4 key actions product managers should undertake to drive successful adoption of data science projects

There is a disruption in the retail business with COVID and advanced analytics capabilities that would have taken a couple of years to implement are being rolled out in months. As always, business looks at AI as the panacea to all challenges gripping the retail landscape. Despite business models being upended , one constant factor will always be people and their ability or lack thereof to adopt to the new ways of working. Adopting data science algorithms is much more than switching to a new technology, it involves a change in thinking and problem solving. While data science is used as a catch -all phrase to solve all problems, motivating businesses to rethink the way they do business and adopt the analytics recommendations can still be challenging. Generally in project management roadmaps, there is a separate swim lane to focus on Change Management. The accountability of ensuring a product adoption lies solely on business stake holders and change management orgs. But with the urgency to implement fact driven recommendations to improve accuracy, looking at change management as a separate activity rather than as a fundamental product management responsibility will lead to delayed timelines and lower adoption rate.
There has been multiple cases where companies have invested heavily in AI , only to have a prototype built and rarely rolling it into production. This is mostly due to the hesitancy on the part of the business to let an algorithm make decisions that impact P & L. This is the case despite spending considerable time on business ideation sessions to hone in on problems and researching the right models to recommend solutions. Data science product managers while bridging the gap between business and engineering should actively factor in change management early in their project and continue to don the role of the business client along every step of the process to increase adoption rate.
Building relationships with business users who can champion your product : While setting up demoes with business stakeholders regularly is the norm, product managers should stay more engaged with business users who can influence their peers and bring more credibility to the Data Science recommendations. Carefully choose users who have an extensive background in their business domain and continue to engage with them often. Set up smaller group sessions with business clients to understand their qualms about the data science recommendations. Genuinely listen and learn from their concerns before brain storming on a solution. Embrace this moment of learning; Opening the mind to see the world from the business optics lent to you will help gain trust from business. Building this partnership and convincing these business champions will be crucial to a wider buy-in during product roll out.
Building transparency in algorithms: One of the reasons business users are skeptical about using analytics recommendations is that the logic is a black box to them. Setting up multiple meetings to explain the details will only add to the business team’s frustration with this transition. Being transparent on the algorithm definitely does not mean walking the users through the python code. To create the right engagement and make sense of the algorithm, it is helpful to build visual representations of the input variables and output . A good narrative completely devoid of esoteric analytical allusions and more glanceable is effective. A well put together visual representation of the analysis helps business stakeholders to quickly determine if this framework aligns with the requirement. Familiarity heuristic can be a barrier to innovation among business users who have done things a certain way for a long time. Visuals highlighting correlations and trends can be a good start. Depending on the use case, building a proof-of-concept of the recommendations and allowing users to play with What-if scenarios can convince them to rethink their methods. Ultimately presenting insights that strike a plangent chord for stakeholders is the key to building confidence among users, that advanced analytics models can provide more precise outputs based on multiple signals that cannot be achieved from a spreadsheet.
Building small wins: Investing in machine learning projects should be thought of as an expedition; to make the outcome of this journey more fruitful, it takes more time and financial investment in data profiling, research on the part of data scientists and scaling and optimization by machine learning engineers. But for organizations looking to make radical strategy changes in a short time, waiting on a matured data science model may not seem like an appealing solution. To build credibility, look to setting small goals that focus on achieving an aspect of the long term data science strategy. If the end goal is to build a machine learning model that predicts demand, engage with data analysts to roll out a time series model in a month, while the data scientists research on external factors that impact demand. Instead of waiting on a tool to automate the business process with advanced analytics, build a power bi or tableau report to surface the data science insights to the business users. Seeing the results without being compelled to use the analytics recommendation will significantly reduce the resistance and allow the users to ingest the output at their own pace; while also providing feedback to the data scientists .
Building an empowered data driven culture : It’s not uncommon for business users in today’s environment to be data savvy and have a working knowledge of analytics. Despite the interest in data science, there is always a bit of mordant skepticism towards the disruption it entails. Also in an organizations where business users are already dealing with an influx of innovation, introducing analytics models in their process will be seen as an added complexity. Shaking the steady quotidian routine even with an effective more improved process creates resentment and pushback. While discovery sessions and design thinking workshops will help the data science and product teams understand the current business process and strategy, its also important for business users to be educated on analytics solutions. Facilitating a Analytics 101 session for business users that covers the basics of supervised vs. unsupervised models can help garner some interest. Product managers can work with the data analysts to come up with business relevant datasets that can be used to demonstrate simple clustering or regression analysis. Equipping business users with easy to use BI tools like SAS enterprise miner or Alteryx and guiding them to build predictive models can also help highlight the gain in adopting an advanced machine learning algorithm that recommends based on multiple data signals versus a simple analysis based on historical KPIs.
Successful product Management journey does not stop with product implementation but with pursuing and achieving 100% adoption rate and target ROI. Business co-operation is critical for any new technology, but its especially important with Data science projects which challenges and sometimes disproves fast and intuitive thinking of business users. Finding an ally in the business who can lobby the cause is important. First step is to build trust by listening and learning from your business partner. Uncomplicate the layers of data science by presenting visuals that allow business to understand and gain confidence in the solution. Focus on short term actions that build credibility among your userbase by rolling out proof of concepts and dashboards. Its not easy to unlearn the cognizant biases and Type 1 thinking that drove the pre data science era processes, but democratizing the analytics skillset through analytics trainings will allow users to create insights on their own that challenge their methods. This in turn builds a fact driven culture that will be more open to a new ways of working and uncovering insights.