As the analytics head, Ana is worried to kick start the project. Although she has finalized Power BI as the best-suited tool for her organization’s insights to drive business growth, she is still trying to define the best practices for the analytics project.
She discusses the same with one of the senior project managers and does her research on what should comprise a successful project for analytics. Here are her findings.
What comprises the cornerstones of a good analytics project?
1. Planning
With the common assumption that Analytics project is not related to IT but only business, plans have failed in the early stages itself. An analytics project needs as much IT involvement as any other development venture heavy on coding and testing. Although coding is not a major part of a business intelligence (BI) project, the implementation needs an equal amount of planning, analysis, design, development and testing phases.
Staffing and resource planning along with infrastructure brainstorming are equally required and should be consulted with the right incumbents. Keeping this in mind, Ana has started looking for a BI expert to build a competent team. Since multiple stakeholders are involved (Business, IT & Source system vendors in few cases), planning the dependencies becomes more critical.
2. Business Involvement
An analytics project needs constant inputs from the business teams. They need to keep pouring ideas and information about data discrepancy, unexpected data variance, missing numbers and right formats after reconciliation.
Moreover, the business team is the right team to define the parameters, KPIs (key performance indicators) and the look and feel of the insights projected on the dashboards. So, one needs to keep them involved from the kickoff phase itself. Due to this, Ana considered the agile method as the desired framework. Refer point No.1 under What should be avoided in an analytics project? for more details.
Having the business team involved from the beginning throughout the project execution phase as compared to having them later at the UAT stage, is a critical success factor.
3. Buffer Requirement
Effort variance would be high in an analytics project being factual and number driven rather than aesthetics. So, this has to be factored during the estimation and planning phase. Since the business and IT involvement needs to be in tandem, with dependencies on either party inevitable, delays are bound to happen.
So, allocate a buffer larger than a regular project in each phase to execute an analytics project with success.
4. Tight Scope Definition
Ana has drawn a pictorial dashboard for the finance head as the first dashboard in the scope of the project’s first phase. Sales lead hopped in and asked to include a similar report for him. Ana has to park aside his requirement by explaining to him the tight scope for the initial phase. She assures him to house his request in the upcoming phase.
Scope creep like the one mentioned above happens in analytics projects like any other development task. But Ana has handled it practically.
Incorporate proper control on the problem definition to fence multiple scope changes. Such measures would save the project overrun. If data sources are diverse and departmental, it is wise to implement a phase-wise project as Ana did.
5. Baselining
BI projects need absolute baselining to establish thresholds and benchmarks. Data being pumped from various channels need to be reconciled, cleaned and baselined according to the standards. This makes the data compatible and easy to manage and represent.
Similarly, in the quality assurance process, the data being compared in reports and dashboards should have a benchmark. If the results are not compared with the right datasets, the correctness cannot be measured and insights fail to meet the purpose. Absence of any benchmark from real scenarios is even worse. So, collate the datasets with the help of business as the benchmark for testing at various phases.
6. Focus on the Business Problem
Translate the business requirements properly into measurable problem statements. Record the expected results in statistical format and not just as a visually appealing dashboard or a calibrating graph line. The owner or the business head of the deliverables should define the problems clearly and explain the expected outcomes. Like, ‘why are the sales dipping in the country’s southern region in spite of an increase in the customers since last quarter’ could be one circumstantial problem statement. ‘What is the churn in the top 10% of customers’ base over a year?’ can be a broader problem statement, the CEO would be interested in.
The project team should get involved to design the desired milestones.
Most Analytics projects appear as a beautification task, which is not true. Ana decides not to lose the focus in the visual appeals of the representation but gets the most out of the data as business insights by overlooking the brilliance of technology, which can be sometimes superfluous.
7. Staffing
Ensure that there is a BI expert/manager who understands the business, is data literate and understands the technology in use. The BI manager should unify these skills to guide the team and accept the challenges from the forefront. Additionally, a business analyst from the business side should work in tandem to standardize KPI definitions and provide clarity on data attribute definitions for building the dashboards.
Give a keen thought to onboard designers knowledgeable about data concepts such as cleansing, profiling and analysis apart from tool expertise. A little awareness in the database and querying language would help in the long run.
What should be avoided in an analytics project?
1. Traditional Project Execution Methodology
Waterfall model—a common software development practice—delivers the desired application in the end while IT team keeps working on it until the deadline. It is considered an obsolete model for analytics projects. Based on numbers and analysis, the system has to be robust and flexible enough to accommodate the changes the business suggests. Hence, an agile framework where a working prototype of the project is enhanced in each cycle to deliver a pre-defined set of goals is best suited for analytics projects.
A hybrid model (best of the waterfall and agile combined) is also gaining popularity nowadays. Both the approaches keep the business and the IT teams in constant touch with each other and the progress of the project, to induce feedback in the best possible ways. Otherwise, in a waterfall model, when the results are out, they can give setbacks because of the gap with intermediate feedback. Business teams come up with a lot more inputs once they see the sample visualizations along with the numbers, which they can correlate to their day-to-day data reporting. Hence, it helps a lot to have iterative mock-ups & review sessions on deciding which visualizations are to be used in the dashboards.
2. Overlooking Data Massaging
Many projects overlook the need for data massaging for the sake of generating quick reports and beautiful dashboards. But the results tumble down to be worthless if the data are not looked, analyzed, reconciled and polished for best results.
A successful project needs to find the gaps in the data, the correctness of the available raw data and the integration of the data coming from various systems. Data reconciliation is an important part to avoid pain points at the later stages. You should handle the missing and unrealistic values at the initial phase. So, don’t underestimate or ignore data massaging. If any manual data massaging or clean-up is being done out of the core system before reporting the final numbers in the dashboards, then such rules need to be completely documented & agreed upon with the management.
3. One Dashboard Fit All
Don’t target multiple audiences with a single dashboard. Ana agrees that advanced analytics allows producing sophisticated reports and dashboards, facilitating distribution with the right information revealed according to the audience.But the same dashboard doesn't need to serve all groups because one size doesn’t fit all.
A cramped output would be clumsy to ingest and may not be fruitful. For example, one analytics dashboard might not help finance, operations, and sales department, irrespective of the customized options you build in it. A dashboard of this kind can suit appropriately the executives for a summarized insight. So, invest time and effort on focused, solution-based dashboards for the intended audience even if it demands multiple designs.
4. Security and Testing
Testing should be a crucial part of the execution plan and it should be rigorous. An analytics project in the new times needs to procure ‘predictive’ (what will happen) and ‘prescriptive’ (what can be done) knowledge and not just ‘descriptive’ (what had already happened) and diagnostics’ (why) results. Hence, rigorous testing comes into play to answer the ‘how to’ and ‘why’ about the projections.
Similarly, the security compels a thoughtful implementation because data should reach only the right and authorized people. With GDPR and other policies governing personal and sensitive information, the project team should know the adversities of data leakage or mishandling.
5. Performance Considerations
Power BI is excellent at the performance side for in-memory and columnar structure but over expectancy with the performance could be a letdown. Do not expect your dashboards to turn up with results in real-time with extreme changes. After all, data processing needs some time.
Querying large datasets for year-on-year calculations or analysis with multiple years of data or huge geography is bound to be slower than monthly or regional results. Likewise, a continent’s level parameter in an embedded map would turn up slower than the same parameter for a city.
So, keep the expectations to realistic levels and your project will satisfy you.
What’s next for Ana’s project?
Ana has completed the planning phase of the analytics project and has all the required resources in place. With the ardent task of kicking off the project at the organization level, she is not only thrilled to use Power BI but also has a dashboard charted out in her mind to gauge the sales of the company.
Interested in knowing how it worked out for Ana? Keep a lookout for our next blog for insightful details.