: Segments are needed to calculate price sensitivities; price sensitivities are needed to identify segments. A new statistical clustering method must use price sensitivities when defining splits to identify discrete buckets for each attribute. We predict that an iterative method will be developed to accomplish better pricing segmentation, although the method will rely heavily on accurate measurement of price sensitivity, which is very challenging in itself.

PRICE SENSITIVITY GENES

We might someday see a story titled 'Price Sensitivity Genes Discovered' in our morning newspapers. As we learn more about the human biology — particularly about the brain and human genome — this kind of discovery may not be just a remote possibility. Perhaps certain genes do control our buying behaviour. How would then our field change? Would governments allow companies to use pricing models that accounted for price sensitivity genes? Would we need a century for deregulation? Or, would we be taxed based on our price sensitivity genes?

There are, of course, different price sensitivity genes for different products. After all, how would your price sensitivity genes behave if your hip needs to be replaced? What if I want price sensitivity gene therapy so that I could be more careful with my money? A talented writer asking these kinds of questions may end up writing an entertaining science-fiction novel; but we cannot get carried away, as this paper is already past the word limit set by the editor-in-chief.

The point is, an accurate and ongoing measurement of price sensitivity could substantially improve the effectiveness of pricing systems. In practice, expert opinion, customer surveys, and market response modelling using experimental and/or observational data are most commonly used. Each of these methods will continue to have its own efficacy; however, successful modelling and engineering of market response models will be significantly improved, given a higher quality and more data.

Market response models are preferred due to their accuracy and low maintenance requirements. With them, price sensitivity is estimated by building a regression-based response model, including a customer response variable, pricing variables, segmentation variables, and other exogenous variables.

Pricing variables can be functions of discounts, actual prices, relative prices (eg, with respect to an average competitive price), and/or promotions. There may also be interaction terms for cross-sensitivities. The main goal is to be able to relate pricing variables to other variables (particularly customer response variable) in a statistically meaningful way. Thousands of segmentation scenarios and numerous forms of market response functions can be tried to identify the most stable and accurate estimates of price sensitivities.

PERFORMANCE EVALUATION

A pricing system sits at the heart of any business. It requires enormous investment and resources, and the pricing recommendations it produces can make or break the company. A pricing system can no longer include black-boxes; complete visibility is needed. In addition, it is vital to have a yardstick that closely monitors the profit performance that a pricing system delivers. As market environments become highly dynamic, a future pricing system will have the ability to assess itself constantly, even making self-correcting adjustments or recommendations as needed.

The make-up of a performance measurement module will be largely industry-specific, but its main components must include monitoring business statistics, predicting demand and profit contributions. Sample business statistics include pricing, utilisation, and user compliance indices. Each statistic can be continuously evaluated using measures such as averages, dispersions, percentiles, skewness, kurtosis, knee-jerkiness, and year-over-year changes.

The performance measurement system should have the ability to slice and dice each business statistic in multiple ways, so that the cause of any perceived issue can be easily understood and acted upon. In addition, real-time customised alerts can be generated when a given statistic is out of the user-specified and/or statistical bounds. Both data-mining and quality control techniques could be used to support this type of functionality.

Price recommendations largely depend on the accuracy of the demand models; thus, the quality of a system's predictive ability must be scrutinised by using statistical techniques such as hold-out sampling. Regressing historical observations to a model and accomplishing reasonable fit scores (such as adjusted R² and F statistic) may be necessary, but not sufficient. It is well known that any data series can be fit into a high-degree polynomial model to achieve perfect fit scores, but this does not mean that the high-degree polynomial model can always predict well. The issue of predictability of demand models should receive more scrutiny by the pricing science community.

In addition, the profit performance of the pricing system must be evaluated using simulations, 20-20 hindsight, sampling approaches, or other comparison methods. For example, the optimal pricing policy could be compared with a naïve pricing method, and the difference continuously monitored.

Furthermore, system-generated prices may not directly turn into income for the company; additional discounts, allowances, payment terms, volume bonuses, and other incentives may be given to customers by downstream systems. These transaction-level complexities need to be understood, and full economies of each and every transaction must be made visible and evaluated to minimise unnecessary price leakages.

In summary, future pricing systems will have a renewed focus on evaluating and monitoring the performance of related business statistics, demand models, optimisation, and additional pricing rules. The performance monitoring aspect of the scientific pricing is mission-critical for its continuous success.

PRICING EDUCATION

Moving beyond practice, teaching scientific pricing as part of a formal graduate-level programme is long overdue. The need for highly skilled pricing experts is increasing exponentially. There are now pricing departments in most major companies. Job titles range from revenue management analyst to vice president of pricing science. Most are skilled pricing practitioners and experts in their own industries, but they are, at best, only half-educated in scientific pricing.

Research in the field is extensive. Two recent books, Talluri and van Ryzin (2004) and Phillips (2005), are excellent resources. Many business schools offer pricing courses as seminars, but courses with pencil-and-paper-tests are still needed. The author developed and taught such graduate-level courses in the Operations Research Program of The University of Texas at Austin. Students had a vast interest in the topic, and a majority of them joined the pricing workforce or are conducting research in the area.

The field is extremely rich in context, covering many disciplines, from applied mathematics to econometrics to statistics to psychology. These disciplines could be synergistically integrated into a carefully crafted pricing-focused curriculum. The curriculum of such a programme is beyond the scope of this paper, but we believe that the potential graduates of such a programme may find jobs waiting right after they receive admission letters from the department.

ADDITIONAL TOPICS

Owing to limited space, we could not explore additional areas that will undoubtedly expand in the near future. At a minimum, they are worth of a brief mention. Again, the following list includes interconnected topics and is not presented in any particular order:

• Optimal merging and alignments of strategic, tactical, and operational pricing decisions
• Coordination of promotion management and pricing
• Joint price optimisation and demand alignments across multiple products, services and locations
• Optimisation of price negotiations under different bargaining mechanisms
• Joint pricing and inventory management
• Joint optimisation of buy-side and sell-side processes
• Contract pricing
• Group pricing and management
• Better estimation of 'true demand' using partially observable data (eg, better unconstraining methods)
• Incorporation of uncertainty into demand models
• Integration of product bundling, affinities, and substitutions
• Incorporation of pricing rules and intelligent heuristics
• Enhanced what-if capabilities

ADAPTABILITY

So far, this paper has focused on the direction of potential changes, but an equally important matter relates to the speed of the journey. We are exposed to the most rapidly changing business environment humans have ever experienced. Therefore, pricing systems must be able to adapt to the changes in data, products, markets, services, promotions, competitors, customers, etc.

The era of one-forecaster-and-one-pricing-model-fits-all is over. If system adaptability is ignored, the limitations of legacy pricing systems will be a major handicap. A company may end up spending prohibitive amount of money to replace or modify its legacy pricing system when the need arises; and the need will arise, sooner than we expect.

SUMMARY

Believe it or else. As part of the information revolution, we are now in the midst of a major pricing revolution. As science and technology rapidly advance, so, too, does the need for the application of scientific pricing.

More products and services will be priced based on scores of interwoven factors. Better information about products and services will be available to sellers, buyers, and customers. Competitive pricing will take centre stage.

Future generations of customers will become more price sensitive, empowered by an increasing amount of information. In fact, more products will be sold as commodities in perfectly competitive markets where customers are infinitely price sensitive.

Advances in segmentation and price sensitivity measurements will firmly improve the way pricing decisions are made. Better yardsticks will be developed for continuously monitoring business statistics and pricing systems' ability to predict and generate profits. Complete visibility will be a must for pricing systems.

Universities will develop programmes with curriculums focused on pricing science, and students will anxiously apply to these programmes. Research in the field will continue to explode. More pricing conferences will attract record numbers of talks and attendance.

The ability to speak with precision and certainty about the future is limited, but that is not a valid excuse for silence. Growing evidence indicates that scientific pricing is bound to experience roaring changes. For the purpose of this paper, we have tried to be more insightful with our predictions than completely correct. At the very least, if reading this paper stimulates readers' thoughts about the long-term prospects of scientific pricing, we have accomplished our goal.

Notes

¹ This paper uses scientific pricing, pricing science, and revenue management interchangeably.

References

1. Biberoglu, E., Ozkan, S. and Kuyumcu, H. A. (2005) 'Procedures for identifying segments based on homogenous price sensitivity', INFORMS National Meeting, San Francisco, CA, USA.
2. Lieberman, H. W. (2002) 'Revenue management: what lies ahead', Journal of Revenue and Pricing Management, 1, 2, 189–195. | Article |
3. Phillips, R. (2005) Pricing and Revenue Optimization, Stanford University Press, Palo Alto, CA, USA.
4. Ratliff, R. and Vinod, B. (2005) 'Airline pricing and revenue management: a future outlook', Journal of Revenue and Pricing Management, 4, 3, 302–307. | Article |
5. Shoemaker, S. (2003) 'The future of pricing in services', Journal of Revenue and Pricing Management, 2, 3, 271–279. | Article |
6. Talluri, K. and van Ryzin, G. J. (2004) The Theory and Practice of Revenue Management, Kluwer Academic Publishers, Dordrecht.
7. van Ryzin, G. J. (2005) 'Models of demand', Journal of Revenue and Pricing Management, 4, 2, 204–210. | Article |