Many who make the Big Data journey are overly fixated on making it to the “Promised Land.”
In far too many cases I see people who plan to build out a complete system and architecture before using a single insight or building even one predictive model to accelerate revenue growth.
Everyone anticipates the day when Big Data can become a factory spitting out models that finally divulge all manner of secrets, insights, and profits.
It doesn’t have to be this way. Creating what I call an “Insights Roadmap” can provide useful insights long before the Big Data build-out is complete. By taking this approach, you can start cranking out hundreds of models that can generate revenue quickly.
Getting from insight to impact faster
As I mentioned in an earlier piece “The one tool you need to make Big Data work: The pencil,” any Big Data journey needs to begin with a clearly defined end game. This approach, “Destination Thinking,” helps to make sure you know exactly what you are looking for before you start your journey.
However, once you have started your journey, it is all too easy to fall into the trap of believing that you can’t harvest any Big Data insights until the long building journey is complete. This doesn’t have to be like a car, which you can’t drive until it’s all built. Think of it more like a tool factory making all sorts of specific tools for specific jobs. Every time one tool is complete you can begin using it; you don’t have to wait for all tools to be complete.
Here’s what I tell my clients to do:
1. Building and rolling out an Insights Roadmap won’t happen by making it someone’s part-time job, or assigning one person without the right skills to do it. It requires a team of people from relevant functions with the right set of set of skills. In most cases, that includes data scientists, business solution architects, infrastructure architects, and software experts combined with business leaders in charge of the marketing campaigns and accountable for results.
2. Focus on P&L impact. Create a detailed business case about the desired impact you’re looking to have so that no one is confused. The business case needs to identify revenue targets and customer behaviors to change that would have the largest impact on the P&L. That level of insight requires an analysis of what the sources of profit and loss are before ranking them by their impact on the business. Most companies will have a sense of this, but a detailed list of sources of P&L impact is often incomplete or non-existent even though it should take just a few days to pull together.
For example, once you’ve identified that stopping churn has a big impact on your business (as measured by the lifetime value of the customer), you need to identify the behaviors that cause churn, e.g. long waits when calling customer service. Then the team needs to create a target , e.g. reduce churn by 4% in 6 weeks, and develop a model that can, for example, predict spikes in customer service calls so that the company can then staff additional agents at those times to reduce wait times and/or improve training to manage question flow more efficiently.
3. Prioritize by ease of data access. Answer this key question: Which models can you use first with the least amount of investment and the greatest ease to build and deploy? You need to essentially start cherry picking data sets. Call this the power of “Small Data”. Find the data you already have at hand, and the fewest number of data sets needed to make the calculations for a viable model. In our example, this would mean accessing customer service center records regarding who called and waited how long on hold and customer billing records which can tell you who is and is no longer a customer. Then prioritize those models based on P&L impact list you’ve already developed. Be sure to test the model once you’ve developed it for quality and integrity.
4. Build a roadmap. I know this step sounds obvious but I’m constantly surprised by how little effective planning takes place. As counterintuitive as it may sound, it takes planning to work fast. Make sure you’re focusing on the models and necessary infrastructure that deliver the most value, and sequencing them all to minimize delays. Put in place clear data delivery deadlines and test timelines for each model to ensure timely delivery.
Let me be clear on one point: this fast tracking with an “Insights Roadmap” isn’t about creating extra work; the models you do first are ones you’d develop anyway.
Real value, real fast
One high tech company followed this “Insights Roadmap” route. By following the four steps I laid out above, the company started generating insights within two months of the project starting, long before the two-year Big Data warehouse build project was completed. Within three months, the company was able to book $5m in additional revenue. By end of year one, incremental revenue exceeded $75m.
Your journey to reach Big Data’s promised land will take time. But realizing value doesn’t have to.
Matt Ariker is the COO of McKinsey’s Consumer Marketing Analytics Center. You can learn more about marketing and analytics at the Chief Marketing & Sales Officer Forum site.
Please follow War Room on Twitter and Facebook.
Join the conversation about this story »