"52% of data professionals say they have trouble demonstrating the impact data science has in business outcomes," according to TechRepublic. A big part of the problem (50% according to a recent study) is the difficulty of quickly deploying models into production.
Since the dawn of the millennium, business leaders have been told that their companies need to be more agile to stay competitive. But, unfortunately, the current state of how Data Science teams work is not optimized for agile operations.
Recently, Gartner introduced the concept of XOPs, which aims to accelerate processes and improve the quality of what the different OPs disciplines are delivering: software (DevOps), data (DataOps); AI models (MLOps); and analytics insights (AIOps). One of the critical benefits of XOPs is to create a collaborative environment where XOPs work together to achieve quick production-ready models for the business units to use.
In large organizations, the situation is unsustainable. Data Science teams working transversal to the organization control most of the flows to put models into production. This situation is causing data scientists who are embedded in different business units to suffer the consequences. Given that organizing the core team to deploy a model becomes a painstakingly slow process when a business unit requires it. As one business analyst in Godaddy puts it,
"Our unit has a ton of models, but there are just too many pain points to operationalize them."
In small organizations, Data Scientists need to patiently wait for their CTO to find the time to operationalize models into the tools their various teams need. However, this waiting game poses a threat by threatening to lose focus from the primary goal when building a product. Or if, on the other hand, the Data Scientist is operationalizing its models, it causes them to lose focus, as one Data Scientist of a vital tech startup mentioned.
To solve the disconnect between data scientists and the organization, our team has gathered several recommendations that will help you in the short and long term:
- As Gartner suggests in this article, reducing the duplication of technology and processes and enabling automation will dramatically improve the time to market. The different OPs units need to work together and establish an orchestrated approach to deliver results efficiently to the other business units.
- Data Scientists embedded in business units need the flexibility with tools to prototype and iterate fast. There are core processes that will take some time to get in place—for example, building a scoring model for a financial institution. But let us say the marketing team needs a clustering model to target customers. The team needs the results today, not tomorrow or in a week. Once the model works successfully, the company can decide to integrate the model into its core stack.
- XOPs will allow data scientists to quickly send the output of models to the business applications that the business units need in a collaborative environment. This aspect is critical in the data cycle, considering that currently, since the time to production is taking too long, interest decays and priorities change, creating frustration in the teams involved. With XOPs, data scientists and team members from other departments can collaborate to deploy the models into the apps needed.
Datagran specializes in XOPs to allow Data Scientists in small companies to quickly upload a model and send the output to specific business applications that the company needs. This lack of friction increases the chance for teams to work together after the primary goal. For larger companies, Datagran allows data scientists embedded in business units to prototype, iterate, and quickly make the model core.
Being agile in the Data Science practice will be a significant trend in the years to come. And the teams that can serve models into the business applications promptly will be the ones that can prove their business value.
For more information on how datagran works, visit www.datagran.io
.jpg)
