Click to learn more about author Aditi Raiter.
Self-service BI is lucrative but demands an eye on
governance and security. In this article, I will stress the behaviors and strategies
needed around managing self-serve BI/reporting tools.
Self-service analytics rotates around business users having access to data preparation and dashboard building tools. There are many challenges when self-service BI tools are open to all users in an enterprise. In the article The Impact of Data Governance on Self-Service Analytics, the author rightly claims common risks — multiple data models with unnecessary changes, multiple truths and analysis outcomes, security audit failures, BI systems maintenance nightmares, etc. In the article Safeguarding Against the Risks of Self-Service Data Preparation, the author emphasizes the impact on data quality and data security associated with such models. These solutions can become critically vulnerable when they are flexible, as they can even lead to enterprise data stored in a server under the desk of a power user!
Governed self-service models rotate around IT in charge of data model builds and preparation while business users are in charge of building reports. IT creates a centralized reporting database environment that is enhanced with new data elements with new business needs. Business users build reports on the same data model, leading to lower wait times and improving flexibility. Below is a commonly used enterprise data flow diagram that depicts the data flow in an organization. Business power users can build reports when they have access to flattened out documented reporting data sets. A reporting warehouse is dedicated to every business line with a documented database design.
Image Source: Aditi Raiter
Some cooperative strategies allow reporting tools to directly connect
with ERP/CRM standbys. The question is: Is this the right long-term approach?
Even though standby environments can take some amount of data retrieval loads,
they are not meant to be used as a playground for self-service BI tools. They
are not reporting replicas!
The success of a governed self-service model is also about the processes needed to coordinate people’s roles. According to Dan Madison’s book Process Mapping, Process Improvement, and Process Management, 85 percent of organizational problems fall into the categories of process, control mechanism, or structure, with the bulk of that being in process category. Some companies these days don’t pay attention to governance processes for their self-service BI tool as there is no easy ROI to justify the value of these approaches. They let power users build reports without verification and are willing for IT to later fix reports if the numbers seem strange. On the other hand, a few organizations have very stringent governance processes that make it hard for power users to build reports, discouraging them from creating their own reports. A governed self-service model is the way to go for most of the world’s leading companies. In How to Succeed with Self-Service Analytics, the author describes the process of creating a report governance process.
The intent here is to curate the right balance between flexibility
and governance. The right reporting governance framework must be able to
support data discovery, data security, data quality, audits, and change management
on an interface that has access to flattened-out reporting datasets for power
users. The platform must also be rewarding to power users so that they are
proud of what they can do and feel accomplished.
The platform description starts off by defining user types.
The article referenced above stresses the importance of the power business user role in the report building process. These users also have the authority to raise requests to manage other users on their business teams like casual users and guests. Power users have the felicity to generate their reports on their timelines with approvals from the right authority.
Self Service Analytics — Enterprise High-Level Process
Image Source: Aditi Raiter
The above process can be described with a practical example.
Joe, our protagonist, has joined a logistics company in the operations
department. Joe sets off in this new work environment as a power user to create
a Key Performance Indicator (KPI) report for a warehouse process. He is
required to build a Receipt of Put-Away Performance (by Date Range) Report that
shows detailed information about the average time taken for orders to be put
away. To build the above report, Joe’s manager recommends he attend mandatory HR
power user training.
1. Attend Mandatory Power User Training
It is a lot to expect from operations — that they’ll be able to cultivate data reporting abilities even if it is as simple as dragging and dropping fields from an area on the screen. In most organizations, these are additional responsibilities outside of their day job. As a part of performance management, this type of training is mandated for power users, encouraging system awareness.
Joe, in our example, begins by attending a training program
to understand what reporting means. He gets answers to some of his questions:
- How can on-demand, real-time reporting help retain
a customer? - Is a prospective client an existing customer
with any other business line? - Are they worthy of a gold or platinum service if
they are already a customer? - Are customer orders shipped on time? People find
data worthy only when they can imagine its benefits.
Joe is introduced to
the self-serve reporting process (outlined in Figure 2). He is trained
on building reports utilizing the reporting tool and its features connecting to
the reporting database environment for his business line. He also learns about a
version control tool that helps him check-in and save his work. He now understands
the IT term “metadata” and its usage in search engines deployed within the
company. He also now understands that he has the IT help desk team on his side
if he needs help.
Joe realizes the importance
of his elevated access — if used in the right way, it can help Joe and his
teammates answer frequent questions for themselves.
2. Document Reporting Requirements
Another important step for power users is to gather their needs
on a standard power user reporting requirements template. This template
documents some of the commonly asked requirements for reports, such as:
- Report update frequency
- Report users and their permission types to view
data elements - Summary/detail reports
- Additional data elements useful for comparison
- Outlier definitions
Joe, in our example, fills in the standard reporting power user template with data elements for each order line and the selection criteria as outlined in the image below.
Joe
realizes the following:
- The template helped him organize his thoughts on requirements.
- The signed-off requirements document facilitated Joe and his business team to be on the same page.
3. Metadata and Report Catalog Search
In the article Business Intelligence Meets Metadata Challenges, the author stresses the importance of the growing concern over metadata. Storing data lineage and reporting table content information does not give a full picture of metadata.
Metadata is the catalog of data. It shows you where to find the information needed, the source, where it’s used, and the formulas that it participates in. Data can be anything — IT reporting table columns, IT ETL processes that describe transformation rules and job dependencies, data steward MS Excel templates, IT reports and user bases, business data quality indicators, data content information, IT data models, access permissions, systems, etc. Any documentation that can give more information about data is metadata.
Data catalog tools attract self-serve platforms as they aid in
building trust around enterprise data.
Some report catalogs are implemented, storing report metadata information in a graph database where resources are data tables, dashboards, reports, users, teams, business outcomes, etc. Their connectivity reflects their relationship: consumption, production, association, etc. Airbnb’s data portal success has an interesting approach. In the piece Disrupting Metadata with Metadata Automation, the author talks about the advantages of metadata automation.
Joe, in our example, sets off on a data discovery platform.
He uses a metadata tool to see if the report already exists or if one that
exists can be enhanced to fulfill his requirements. He finds reports that can indicate
a put-away performance, indicating the volume of stock put away per warehouse
clerk per hour. This is close to the logic he needs but not the same. On other
reports, he notices the right logic for his need, but they are not built for
his customer. He begins by testing a
similar report for a different client that satisfies his needs. As the database
and the tables are similar, he can get in touch with the business team who
built the report. Alternatively, Joe has access to the Reporting Warehouse
Database Design document, which indicates the tables that can be used to build
the report from scratch. Joe chooses to enhance a standard report instead.
Joe realizes:
- He no longer has to spend days looking for data.
- He is no longer dependent on IT old-timers who have company experience in metadata because they were partially involved in all projects related to building reporting tables.
4. Collaborate with
Other Power Users
Once the users are aware of built reports and dashboards and have access to the above metadata discovery database, they usually need to collaborate with the teams that have built these reports. A collaboration tool can be used to get like-minded people together to build a business report type. They can interact with the people who have already built or are in the process of building a similar report.
This collaboration platform can be as simple as Microsoft
Teams, Skype, or a more advanced one that enables sharing analytics dashboards
with some of the features like the ones listed below:
- Different streams for different business lines —
a business line lane that could further fork into report groups for that
business - Ability to share experiences, best practices,
discussions around potential improvements, and hurdles to build a report kind - Ability to share dashboards and visuals
- Ability to view, write, edit, or delete
annotations and start new discussions around the data - Option to share dashboards with colleagues or
customers who do not have a tool account by sending them a session link via
email
Joe, in our example, starts off on a collaboration analytics
platform. He sends a request to join the reporting lane to another client. He
reads previous comments on the report experiences. The comments hint that he
should get in touch with a certain person on the team. After getting in touch,
he learns that the existing report built, unfortunately, has hardcoded customer-specific
SKUs and intended locations only.
Joe realizes:
- The collaborative platform helped him quickly get
answers to all his additional questions from the person/team who originally
built a similar report.
5. Assistance from IT Help Desk Support Team
Support
teams can provide technical assistance for solving a software or hardware-related
problem. Part of this group, called the IT help desk, is dedicated to
supporting data-related issues/requests as they support IT applications within
the organization. Obviously, if documentation is not in place or metadata
cannot be found, the task will need to be redirected to core IT teams.
Automated metadata tools can help minimize these concerns. IT help desk support
teams also have additional access to view data lineage from source systems to
reporting platforms.
Data Lineage
and Profiling
Most data catalog or ETL tools have visibility into data profiling, which is the process of collecting summaries and statistics of data from the source. It’s an audit that details the number of null columns, the maximum and minimum value in a field, data quality indicators, etc. Data lineage tools track business data flow from the originating source through all the steps in its lifecycle to destination. Some tools can go all the way to keeping a time slice of the lineage. For example, if a requirement is to know how a field was calculated at some point in time in the past, an option to view previous and current lineage is possible. There are also other open-source data lineage tools if Spark is used for pipelines.
Joe, in our example, sets off by raising a ticket to the IT
help desk. The support team helps
him check the data lineage of the report field between various systems. They
let him know that the report is pulling the right field from the source systems
even though it has some hardcoded SKUs. Joe also verifies a few orders from
source WMS screens rightly pulled by the report. The support team also confirms
that there are no additional transformation rules if Joe builds the same logic
for his customer. With the solution now researched, Joe begins preparing
standard report development documents for the change approval board. He proposes
modifying the existing report as the board appreciates a new customer report
only when necessary. He updates a standard Report Design template.
Joe realizes:
- Approaching the IT help desk was a very convenient option to
finalize his design. Though their help was not mandatory, the process proved to
be beneficial.
6. Center of Excellence (CoE) Report Design Approval
The CoE team
validates the proposal for enabling the report cube for both customers. Their
intent is to verify whether the report tables suggested meet the design
requirements. Some business lines like Freight Management have mandatory user profiling
rules for shipment visibility. The CoE team determines whether the report
satisfies core data visibility authorization concerns. They review the SQLs and
indexes proposed and consider any improvements suggested on the reporting
database.
Joe, in our example, attends the CoE board meeting with the report design template proposing an additional parameter for customer name on the existing report. He submits the SQL used and verifies that the report is performant. User groups can see their own version of the report variants for a customer. The board then approves the report enhancement process.
Joe realizes:
- This kind of strict governance often plays an influential role in curbing broad data mapping issues, incorrect data-related decisions, and redundant evaluations. It was an opportunity for effective self-serve analytics as the design team approved (and initiated) database disk space additions for forecasted volumes. Attending a one-hour call was not burdensome red tape, after all!
7. Report Creation/Enhancement
Some reporting and BI tools come with easy access to online
help and community forums. Online help has some sample reports and dashboard
demos that can come in very handy. A few BI strategies allow relevant master
data loads on reporting databases to add additional details around reference/master
data on the reports.
Joe then modifies the existing standard report on the reporting
tool’s UAT environment. He chooses to involve other teammates so that they can
help with testing. He enhances the KPI report by adding the warehouse id,
country, region, and cluster details.
In a very rare scenario, Joe realizes that the put-away for
a pallet type is not updated at all in the system. He decides to open a ticket with
the IT engineering team to fix the bug. As the bug is not very critical to his
business case, he also considers the option of fixing the bug in release 2.
Joe realizes:
- He can manage and negotiate report delivery
timelines with the customer himself. He is not dependent on IT to quote on fixing
any unexpected problems. - Thankfully, with the IT help desk team’s support,
Joe was not alone in resolving this issue.
Sum (ID)
8. Report Migration to PROD
After the UAT
phase, like any other object, the report is moved to PROD via a help desk and
ticketing software with workflow management. It’s a great deal to consider that
the power user can manage IT tasks that involve documentation related to change
board approvals and releases. This task is better handled by the IT help desk.
Report
enhancements or new report builds are not critical releases involving any major
downtime. In most organizations, the change board does not mandate IT
representation for such requests.
Joe, in our example, opens a help desk ticket to get the report cube released to PROD. His final UAT tested report is checked-in to GIT.
The IT help
desk validates the following details received from the ticket:
- Report cube object to be migrated to PROD
- Guest user list that has execute-only rights to the
report in PROD - Casual users who can use the migrated report
cubes to build a report version in PROD - UAT approval from all business teams who tested
the report cube along with the business need and test case documents
They
additionally verify that:
- The report columns do not have personal
identification information - The report satisfies expected performance
indicators/timelines for common parameter executions - New report metadata is registered into the
system
The IT help desk fills in other typical change board
documents on Joe’s behalf confirming that they have the project details and
critical assessment details for the change board’s review. Once approved, they
get in touch with the release team to move Joe’s objects to PROD.
Joe realizes:
- The request ticket form was in place to maintain
stable and secure systems. His PROD work environment is stable as any changes
to the reports or user access need his approval.
9. Advanced Predictive Analytic Responsibilities
Some
companies find it worthwhile to invest in granting predictive analytics
capabilities to power business users.
Popular uses
of predictive analytics answer two kinds of questions:
- Regression answers “how much”
- Classification answers “which one”
A predictive model is like a car; you need to upgrade it from
time to time, changing its features. Predictive models follow the same
philosophy, but the magnitude of the upgrade is not as drastic. Improving a predictive
model’s accuracy can be achieved by enhancing a couple of features or
predictors.
A predictive
analytics project includes the following tasks:
- Select the target variable
- Get the historical data and prepare it
- Split the data into training and test sets
- Experiment with prediction models and predictors
(features) — pick the most accurate model - Train the model on the training set
- Validate it on the test set with standard
accuracy measures - Implement it — in addition, do not forget to
follow up on it from time to time
A BI tool
that has the following features might thrive in the area of self-service report
generation:
- No coding or scripting
- GUI to common R/Python machine learning packages
- Workbench for advanced analytics
- Stats analysis, visualizations, transformations,
built-in models for pattern discovery and model testing
The above tool features can help achieve rapid model
creation. Product features or power user-training in this area are not
solidified yet. This function still demands an eye to effectively manage a data
scientist role in the hands of a smart power user.
Summary
If the right processes are not in place, even the best brand
tools cannot help to accomplish the insights when needed. Data quality and data
catalog features help build trust around enterprise data. Required governance
helps build confidence in IT operations. The true value of a self-serve
capability is not in a tool but in a framework.
And finally, Joe
realizes that designing great looking BI reports and dashboards isn’t just
about spectacular charts, but the goal is to share easily understandable
information by efficiently getting to the right numbers at the right time!
