Capabilities

null

We have built our data science capabilities around the following three areas:

Artificial Intelligence

At Autometrics, we use a branch of Artificial Intelligence (AI), called Machine Learning.  Machine learning is a method of data analysis that automates analytical model building. Using algorithms that iteratively learn from data, machine learning allows computers to find hidden insights without being explicitly programmed where to look.

Some of the most widely adopted machine learning methods are supervised learning and unsupervised learning and clustering:

  • In Supervised Learning the machine is trained by being given sets of input values and told what the corresponding output values (the answer) should be. The machine gradually learns a mapping between the input and output and when given a completely new set of input values produces an appropriate output answer (based on the mapping it has learned). For example, given Autometrics Shoppers data for the last 7 years (as the training input), AND the actual retail sales figures over the same period (the output), it may learn a mapping from one to the other and predict retail sales next week (this is a simplistic example – many other factors such as ad-campaigns and incentives have to be added to the input).
  • In Unsupervised learning, there are sets of input values as before but there is no output or answer that we can provide. Rather the machine tries to find hidden structure in the input values, i.e. it discovers recurring patterns, or clusters in the data. Certain combinations of variables co-occur more often than they should just by chance. For example, a simple pattern would be a correlation between certain zip codes and certain models, or a more complex one would be that at a certain time-of-year, given certain weather conditions, and if the oil prices are within some range then there is more interest in certain models. By definition, Unsupervised learning is much more complex than Supervised learning and requires significantly higher volumes of data. Autometrics highly granular shopper data at the zip code level, for every minute of the day since 2009, is ideal for this purpose.
  • In Clustering, the system takes a high-dimensional space of variables (e.g. the location of shoppers and economic conditions under which they express an interest to buy) and learns that the majority of the data actually clusters (i.e. naturally clumps) into several distinct groups. For example, buyers in Texas, during periods of drought may show interest in a different set of models than buyers in Michigan during harsh winters. Clustering algorithms naturally discover and update these clumps dynamically from the data rather than relying on preconceived human notions about shopper behavior. Again, Autometrics’ shopper data is ideal for identifying such clusters.

Statistical Inference

By statistical inference we mean drawing conclusions based on data, putting hypotheses and gut feelings to test, separating signals from noises, leveraging empirical regularities as opposed to chasing red herrings.

At Autometrics, some of the common questions addressed through statistical inference include:

  • Which makes and models are most frequently comparison-shopped by Shoppers? How does the pattern of comparison-shopping vary across markets and over time? By tapping into the fact that Autometrics Shopper data is available by minute and zip code, we have developed a statistical algorithm for inferring the degree of co-consideration between any pair of vehicles. The resulting pair-wise co-consideration statistics are then turned into a perceptual map, allowing automakers to visualize and monitor the competitive landscape of the automotive market.
  • How many incremental Shoppers did a TV ad spot generate? By aligning Autometrics Shopper data with media placement data, we have developed an event-study approach for inferring the number of Shoppers that can be attributed a TV ad insertion. By examining the relative impacts of a large number of ad insertions, we can help automakers determine the optimal ad creative and media placement strategy.
  • Which markets present the most potential for near-term Shopper growth? By comparing Shopper data across markets, we can infer the “demand frontier” for each make and model in each market, after taking into account each market’s demographic and socioeconomic characteristics and demand patterns in lookalike markets. The gap between the actual demand and the frontier shall indicate the potential for Shopper growth in each market.

Econometrics

Econometrics is the science of quantifying relationships between observed variables, often using time series-based modelling techniques.

  • Car buying represents one of the largest decisions that consumers make and is affected and influenced by numerous factors, including marketing, sales incentives, competitive effects, pricing changes, as well as a range of economic factors, such as interest rates, oil prices and environmental factors, such as weather patterns. Further, there are complex lags between the time that these factors occur (e.g. the launch of a marketing campaign, a change in price, a competitive model launch or a hurricane) and their impact on consumer purchase behaviour, which can often be long lasting.
  • Making a direct connection between these marketing, economic and environmental factors and retail sales is often impossible as too many competing factors obscure each other before a sale is made. By contrast, the immediacy of Autometrics Shopper data, which captures changes in consumer demand in real time, along with advanced time series modelling techniques, will reveal the true impact of each of these factors on future retails sales, by region and even DMA. Using such techniques, Autometrics Shopper data will find historical patterns and inform automakers what actions to take in response to current changes in demand and how these changes will impact future sales.
  • Which markets present the most potential for near-term Shopper growth? By comparing Shopper data across markets, we can infer the “demand frontier” for each make and model in each market, after taking into account each market’s demographic and socioeconomic characteristics and demand patterns in lookalike markets. The gap between the actual demand and the frontier shall indicate the potential for Shopper growth in each market.