Turn self-service BI into a success story
Business Intelligence is essential to the effective and efficient manage of any organization. Managers can’t get insights into operational performance fast enough. Fortunately for them, BI software is becoming easier and easier to use. Managers don’t have to rely on stressed IT staff to provide the right data anymore. Long live self-service BI! Right? Well, yes and no. There’s a danger to using data in this way. If you don’t take the limitations of self-service BI into consideration, its success rate is very slim. Our 6-step improvement plan reinforces the foundation of self-service BI.
In multiple reports, research and analyst agency Gartner presented its expectations for the future of self-service BI. As always, there are two sides to every coin.
Two sides of a coin
On the one hand, Gartner expects more data to be analyzed using self-service BI than Data Science (meaning) in 2019. On the other hand, the researchers are pessimistic about the chances of success. Earlier, they predicted that probably only 1 in 10 projects is set up properly. Most projects run into data inconsistencies, which have to be avoided at all costs, as they negatively affect management decisions. Managing the data chaos that results from self-service Analytics can be a full-time job, Gartner warns us. Keeping the data secure and preventing data leaks also poses a great challenge.
Four practical problems with self-service BI
Besides data chaos, there are several persistent practical problems with self-service Business Intelligence. Below you’ll find four of them based on my own experiences and what I’ve encountered in the literature.
Managing the data chaos that results from self-service Analytics can be a full-time job.
1. Inconsistent definitions
More people get access to the source data themselves, allowing them to make their own calculations. That leads to everyone having their own definition. Especially when certain definitions work to the advantage of the person in question. These definitions are often also not shown in the reports. That’s a shame, because you’re striving to maintain one version of the truth across the organization.
2. Wrong interpretations
Multiple definitions also lead to wrong interpretations, especially as data gets more complex and speed becomes an important factor. In the case of self-service BI, it’s likely that the reporting person doesn’t consider every factor in their calculations. For example, when British soldiers started wearing helmets in WWI, the number of reported head wounds increased. A wrong interpretation of this statistic could lead to the conclusion that the helmets were a mistake.
In our ongoing project in the municipality of Rotterdam, we experienced complex data leading to wrong interpretations more than once. When citizens who apply for benefits are supported by offering education and other options from the first intake onward, you expect commitment from those citizens.
Yet, the data seemed to show the opposite, at first glance. In the pilot program, it turned out that three times as many citizens “dropped out” and stopped showing up to follow-up appointments compared to the traditional approach. This high percentage of dropouts was seen as a negative at first, until further analysis proved that the new approach led to fewer citizens feeling the need to apply for benefits, because they got the help they needed. There was a lot of daylight between definition of the calculated numbers and the way they were interpreted.
3. Trivial KPIs spreading like wildfire
Managers love measurements. If the data is available, they can calculate to to their heart’s content. This leads to them developing tons of KPIs. Everything that can be measures is used to manage.
Focus on the vital few KPIs, not the trivial many.
Whoever took the Performance Management and KPIs training course knows that more isn’t always better, especially when it comes to KPIs. Not everything that can be measured is relevant from the perspective of management information. It’s all about the vital few KPIs, rather than the trivial many.
4. Inconsistent data
As we said earlier, speed plays an important part in self-service BI. The manager doesn’t want to wait for IT to deliver the numbers, which can take weeks. Speed wins over purity. If the data is available, they can immediately start calculating.
But herein lies another treacherous pitfall: the assumption that the data is correct and that the manager understands the data.
For a long time, the trend was to let everyone register as much data as possible, without checking to see if the data is actually necessary. Until recently, every employee in healthcare had to log a diary every 5 minutes, detailing everything that happened, as described in the Dutch Volkskrant. The healthcare secretary fortunately put an end to that. Seeing as the person registering the data often doesn’t benefit (directly) from the data at all, the odds are good that there are inconsistencies in the data, or at least that the data was registered incorrectly.
Apply our improvement method for BI self-service
How can you avoid these practical problems and pitfalls? A solid methodology when setting up self-service BI is a must. To make a project succeed, it’s essential to make commitments with each other. This can be done using the methodology described below.
6 steps on the road to manageable self-service BI
The methodology covers the entire process of calculating, validating, and displaying the data. Using this methodology, you can quickly implement self-service BI, without sacrificing data purity. The various roles and responsibilities are clearly registered in this process. This is an important condition to avoiding most of the problems above.
- Existing insights. The cycle always starts with the existing, validated insights. These insights keep leading to new questions. In other words: this is an organic development process, where a new step is only undertaken when the previous version is working correctly.
- Suggesting a new data point. These new questions have to be made concrete (for example with a practical example). That makes it clear to everyone what the question means. It’s important that everyone involved agrees about the question to be solved, so that everyone can interpret the data correctly. That makes the next step easier.
- Formulating a definition. The question has to be turned into a definition. You might think these are one and the same, but they’re not. The definition has to consider the question of which data is available, and how it’s registered. The definition contains, for example, the exact names of the tables and columns the data is pulled from, and how they have to be processed together. This is important, because large organizations can make the same kind of data available in multiple places in the (enterprise) data warehouse, but with small differences in key places. It’s also important that the definition is relayed back to the stakeholders of step 2, and agreements have to be made again, to once again prevent wrong interpretations.
- Calculating. Once the definition is established, it can be calculated using the proper BI tools (see our BI Tools Survey to find out which tool is right for you).
- Validating the data point. As stated earlier, it can’t just be assumed that a data warehouse contains “pure” data. You also can’t assume that the data is exactly where you might expect it. The data has to be validated before releasing the calculated data. This can be done by those who registered the data, for example. This step is crucial, because the data is assumed to be true after this step. It can still happen that people don’t accept the data as true, especially when the numbers are disappointing. Positive numbers, on the other hand, are accepted as the truth, even if they’re not. But if this step is executed correctly, you can avoid debates about the origins and credibility of the numbers. That saves a lot of time!
- Adding to reports. Finally, the calculation can be re-added to the reports, and the cycle starts anew.
TIP: the definition and validation phases are crucial – never skip them.
The improvement method described above can be applied to different self-service levels. It’s possible to play every role for all six steps yourself. Of course, the method can also be used in a more traditional organization. A BI team could, for example, take care of the definition, calculation, and validation. The business intelligence manager then processes the calculated data in the reports. It also ensures that managers who prefer speed can no longer skip the crucial steps, such as definition and validation. To ensure that the steps are followed, someone should be appointed as the person responsible for every role. When implemented well, this method ensures a thorough foundation for a manageable self-service BI process.