How to define self-service in analytics?
Enabling all users to use analytics tools, whatever their level: that's the promise of self-service. But is it achievable, or even advisable? We've taken a closer look at the question in an ebook you can download here.
Today, the word self-service is thrown around like a marketing slogan by just about every BI tool vendor. But it is rarely defined in concrete terms. It's hard to identify the essential functionalities required to consider a tool a self-service tool. Generally speaking, the promise is to enable business users to better exploit their data without the intervention of data or IT experts, via a user-friendly interface.
But here's the thing: “business users”, “data exploitation” or even “intervention” never mean the same thing. Are we talking about developers who may not be data experts, but who know how to formulate SQL queries to extract the information they need? Or busy floor managers who need to see their KPIs clearly and quickly on their tablets while managing their stores? And what do we mean when we talk about minimizing the intervention of experts? Will they always be available for support when the self-service tool reaches its limits? Do they clean, format and set up data sets before integrating them into the self-service platform? Is their help needed to determine new calculation rules, or can end-users modify the dataset themselves?
Here at Toucan, we distinguish between 4 levels of analytics self-service. Find out more in this article, or dive into our ebook on self-service.
Level 1: Dashboard customization
Where does a user's analytics experience begin? With the dashboard in front of their eyes. The presentation and arrangement of this dashboard usually depends on the choices made by the product builder. They can decide to place certain visualizations before others, or to highlight certain KPIs. Let's take the example of personnel management software used by a hotel chain. The product builder may have chosen to place at the top of the dashboard a leaderboard of branches according to their rate of presenteeism, followed by a graph summarizing the number of overtime hours worked by each employee. It may also be possible to apply a geographical filter, based on the branch's region, or to filter according to job type (restaurant or reception staff, etc.).
The problem? The product builder who delivered the dashboard doesn't know the day-to-day experience or habits of the branch manager who's going to use it. So it's a good idea to let the branch manager determine the layout of the dashboard themselve (which visualizations appear before others) and set their default filters. Just as when you configure your own home page on your browser, this can save time and make the experience smoother. This is the first level of self-service in analytics.
Level 2: Removing or adding existing visualizations
In business analytics, the how counts as much as the what. In other words, the way in which information is presented is vitally important. If a floor manager wants to know which products are the best-sellers, a pie chart may not be particularly useful, and a histogram may show the differences between products more clearly. The product builder can't know this, because they don't know the end-user's preferences. The product builder will therefore have planned several visualizations for the same indicator. The user can choose the most relevant visualization and integrate it into his dashboard. They can also freely delete those that don't suit them. This is the second level of self-service.
Check out our ebook: self-service in analytics here.
Level 3: Creating new visualizations
What happens when the product builder hasn't provided the right visualization for a particular indicator? Let's take the case of a call software application for a customer service department. The team manager needs to know the average waiting time before an operator takes charge of the call. The calculation rules for this indicator are clear, and the data set has already been processed by the data analyst. The only problem is that, to see how this indicator evolves over the weeks, the manager needs to see it represented on a line graph. When the third level of self-service is reached, the user will be able to design this visualization themself.
If this requires a strong simplification of the UX so that non-technical users are at ease with configuring visuals - on the model of reference tools such as Excel in large groups or Notion in SaaS - it's clear that generative AI appears to be a real unblocker: just as models can create an image from a prompt, they can create visualizations.
Level 4: Creation of new indicators by directly manipulating data sets
So far, when we've talked about self-service, we've mainly been talking about how to present the results of calculations that have already been made, rather than the means to perform the calculation itself. For example: suppose the manager of a customer service team needs to know the rate of calls not picked up because they make for particularly unhappy customers. They'd like to build a new indicator to monitor the number of daily calls that would not be answered by an operator. This indicator was never foreseen by the experts who built the platform, so it will have to be invented.
This would require direct action on the dataset. This requires a level of expertise that only data experts possess. They are the guarantors of data quality and security: they clean it up, remove duplicates and format it. They determine the calculation rules for each indicator, based on the headings they themselves have defined.
It's an illusion to think that, one day, anyone will be able to do this work thanks to self-service, even without any particular expertise. There will always be a need for an expert to guarantee data quality and calculation rules. That's why, at Toucan, we don't consider level 4 self-service to be an attainable goal.
