This is part two in a series of articles about effective analytics implementation. The first part “The Five Faces of Analytics” explores the roles necessary to develop a successful analytics team. So you’ve assembled a team of world beaters and they’re chomping at the bit. They’re ready to transform your organization into a data-driven decision-making juggernaut. But where should they start? How do you coordinate this team in such a way that they’ll actually be effective.
Remember that some estimates put the failure rate of analytics projects at 80%. That’s almost twice as bad as the average IT project. How do you ensure your initiatives are firmly planted in the 20%?
Analytics lives and dies on adoption by decision makers. These are managers who have spent their lives processing information and deciding based on gut feel. You are trying to get them to augment or in some cases abandon decades of received wisdom. No wonder your results are met with skepticism. Analytics is the outsider, the newcomer. The Disruptor.
And that’s the key. You need to recognize that what you are doing is a disruptive innovation. And like all successful innovations, you need to start small and work your way up. You’re not going to slay the dragon while you’re still just a squire. You need to win some minor battles, solve some smaller problems, and collect a bit of credibility.
Step 1: The Brain Storm We start at the end: the decision. Write down all the decisions that the organization makes, whether big or small. Add to these all of the decisions that they could be making (but aren’t) because of uncertainty or laziness. Document every rule of thumb.
Think of strategic, one-off decisions like “should we buy this other company”, or “should we build a second store” right through to the more tactical and ongoing “how many flyers we should send out this week” or “what routes should our drivers take today”.
At this point, you’ll probably realize that there are more problems to solve than you have time in your career. That’s a good thing. The next steps are how we’ll pare the list down.
We’ll do so by eliminating those with high risk in data inputs, research, and implementation.
Let’s cull the herd.
Step 2: Evaluate the Data
Using these “decision problems” as a guide, take an inventory of all the data resources that could be used to support or solve them. Don’t limit yourself to data that is behind the firewall. Look for open data, proxy data, and other sources available outside of the organization. The world is swimming in data, so be a little crazy here. You may not find exactly what you’re looking for, but there’s often a pretty good proxy that you can use. Creativity here will make you look like a hero later.
Next, match each data set (or sets) to each decision problem.
Now data can be a killer. More than one analyst has lost his job when he discovered that garbage in actually means garbage out. So it’s time to be ruthless: eliminate all the decision problems where the data is expensive, suspect, or unavailable.
Wow. That chopped the list down to a much more manageable size. But we’re not finished chopping.
Step 3: Scope the Projects In looking at what remains, you can start to estimate the difficulty or uncertainty associated with finding a solution. We’re turning our “decision problems” into potential projects. Talk to your analytic explorer. Ask her how many weeks or months each would take to “solve”. Have her think through her approach including data collection, cleaning, modeling, verification, visualization, tool or metric development, and implementation. This doesn’t have to be super-accurate, but you want to know if it’s days, weeks, or months.
Now double all her estimates and get rid of any that are longer than three months.
Step 4: Scope the Implementation Implementation and adoption are tied at the hip. So we’re going to do a little pre-screening based on likelihood of acceptance. Look at each project and try to envision how many people would be involved in actually using it and supporting it. Who will need to sign off? Are there multiple end users? Will it require some kind of software tool? Who will care for and feed the model new data? How often will it need to be updated with fresh data? How often will it need to be recalibrated?
Some projects will require one or two people, others will require people from half a dozen departments in the organization.
Eliminate anything that involves more than three people.
And BAM! You have half a dozen potential projects that have data available, can be solved in a few weeks, and won’t require a change management consultant to implement.
Step 5: Engage the Decision Makers At this point, you’ve been on the job for a few weeks, and your boss is probably wondering what you’ve been up to. It’s a perfect time to show her the list of projects. Pull together all the decision makers who are represented on your list and let them digest the implications of each one.
Describe each project in terms of how it will support decision-making and make them look like heroes. Have them envision the upside of each project and the value to the organization.
Remind them that your solutions won’t tell them what to do, but will simply reduce uncertainty. They’ll still use their gut, but they can now supplement it with their heads.
They’ll prioritize the ones that eliminate the most pain and help them sleep soundly at night. Voila! You have found the low-hanging fruit.