For Analytics to Be the Answer, You Need the Right Use Case
In this article
More than ever, companies are looking to incorporate big data analytics to help drive value throughout their corporation. They are creating more data every day and looking to use this data to enhance their efficiency in a variety of areas. From marketing to supply chain and security, using big data analytics to make sense of data, drive insights and automate decision making is proving to be a valuable endeavor for enterprises.
Getting started on the right foot with your analytics process is more important than ever. Internal organizations tasked with making this all work are in the spotlight and under pressure. They must sift through mountains of messy data, match that data to the right business use case, apply the right algorithms and ultimately decide the best technology platform to tie it all together.
As mentioned in a previous blog post, WWT recommends always starting with the business outcome you are trying to achieve. Once identified, the business outcome will help the organization stay focused on finding the right math for the business challenge. From there, they can clean and link only the most pertinent data. The result is in initial analytics product that is delivering actual business value. No one can argue with a team bringing business value, and you'll have the power to continue on to your next use case.
This approach has worked extremely well for WWT, so let's dig in a little more on how you choose the right use case.
The first step in choosing the right use case is brainstorming all possible use cases and business outcomes. The goal is to develop a list of about 20-40 use cases with clearly defined business outcomes (e.g. increase revenue through more targeted marketing, reduce call center costs by decreasing call volume, decrease fuel/wear and tear costs on our fleet by optimizing routes and predicting maintenance issues). Once the list is created, then the fun starts.
To prioritize the use cases, the first question that needs to be asked is whether or not the use case lends itself to a big data analytics solution. Not all use cases do, and when a company goes after these use cases, it is like fitting a round peg in a square hole. A good example of this would be an advertising organization looking for a more creative way to humorously market their product. The only way to get more creative and humorous is to have more creative human brains making far-fetched connections that only humans can understand the humor in. This is something computers just cannot do no matter how much algorithmic training is involved.
Once all of the use cases that lend themselves to analytics solutions are identified, the expected impact for each use case should be estimated. Impact can be defined by a variety of factors such as increase in revenue, decrease in cost, decrease in risk or improved efficiencies. This exercise is more of an art than a science. But getting expert opinions, clearly stating your assumptions and giving plus/minus ranges are all good practice. The idea is to get an order-of-magnitude feel for which use cases could make more of an impact than others.
Finally, each use case should be rated on their ease of implementation. This will be a subjective categorization, but there are five major factors that should be taken into account.
- Data source complexity – What is the number of data sources that need to be combined, the cleanliness of the data, the granularity of the data, etc?
- Analytical complexity – How challenging is the solution from a modeling perspective?
- Political complexity – How much internal support will this solution have?
- Regulatory complexity – Are there special considerations such as HIPPAA compliance or SEC regulations that need to be taken into account?
- Implementation complexity – How difficult will this be to integrate into everyday operations?
Overall, you want to evaluate how much time will it take to get all of the necessary data, evaluate it, build predictive models and put them into production. To properly evaluate the use cases it is recommended that they be put on a continuous scale rating them from 0 to 100, where 100 is the easiest to implement.
Once you have gone through the steps above the data should be aggregated as shown below. This chart will allow you to prioritize which use cases to go after first. Once plotted, you can see which will give you a quick, high-impact win using big data analytics.
It is recommended that only 1-3 use cases be tackled at a time, per business unit. Each use case should not take more than four months, from idea to implementation. With that, business value should be gleaned within the first two months of production. Rather than saying you implemented a new technology, ingested the data, explored the data and ran some jobs side-by-side with your current data warehouse, just imagine the power of a statement like:
"In six months we were able to show we can save $2M/year using this analytics process on a subset of our population. If we scale it out across our whole customer base, we expect to save about $20M/yr. Because we have already implemented this in one area of the country, scaling out should require minimal effort. We just need executive support to get the right people on board!
Taking the business-outcome approach will eliminate skepticism at the executive level. It also helps build support for you to go tackle more use cases while your team is implementing the right technology to scale the whole effort. With this approach, you can directly compare how different combinations of software and hardware perform something you know is mission critical. Ultimately, you'll find yourself in a place where you are more prepared to evaluate which technology is right for your business.