Crossing the chasm between data science and impact | by krishna shrinivas | Jan, 2021


Deep analysis is only half the job. Win trust with decision makers and inspire actions to deliver impact.

krishna shrinivas

Most analysts have been part of failed projects, or have encountered the dreaded “I am not sure I understand it…” followed by an awkward, unspoken “and, I can’t do anything with it!”.

Research reports say that many (nearly 87%) analytics projects fail. Several reasons have been cited such as organizational structure, culture, talent, data availability, etc.

However, translating insights to trustworthy recommendations, and inspiring actions is a big stumbling block for many analysts. Often stakeholders aren’t able to quickly grasp the analysis, develop misgivings and abandon ship without taking any actions.

It appears that the last mile poses a significant gap!

Photo by Lubo Minar on Unsplash

“Analytical translators” have been proposed as someone who can help structure problems and translate insights. However, that would move the analyst one step further away from the context, causing other complications.

I posit that with a bit of attention, planning, and focused effort, data scientists can begin bridging this gap.

Insufficient time for insight generation/ storytelling –
Insight generation is an overlooked step in the analytical process — We allocate a disproportionately small amount of time to building insights & storytelling. We feel an onus to spend most of the time digging through the data, and expect that a few charts pasted onto slides will communicate the story.

This last minute effort is a key failure point, where data scientists struggle to communicate complex analysis, insufficiently bridge the analyses to the objectives or recommendations, failing to win trust with decision makers.

Complexity tradeoff
As data scientists, we are tasked with finding insights or building an accurate prediction model. To that end, we might dig deep and chase down a complex train of thought, or apply an ensemble of algorithms to achieve the desired outcome. However, the more complex the analyses, the harder they are to explain, especially to an audience less experienced with data science and algorithms.

A lack of shared context
Having shared context with business folks, knowing the motivations is critical to creating hypotheses, identifying model features and ultimately connecting the dots to meaningful recommendations. Often, analysts & stakeholders spend little time sharing context and walk away with different visions for the project. An analyst might not know what to ask for, and the stakeholder may not realize there is insufficient prior knowledge. Ultimately, the effort ends up surfacing something rudimentary or missing the mark completely.

Allocate sufficient time
I propose that analysts budget at least 50% of time in the project plan for insight generation, and structuring a narrative. The shift to telling the story, rather than continuing to dig the data will also serve as a check against analysis paralysis, bring focus back to the exam question (reduce scope creep), and at the same time enable the analyst to address any potential holes or follow up questions that might come up.

Iterate and polish
The process of creating meaningful recommendations and building a crisp narrative is iterative. It can at times be frustrating to keep making changes to slides, but with the right feedback and direction, the effort pays off. With each iteration, you are looking to address any insight gaps, and improve the effectiveness in conveying the insight and recommendations.

It helps to identify the key decision makers and their style and appetite for adopting data based recommendations. You may want to anticipate potential follow up questions they might ask and preemptively address them. Improvements you might make include verbiage, content flow, imagery and even things like font, colors etc. You want to eliminate any distractions to the key message such as ambiguous statements, or erroneous facts that can derail the presentation. Higher the stakes, the more refinement you’d want to the overall presentation. It isn’t uncommon to go through dozens of iterations to get a polished final product, ready for a CXO presentation.

Find a sparring partner
It is extremely useful to find a trustworthy sparring partner — someone that has tenure in the organization, or has expertise in the business area you are analyzing. Most importantly, they can lend a critical eye to your work, provide direct feedback on whether your insights are new and impactful, whether your recommendations are meaningful and feasible, and whether or not the presentation communicates effectively. You also want their counsel on how best to win the trust of key decision makers and potential areas that can derail the effort.

Invest in building context
Context is acquired over time and every project serves to bring more. It can be intimidating for an analyst to seek to understand the inner workings of a product or a business. However, without it, your ability to answer the question gets compromised. It is ok to be vulnerable and ask for context around the underlying business, the motivations for the project and any potential hypotheses. Request any prior studies that might exist, build time for exploratory analyses to understand key business drivers. Most importantly, foster relationships to continue building context over time.

Continuous improvement
As with anything, effective communication is an area for continuous focus. After each presentation, reflecting on the outcomes and identifying areas for improvement will help make future reports more impactful. However, the key step is to acknowledge that it takes focus and effort to bridge the last mile gap. Once you have identified the gap, it is a matter of time before we cross the chasm safely.


  1. M.Henrion, Why most big data analytics projects fail (2019), ORMS today
  2. C.McShea, D.Oakley, C.Mazzei, The Reason So Many Analytics Efforts Fall Short (2016), Harvard Business Review
  3. T.Davenport, Keeping up with your quants (2013), Harvard Business Review
  4. N.Henke, J.Levine, P.McInerney, Analytics Translator: The new must-have role (2018), McKinsey

Read More …


Write a comment