Business intelligence (BI) can be hard to do well. Why? Because our systems are in a constant state of change, our choices are rarely as simple as right and wrong, and perfection comes at a cost.
When I set out to develop a relatively new BI practice at Twitch just over two years ago, I sought advice from other experienced BI leaders. Here are the top things I learned.
Managing BI is like painting a moving bus
Do not get overwhelmed. Accept that you can’t solve every problem. Instead, be opportunistic and place bets. You will always be pulling pieces out, putting pieces back, and the results may still be a little sloppy at times. That shouldn’t stop you. Instead, keep moving and apply touch-ups later.
When we first established business performance reporting at Twitch, we didn’t yet have exact definitions of all metrics in our report set, but we also couldn’t afford to wait until everything was perfectly baked. So we launched the report set along with a governance process to manage change. Teaching our stakeholders how to swap metrics in and out became a valuable part of the program and allowed us to keep our momentum despite the moving parts.
Your BI strategy should reflect company culture
In BI, there is no right and wrong way to do things, nor a magic template with check boxes of everything you need. Instead, you have a series of decisions to make and ultimately your choices should be tied to your company culture.
Take data interface ease of use, for example. If you have a technical employee base, you may choose to provide only lightweight data interfaces. Alternatively, if employees place high value on exploring data quickly without writing code, you may invest more in full-featured, self-service analytics. Realistically, you will likely need both types of interfaces, but the right investment mix is a subjective decision reflecting your corporate culture.
In BI, there is no right and wrong way to do things, nor a magic template with check boxes of everything you need
High-quality data comes at a cost
Data quality is a justifiably hot topic in BI. Starting out, you might think the goal should be to make all data in your company as close to perfect as possible, but that approach can be expensive. Ensuring exceptionally high-quality data takes engineering and analyst effort, specialized tools, and dedicated program ownership. If you invest enough time and money in the equation, you can boost your data quality considerably. However, you must first decide whether it’s worth the cost.
Whenever the inevitable quirks of data surface, I ask, “How good is this data, and how good does it need to be?” The answer isn’t always, “Perfect.” Rather, acceptable data quality varies depending on the risk profile. For us, if data is tightly linked to customer experience, revenue generation, or topline performance metrics, only then will we invest in controls to ensure the highest possible quality.
The advice I was given amounts to a few simple rules: falling short of perfection is sometimes necessary, strategy should match your culture, and ROI should always be evaluated. Leading a BI practice may never be push-button easy, but getting the fundamentals right shouldn’t take more than basic common sense.