The move towards cloud computing and analytics is creating a world in which every possible cut and aggregation of a dataset can be done automatically. The problem is that while analytic engines can scale infinitely, human attention does not.
With hundreds of possible models to apply, and infinite possible insights from each, the process to go from statistical results to clear action is painful, murky, and lacking a common language for relaying insights. In the end, models with even the highest accuracy will run into a wall of objection — “What am I supposed to do with this?”
Today, analysts and business leaders frequently make decisions by squinting at myriad charts and visualizations, eyeballing trends in the search of useful insights. Unfortunately, many of these supposed insights are hard to understand by the business, and most are never acted upon due to non-interpretability, irrelevance to the stakeholder, or simple mistrust.
Top data teams can address this gap by marrying these observations with a clear understanding of how a statistical result can map to a specific action. This means establishing common ground with their stakeholders and using that to drive the conversation.
When data is tied to action, analysts can lead with recommendations for action and then use visualizations as a complementary mechanism to validate the findings. For more teams to unlock this style of action-oriented analytics, three key shifts need to occur:
Moving from numbers to recommendations
In many organizations, there’s typically a disconnect between analysts and business operators. Analysts present slide after slide of time trends and bar charts, often leaving operators without a clear understanding of how to respond, or even what responses should be considered.
Let’s use a restaurant franchise as an example, who had developed a weekly email report on key individual store metrics, sent to managers. One report in particular described “a drop in the week-over-week CSAT on the warmth of breakfast pastries in a particular restaurant that exceeded two standard deviations.”
Perhaps this was easy for analysts to understand, but it was gibberish for the restaurant managers. No one translated this fact into a clear recommendation. Exasperated at looking at a wall of numbers and being expected to act from it, a restaurant manager exclaimed: “If you want me to leave the muffins in the oven longer, just tell me!” The analytics team had skipped translating statistical jargon into a clear and concise recommendation in a format that is accessible to the restaurant manager, leaving the manager to draw her own conclusions or ignore the finding altogether.
This frustration made it clear that while an analyst may uncover a critical statistical result, it still has a long way to go before creating value for the business. The key is to map these statistical results to clear recommendations by deeply understanding the levers that operators have to move the business. Then, it becomes easy to translate how data like week-over-week CSAT variations on the warmth of baked goods maps on to an action like leaving baked goods in the oven for longer.
While many statistical results might not have a corresponding action and vice versa, there is tremendous opportunity where there is a regular and repeatable intersection of action and data. Even in prototype forms, these kinds of “intelligent actionboards” that map certain statistical results into clear actions have resulted in wide business gains everywhere they’re implemented. In the study above, McKinsey measures how one Latin American telecommunications company implementing this action mapping resulted in productivity gains of 18 percent.
Shift from actionable data to insights that drive action
With the increasing proliferation of Customer Data Platforms (CDPs) like Segment and mParticle, as well as tools like Census that aid with piping data from warehouses back into operational systems, we see a small but rapidly expanding set of use cases in which actionable insights will be replaced with “insights that drive action.”
The simplest kind of “insights that drive action” would be a precise alert or email when an automated analysis indicates that something is wrong. Devops tools, for example, can alert teams of developers and product managers when page load times surge above a threshold. Pager duty alerts are great examples of translating some occurrence in the data table into a pithy and effective action.
There is opportunity for simple, precise alerting to spread to more business use cases, but also for tools to help with the steps after the notification. For example, with respect to business use cases, rather than just notifying email marketers of a highly engaged segment, a CDP could share recommended messaging and target segments for email marketers to use in their campaigns.
We still believe an employee, as opposed to a machine, will be the one conducting the final action. This is because in the end, it will be the employee that is accountable to the marketing email getting sent out, and its either positive or negative consequences. In most scenarios, the human will want to review and have the final say.
But recommending different actions that the employee can take will enable the acceleration of data-driven action. For example, in the devops use case above, the developer can be met with a set of buttons within an analytics platform that correspond to a different action that can be taken to fix the problem.
Changing analysis from pull to push
Dashboards and executives go hand-in-hand. These business stakeholders require analytics teams to build them a bird’s eye view on how key metrics in their business are changing to make informed decisions and guide execution. While dashboards make the insights available to stakeholders available at a glance, they are only glanced at via a forced review at a weekly business review.
Nearly every week in organizations big and small, we see an hour to two-hour long roundtable meeting with 10+ people (the weekly business review, or WBR), in which business stakeholders and analysts stare at one or multiple dashboards reviewing their yearly metrics and ask unanswerable questions as to “why sales dipped” or “why conversion changed”. With both C-Suite and VP participants and several hours of prep, these meetings are the most expensive in the business. And through our interviews, the verdict is near unanimous – these reviews are painfully inefficient, since the WBR conversation revolves around high-level changes in trends, not what anyone can actually do to respond. One mobility platform operator emphasized, “It’s the most painful thing we do, and everyone leaves a bit frustrated.”
In innovative companies, we see analytics moving from a weekly WBR “pull” model to much more of a push model, in which data drives front-line action on a near continuous basis. Insights and action recommendations are pushed to small, nimble teams of analysts and operators. The analyst digs deeper and validates these insights, while the operator takes action based on his scope of control.
With more business users supported by AI and unprecedented insight generations, a “weekly business review” in the future will feature operators and analysts presenting on what they did, or even which action recommendation they selected – not wasting time head-scratching as to why a KPI moved up or down.
Bringing in the business intelligence of tomorrow
We’re looking forward to a business intelligence that’s simpler to understand, with a focus on actionability and timeliness.
As a first step, we want to usher in a clear mapping of statistical results and calculations to the specific action that a business might take. If the data implies the need for baked goods to be left in the oven for longer, to refer to our earlier example of the restaurant manager, then that should be the deliverable, not a set of charts and graphs.
With this mapping in place, we’ll see an increasing efficiency of how a statistical result in the data gets turned into action. Perhaps this starts with just a precise alert to a line operator. But there’s also immense opportunity for tools to help with the steps after the notification
And finally, the center of gravity of analytics will move towards an asynchronous, always-on exchange between data warehouses, analytics systems, and front-line business operators. Executive business reviews will still exist, but instead of time digging into data to figure out what to do, insights can be delivered automatically and most meeting time will be spent weighing the cost/benefit of each recommendation.
Charles Zhu, Director of Solutions and Data Products, Sisu