Sep 9, 2014

The Winding Roads of Car Sales Forecasting


 Forecasting sales of a car dealer is a tough business, much tougher than predicting the total market. Winning a car race is matter of right combination of engine tuning, tires type and pressure, quality of petrol and the other fluids, the race track parameters, weather conditions, the pilot mental and physical status and many others as well as all these of the competitors. Similarly, sales results depend on plethora of interconnected factors that makes forecasting it Heracles-grade labour. However, the challenges in predicting car sales are not unique and it is a good illustration of some of common problems.

The Stig of predictive models
 Find more about the mythical Stig here. A good sales forecasting model should include all aspects of the the business grouped in:
- Brand and car model related
Appeal of models, brand image, brand and model preference, price and availability of spare parts, etc.
- Marketing related
Customer reach, quality and quantity of advertising, promotional activity, special deals on offer, participation in special activities or events, sponsorships, etc.
- Dealer related
Product mix, number of showrooms, coverage of population, sales force, quality of pre-sales, sales and post-sales activities, etc.
- Demographics -
The dynamics in the customer base, changes in attitudes, etc
- Financial
Prices, discounts, interest rates, discounts
- Market trends
Sales of a dealer cannot exceed the market demand as an obvious restriction as a first, changes in preference of car types and parameters is another
- Macroeconomics and social framework

The perfect model should also include the competition activities across all the groups.

Flat tires and oil leaks
It is easy to come out with the characteristics of a good model but reality has its ways to deal with the perfect plans by serving a load of problems that do to fit even in a Toyota Sequoia. The simplest and most common of them are the data availability and data granularity. The data series that would serve best value for the model are often not available or do not come with a granularity to allow generating useful insight. Data collection methods and definitions change over time to render as not comparable the figures from different periods.

As the race track and weather conditions limit the speed on the race track, market trends and changing context directs the sales one way or another. The problems come with and expressing these influences in numbers. The mix and variety of car models in the segments is an example for changing market trends as new sub-segments keep emerging and existing ones growth in variety in response of changing customer preferences and search of new markets. Not so long ago there were no cross-overs and a variety of SUVs while now they are affordable, in fashion and part of product line of every car maker. Sales of a new car type cannibalize the ones already in place - like the sedans for the last example. The transition of preferences from one car type to another is a problem that could give you an empty battery on a remote country road. Another example for changes in market trends is the increasing preferences less polluting cars - Prius would have never been such a success without it. The question of the weight of lower CO2 emissions in the decision to buy however is difficult to answer. I guess it depends on the hipsters per capita and green party parliament representation but there is much more to that. Market trends seems to be relatively well defined as there are many organizations that release outlook reports on that but for a forecasting model their depth and coverage is not good enough.

Obviously, the car price is one of the most important factors. However, including it in a model is not straight-forward as one might think as it is just a part of the total offer that also includes promotional discounts, extras and leasing options as down-payment, residual value, installments size, interest rate and period. These components are in a constant move to hamper finding the combination providing the best insights and contribution to model accuracy. As an additional "bonus", customers in different segments differ in their preferences for the offer component mix.

Transition of brand preferences is another lorry on a mountain road as they are influenced by fashion, magazines, movies and TV shows. Top Gear or Fast and Furious franchise have significant impact on a brand and model image and are uncontrollable factor therefore impossible to include in a forecasting model. Changes in preferences are part of the more general type of problems with difficult to quantify but influencing factors. Fleet sales are good example as they depend on the quality of the salesperson, the buyer, previous experiences with the brand or the dealer, personal relations, politics and many others, all of which are difficult to impossible to quantify but have significant impact on the outcome. Relationship with the partners is another as there are many intricate details related to doing business between the parties that make the difference between good, very good and excellent sales but are difficult to put in Excel formula.

Engineering a good sales forecasting model also suffers from the fact that business is run by people - their motivation, attitude, respect to company values and procedures matter a lot. Even a Veyron would finish after a Polo if the pilot is more interested in adjusting the radio than winning the race.

Forecasting is complicated by other limiting factors like budget on marketing or available advertisement space. The non-linear response of sales to marketing and other activities add another dose of uncertainty in the forecasting model. There is also anti-monopoly legislation in some countries that restrict a brand market share and it brings out the unknowns of activities to meet them.


The finish line
The general and specific problems faced in forecasting a dealer sales shapes the general approach for forecasting, the applied methods, the forecast span in the future, forecast reliability and the place of the forecast in the decision making process. As George E. P. Box stated it , all models are wrong but some are useful. The success of a predictive model is not only in accurately predicting the sales but in the quality of generated business insights as well.

Update: Read some details about the used car market in Bulgaria I dug up and compiled here.

No comments:

Post a Comment