In the realm of AI and analytics applications, a recurring challenge plagues companies – models built by data scientists often don’t make it to production. A recent survey revealed that the majority of data scientists witness only a fraction of their models, around 20%, being deployed.
The Power of Data Products:
In response, companies are turning to the concept of data products – reusable datasets designed to tackle specific business problems. These products, whether incorporating AI or analytics, prove potent for large, legacy companies. At Vista and Regions Bank, data products have translated into significant profits, with millions in recurring earnings.
The Dilemma for Legacy Companies:
Yet, legacy companies grapple with the transition to data products. Accustomed to tangible products, they face internal and customer implementation challenges. Chief Data Officers (CDOs), despite understanding data well, lack inherent product management skills. Data scientists, while adept at creating models, often consider their job done once the model fits the data.
Enter the Data Product Manager:
To bridge this gap, legacy companies need a new role: the data product manager. Unlike CDOs and data scientists, these managers don’t possess the technical expertise to create models or engineer data. Instead, they excel at managing cross-functional product development, leading diverse teams, and effectively communicating with business leaders affected by the models.
Unique Skill Set:
Data product managers share skills with software product managers, such as coordinating across functions, managing diverse teams, influencing without formal authority, and understanding customer needs. However, they require additional expertise in data capture, extraction, quality improvement, integration, analytics, AI, and basic statistics. They navigate the unique landscape where data and software intersect.
Development and Iteration:
Data products typically follow a Minimum Viable Product (MVP) approach, evolving through ongoing iterations. The initial goal, as suggested by Vista’s Klapdor, is the creation of a “minimum lovable product” before scaling up. Successful data product management extends beyond deployment, ensuring ongoing use and value measurement on a quarterly basis.
Ideal Candidates:
The consensus among experts is that data product managers need both business orientation and familiarity with data and analytics. They must lead diverse teams effectively. Data scientists, often focused on model optimization, might not be the ideal fit, except for highly technical data products. Candidates with domain expertise or seasoned product managers from software companies are preferred.
New Economy Job:
Data product manager emerges as a new job in the digitized economy, necessitated by pervasive data and analytics. Despite the rise of specialized roles like data scientists and engineers, the need for generalists capable of bridging diverse roles and delivering value remains indispensable.
Credit to: Thomas. H Daveport
