Skip to main content Skip to footer

With 'black swan' events arriving more often and with greater impact, businesses must monitor the perishability of their data and the analytic models applied to it.

That means assuring the currency, accuracy and relevancy of their data and models. Knowing the lifecycle of these elements is critical for organizations seeking to anticipate and respond to unpredictable, severe changes more quickly and effectively than their competitors, perhaps even ensuring their salvation in the process.

Perishable data and analytic models make it much harder to anticipate and respond to sudden shifts in demand for products and services, the price or availability of raw materials, consumer sentiment, and whether customers and employees can access the business.

Businesses that keep their data and analytic models fresher can substantially increase not only their chances of survival, but of capturing a larger share of revenue and profits in the process. For example, we helped one global convenience store chain identify which products were selling most quickly at their stores during the pandemic. This allowed them to be sure they had enough of those goods in stock, and to place them near the checkouts so customers could spend as little time in the store as possible. This one insight drove about a 25% per customer increase in purchases of those products.

The deeper and more accurate insight a business can generate, the better it can provide more value internally and externally, such as providing disease-tracking data to public health agencies to help curb pandemics. The good they do can also make such companies more attractive to customers, employees and investors.

Measuring perishability

Understanding data and model perishability goes far beyond traditional measures such as age or recency. It requires continually assessing the three criteria that determine whether an organization is getting the real-time intelligence it needs to understand, adapt to, and even anticipate shifts in the business environment.

'Current enough' means:

  • Data that reflects the most recent changes that could have a significant impact on the enterprise. These might include the loosening or tightening of COVID-19 lockdowns, a tweet by the head of OPEC about oil production cuts, or a call on social media for a protest near (or about) the business. Advanced artificial intelligence (AI) techniques such as machine learning can help identify such data by, for example, identifying which data sources were used to generate the models most useful to your team.
  • Data that has been acquired and analyzed with enough lead time to allow the business to respond in a meaningful way. If a manufacturer of hand sanitizer, for example, needs two months to ramp up production, it needs at least two months and one day’s warning of booming demand. And if takes that business an average of two weeks to hire and train delivery people in a given region, it needs at least that much warning that infections in that area will sideline some drivers so the business can assure product delivery to all of its markets.
  • Models that are continually refined, using advanced AI, against “virtual twins” of the real world, rather than trained once on static data. Many predictive models based on pre-COVID consumer or employee behavior are obviously no longer relevant. But models based on today’s behavior patterns will also change as lockdown levels, infection rates, and other factors (such as protests and natural disasters) change the status quo. Cognizant’s Learning Evolutionary AI Framework (LEAF) can cope with such uncertainty by developing and testing the accuracy of millions of predictions against virtual models of real-world data.

'Accurate enough' means:

  • Data that has been cleansed and validated to ensure it comes from an accurate source, has not been compromised and is in a usable format. This is especially important for data in untraditional forms, such as unstructured data, or from newer sources such as social media or the Internet of Things (IoT), whose ephemeral nature on the network’s edge makes it exceedingly perishable. Such data can often be the source of dramatic insights, such as when cell phone location tracking data is used in addition to COVID-19 testing data to better track the spread of the disease and new infections.
  • Models that have not only been tested for accuracy under current conditions but that can, using advanced AI, provide more accurate predictions or prescriptions. If a prediction is low-confidence but could have massive consequences, such as an outflow of millennials from urban areas due to COVID-19, a business could get a low-cost jump on competitors by being the first to plan for such a trend. Cognizant’s LEAF assigns a fitness score to each model and then, borrowing from biology, creates new generations of the model, using the most accurate models in each generation to create new, more accurate offspring.
  • Models that have been trained to disregard patterns in the data when those patterns cease to be relevant, either because conditions have changed or because competitors have taken them into account. An analytic model that “discovers” weekly or seasonal patterns in financial trading is of little use if competitors have already found that pattern and adjusted their own trades to account for it.

'Relevant enough' means:

  • Data that is significant, or causative, enough to have a meaningful impact on predictions of future conditions and/or the steps it recommends to respond to them. As with other analytical endeavors, the definition of “causative” must be continually reassessed as conditions change. Before the “Me Too” movement, for example, an insensitive tweet by a CEO might not have been a data point worth tracking. Today, the boycotts such a tweet could spur could drive material changes in revenue, market share and brand value. In a recent Edelman Trust Barometer pollpoll from Edelman Trust Barometer, nearly 60% of respondents said how a brand responds to Black Lives Matter protests will influence whether they buy that company’s products.
  • Data drawn from anywhere, within or outside of the organization, that can help the enterprise be the first to sense and respond to change. This requires data scientists and their business partners to look beyond their familiar ERP and core business systems for any data that could signal a change that could affect their business.
  • Developing and refining models that can determine, with the help of machine learning, which data is most useful and disregard less useful data. Focusing on only the most useful data saves on data storage. Even more importantly, it reduces the waste of money, time and effort training the models on irrelevant data.

In our next article, we describe what’s required to continually assure the currency, accuracy and relevancy of both your data and analytic models to achieve the agile analytics needed to model an unpredictable world.

This article was written by Babak Hodjat, VP of Evolutionary AI and Jason Kodish, Global Guild Lead, Data & AI.

For more insights, visit the AI section of our website or contact us.