How does your organization decide which Data Science projects to pursue immediately, out of many proposed projects? Have you developed any particular decision framework for prioritization of Data Science projects?
Sort by:
Excellent acronym, Maritza; thank you for sharing! As a former educator, I always like using tools like rhyme, alliteration, and acronyms as mnemonic aids to make concepts easier for people to remember, categorize, and articulate.
That's a nice and crisp framework!<br>
I can speak from the standpoint of running a data warehouse and prioritizing data analytics, and other stakeholder-driven projects related to data science and other business intelligence related initiatives. We always try to follow a strategic and data-driven process.
First, we line these projects up with our overall organizational goals and objectives, collaborating closely with key stakeholders and management alike. This enables us to understand their priorities and, consequently, identify projects with the highest potential of creating value and having a positive impact.
That being said, we start putting it through a decision grid or framework that takes a number of factors into consideration including:
* Business impact - assessing potential impact of each project on those organizational objectives I mentioned earlier. To do that, we need to evaluate projected ROI and/or other relevant metrics
* Data quality & availability - This is done for each proposed project. Those with reliable and accessible data sources typically will get higher priority (they can be executed more smoothly)
* Technical feasibility - This includes assessing what infrastructure is available, along with both the tools and the needed expertise to deploy the project within the given timeframe.
* Resource Allocation - We look at what resources are available (developers, analysts, DBA for any database structure changes needed or new tables, etc.) to ensure there's enough available for the project
* Time-to-Value - This goes along with the "low-hanging fruit" concept that T. Scott Clendaniel brought up in his answer. That is, we assess the expected time it would take to start getting value from each project. As a result, project with either shorter timelines or less work required (AND from which we would still derive meaningful value) will get the nod before those that do not.
Thank you for sharing Joe. It is quite insightful and comprehensive.<br><br>I used a similar approach and practice that you mentioned above. In addition, there are 5 questions to speed up the decision with key stakeholders and execs, in a sequential order. Any "No" along the way would move the initiative into the "Backburner" corner.<br><br>1. Is it a business pain point*?<br>2. Is it technically feasible?<br>3. Is it financially feasible?<br>4. Is it an instant impact?<br>5. Is its benefits growing over time?<br><br>*This is a bundle question. The answer must address three things: Who was impacted, the Timing (by when the issue needs to be resolved) and the Size of Loss / Opportunity Loss. Note, Timing is everything. Hence, when the timing is right, these initiatives can be brought back to the table for endorsement and rapid execution.
Thanks, Eric; that's a VERY well-thought-out, yet concise strategy you bring up. And thank you also for your kind words, sir.
It has been slightly different in various organisations that I have worked with. However, similar to many other tech investments, it has been mostly driven by the business impact/value vs cost (complexity, time, investment, risk) dimensions.
I guess there is no need to emphasise on the availability of acceptable quantity/quality data as a key to that decision. But I would add that personally I would prioritise projects that the advanced and complex aspects of DS/ML is a value add to a larger initiative over those which are putting all of the eggs in the basket of a complex AI/ML concept.
Thank you. Makes sense.<br>
Depends on the current priorities of growth, value or sustain. Data science projects are generally costly and take time. so depends on multiple factors.
Frankly speaking - unfortunately political and opportunistic interests ... even when it is obvious, that the data is or the corresponding stakeholders are by far not suitable / sufficiently prepared for.
That is very often the reality.<br>
A widely used tool for prioritizing analytics and data science initiatives is the RICE score: (R)each * (I)mpact * (C)onfidence / (E)ffort. This is an opportunity score that helps the organization objectively prioritize the initiatives that will provide the most value with the least amount of effort.