Throughout the business, marketing and insights industries there’s increasing excitement and buzz around the potential for data to transform our world. But what do terms such as big data, machine learning and artificial intelligence really mean in the context of packaging design? And how can data help marketers to develop more effective packaging and expedite and improve the processes?

A data mining effort can yield specific insights to these questions and guide future design efforts while creating a path forward to accurately predicting packaging success.

Conducting Two Levels of Meta-Analysis

LEVEL 1: The linkage between specific performance measures (KPIs) and sales (Purchase from Shelf) The goal was to dive deeper to understand the relationships between different dimensions and, most importantly, to understand which metrics most strongly and consistently drive pick-up and purchase. As expected, this analysis reinforced a key learning from in-market validation work: On-shelf behavioral performance measures visibility and shopability are much more closely linked to sales than attitudinal measures such as aesthetic appeal, imagery or communication.

Thus, one important point is that any form of data mining and more advanced machine learning is most valuable if it is rooted in behavioral data, tied to shoppers’ actions at shelf. Across many studies, the analysis revealed that visibility—gaining attention within a cluttered shelf—and shopability—a consumer's ability to quickly and accurately find a desired product—work in very different ways. Visibility, as a positive driver of sales, works in a largely linear manner. Simply put, if more people see a product within a cluttered shelf, this increases its likelihood of purchase and, conversely, “Unseen is Unsold.” And while the linkage between visibility and sales is strongest for smaller brands with few shelf facings, increasing visual attention always has some positive impact.

Shopability, as a negative driver of sales, works in a non-linear manner. As many brands have learned from painful experience, confusing people at the shelf is the surest and strongest path to sales declines. However, the analysis showed that shopability works quite differently than visibility: Making a brand a bit faster and easier to shop may not help sales, while a slight decline may not hurt. Instead, there appears to be a confusion threshold or inflection point where shoppers become frustrated and walk away.

LEVEL 2: The connection between specific design elements and different performance measures visibility, shopability, appeal, communication, etc. This effort was rooted in painstakingly coding changes to design elements, i.e. logo, color, visuals, structure, claims, etc., when comparing proposed design systems to current packaging. For new products or line extensions, test packaging systems were coded relative to the existing product line “How does the line extension differ from the packaging of the base brand?”

This new lens of analysis revealed some interesting patterns, including: • Changes to brand identity such as logos and variant descriptors were more strongly correlated with shopability problems and sales declines. Changing these elements appeared likely to confuse shoppers and/or create hesitation at the shelf.

  • Changes to pack color, visuals and shape were more strongly connected with improvements in visibility and purchase from shelf. This suggests that design changes need to be significant enough to be noticeable from several feet away if they are to make a difference in shopping patterns.
  • New design systems that included new claims along with both graphic and structural changes were more likely to drive sales gains. This speaks to the importance of linking design (a new look) and messaging (a new benefit) when approaching packaging re-stages.

To be clear, the analysis certainly uncovered cases in which logo changes have improved packaging performance, as well as other cases in which misguided color changes led to sales declines. The idea is not to suggest that effective design can be reduced to a simple formula. In fact, the same solution is unlikely to work in two very different situations. However, data mining shows that by learning from experience and identifying patterns we can better inform design briefs and help marketers avoid making the same mistakes time and again.

What’s Next? PREDICTING SUCCESS by Moving From Data Mining to Machine Learning

While this recent meta-analysis was valuable in identifying guidelines and guardrails for design, it was only a first step—and an initial indication of the power of data. In fact, a larger goal is to apply data-driven learning to accurately predict which design systems will be successful in market. This involves applying the learning to develop a database-driven process to screen new design systems and identify those most likely for success.

The underlying idea is to apply a consistent framework and process to review designs across key KPIs with ratings and weightings informed by database learning. This screening system, in turn, will further feed the network and provide more data and learning upon which to develop, enhance and refine a predictive model. This is where data mining transitions into the field of machine learning, in which a system builds knowledge and accuracy through increased experience.

Of course, machine learning and predictive algorithms of this nature are not a full substitute for ultimately testing designs on shelf with shoppers. However, they can potentially be a powerful complement to more traditional research. On one level, data-driven tools will provide nearly instant guidance and feedback and help remove much of the subjectivity impacting design decisions today. If applied properly, these tools may also foster creativity by allowing for the quick, inexpensive screening of many design directions. For these reasons, both marketers and designers should embrace the data revolution and potential to learn from experience. Those companies and agencies that do so are likely to be well-rewarded.