Decision Science KPMG helps retailers double the accuracy of their revenue predictions for new branches, saving them CZK 950 million in investment and operational costs within a three-year horizon.
After a speech in 1974 at the University of Texas at Austin, the founder of the McDonald’s empire, Ray Kroc, asked his audience in which field they thought he was doing business. The students laughed and one loudmouth called out that everybody in the world knew that Ray Kroc was selling hamburgers. Mr Kroc just smiled and answered: “Ladies and gentlemen, I’m not in the hamburger business. My business is real estate.” During the resulting discussion, he explained that even though his main goal indeed was the sale of hamburgers, pieces of real estate and their location constituted the main key to the success of the individual franchises and of the company in general.
While franchising models and retail chain concepts have been successful for many decades, their success still fundamentally depends on the entrepreneurs’ ability to select the right location for their brand. Growing competition, changing customer behaviour and growing real estate prices and rents significantly increase the necessity of accurate estimates of demand and revenue in given locations as well as the resulting return on investment of a new branch over time.
Inaccurate financial estimates can be fatal for the financial operations of a newly opened store. If turnover has been overestimated, the entire investment may be bound for failure, as the branch will never be able to operate in the black. Demand that has been underestimated may in turn result in forgone profits, as the branch could have easily been larger with a wider assortment of goods or, if nothing else, with more advantageous financing than initially planned. Several branches opened under such unrealistic expectations can easily drive an otherwise healthy company into the ground.
Similar problems were well known by KPMG’s client, a company annually opening ca. 30 new natural fast-food outlets. This client’s revenue estimates for one to three years in advance were on average off the mark of reality by 30%. In fact, It may also happen that a retailer’s expert estimates will under- or overestimate sales by more than 100%, which can easily come to pass if the new branches are opened in locations where the company does not have much experience. If you are successfully running a chain of snack-bars in a city, then the opening of a branch in a small town or of a small kiosk at the train station will confront your business with completely new challenges in the form of, e.g., different customer segments and different turnover rates of the goods you sell. These will translate into the need to change your purchasing as well as selling strategy. Your business approach will also have to differ depending on your local competition. Simply put, a student waiting for the delayed Prague-Brno train will display different behaviour and have different needs than the manager who is late for a meeting and hasn’t had the chance to take a bite out of her morning croissant or to take a sip of his latte.
To deal with the problem our client had with their newly opened branches, we used KPMG’s Decision Science approach, combining Big Data, machine learning and cloud technology. The first phase involved training a predictive model, which based on more than 10 000 comprehensibly variable signals estimated weekly revenue for the first, second and third year after opening and took into consideration growth as well as seasonal influences. The model’s prototype reduced the average error rate of the turnover predictions to 15%, which meant a double improvement in comparison to the expert estimates by the company’s management. Subsequent model versions continued to drive down the error rate to close to 10 %. This entire process took about ten weeks.
The main reason for the massive reduction in the error rate lies in the model’s ability to maximally utilise information from all 10 000 signals which in detail describe the characteristics of the real estate and location under consideration and together create a comprehensive network of links and interactions unique for every point on the map. The majority of signals stemmed from external data sources and included the social demographic and economic characteristics of a location’s inhabitants and workers, the numbers and trajectories of pedestrian and vehicle movements, the presence and type of competitive companies, retail shops and other entities, public transportation stops, photos, comments and sentiments in social media networks, local weather conditions, regularly occurring events and the distance to any of a thousand points of interest (e.g. road crossings, subway exits, ATMs, mailboxes, bicycle stands, etc.). All signals were also compared against the local direct and indirect competition.
At KPMG, our work is governed by the principle that the even the best predictive model is only useful if it can be applied in practice and if it supports its end-users in their every-day decision making. For this reason we designed and implemented the Analytics-as-a-Service approach within the Microsoft Azure cloud environment, helping our client to predict any planned location’s revenue. All it takes is the input of an intended location’s basic information, i.e. address, size, planned seating numbers, into a simple user interface and to submit this data to the system. From the outside, this may seem like magic as the system takes care of everything else: download and preparing of external data, counting of signals and the final scoring of the given location. The subsequent detailed report includes not only predicted weekly revenues for the first, second and third year of operation, but also estimated turnover according to individual products or operating hours as well as a list of already up-and-running stores which are similar to the ones in planning.
“The advantage of our approach is that from the client it does not require a huge investment into expensive technologies or the costly assembly of an analytic team. All remains outsourced at KPMG, including our data scientists”, so David Slánský, KPMG’s Partner in charge of Data Analytics. “For a monthly service charge, clients receive a tailor-made application which they may use in the company’s decision making processes as they see fit, without any other concerns. And we continuously calibrate and adapt our approach to current market conditions to assure maximum accuracy,” he adds.
In closing, it may serve well to repeat the mantra about the impact of accurate revenue predictions on a company’s effective investment into its own growth. Good, well prepared and properly timed investment decisions lead to positive economic results, while bad, incorrect, rash or insufficiently planned steps may mean the end of a company, regardless of management’s best intentions. In the case described above, thanks to the application of KPMG Decision Science, we helped the client reduce such risks to a minimum, while saving CZK 950 million in investment and operational costs within a three-year horizon.