The Declining Era of Data Science Teams as Mere Service Providers
What would you want? Just AI, or a solution that delivers value?
#beyondAI
In many in-house Data Science teams, you will rarely see an AI Product Manager being involved. Their current operating model is focused on delivering outputs, not outcomes. And for this business model, well, indeed, there is no need for an AI PM.
But if you simply deliver AI solutions based on the customer's request, you shouldn't be surprised if you are replaced by others or even completely dissolved 😲. If you see yourself as an internal AI service provider, you might think that you are not responsible for the value of your deliverables since you are delivering what was requested, right?
It’s the business that is actually requesting something, so they should be fully aware of what they are asking for. Honestly, it’s quite evident that we, as AI experts living in our own bubble, may have a bit of a confirmation bias; we rarely see anything beyond Algorithms, Models, and distributed computing technology. Of course, sometimes we also need a coffee break and might accidentally bump into other colleagues in the office kitchen. I know I'm exaggerating, but I hope you understand my point. It's easy to forget that there are others out there who don't speak our language.
If you are "selling" your expertise and know-how, turning your service into the actual product, AND if your customers are happy and actively using what you’ve delivered, congrats on your working business model. Continue!
There seems to be someone very AI-affine business guy on the other end, handling all the necessary upfront and downstream work, kind of an AIPM but located on the business side. Potentially, that guy isn’t even aware that he is silently doing this kind of job.
Anyway, this is really rare, not to say a unicorn moment.
But what if there isn’t someone on the business side? What if your customer isn’t happy with what you have delivered, and is not giving you honest feedback? 🤔 What if your deliverables are not being utilized as intended or, worse, not used at all?
Unfortunately, this is more often the case than the first scenario. It is a common pitfall for AI teams operating as internal service providers. It highlights a crucial aspect often overlooked: the importance of aligning with the organization's strategic objectives and understanding the end-users needs. These needs, in 99% of cases, are not yet satisfied by delivering just an AI Model. It requires the incorporation of all components, necessary to knit all components together so that the chain of components can reveal its real value. Thinking of AI Products as part of a larger value chain has helped me a lot to understand where value really materializes.
Merely fulfilling requests without a deep understanding of the underlying business problem can lead to solutions that are technically sound but lack practical applicability or user acceptance. This disconnect not only devalues the work of the AI team 😥 but can also lead to questions about its relevance and sustainability within the company.
An Example from the Real World:
Low Adoption of an AI-Driven Risk Assessment Tool
A fintech firm found itself at a crossroads. Despite having a talented team of data scientists and AI specialists, the company noticed a disturbing trend: their AI initiatives were technically advanced but failed to make a significant impact on the business's bottom line. The data science team operated in a vacuum, delivering projects based on internal requests without a clear understanding of the end-users' needs.
The realization came during the development of their AI-driven risk assessment tool. Despite its sophisticated algorithm, the tool was underutilized and didn’t resonate with the end-users in the risk management department. Feedback was scarce, and what little was received pointed to a lack of alignment with actual work processes and user expectations.
It was time for a change.
They introduced an AI Product Management role, aimed at bridging the gap between the data science team and the business units. The first task for the newly appointed AI Product Manager was to conduct a series of workshops with stakeholders from the risk management department to understand their daily challenges and needs.
In these workshops, the AI Product Manager and the data science team discovered a critical pain point that had been overlooked: the risk management department needed a tool that could not only assess risks but also provide actionable insights and recommendations within the context of their existing workflows. Previously, the tool delivered complex data and predictions that, while accurate, required further analysis and interpretation by the team, slowing down decision-making processes. This need for additional steps was a significant barrier to the tool's adoption and usefulness.
Armed with this insight, they redesigned the risk assessment tool to integrate seamlessly into the risk management workflow. The tool was enhanced with features to automatically translate its risk predictions into actionable insights and recommended next steps, tailored to the specifics of each case it assessed.
This functionality significantly reduced the workload on the risk management team, enabling them to make quicker, more informed decisions. It directly addressed the users' pain points.
Nothing highly sophisticated, as you might have noticed, but it was something that needed to be revealed first. However, this task becomes quite difficult if no one feels responsible for identifying this gap.
The era where Data Science teams could lay back and act merely as service providers is ending.
It was tolerated for the last decade, due to AI's complexity and novelty. Companies consciously or unconsciously granted a field where AI experts could play and try different things out.
But now, the time for accountability is coming closer.
Luckily, I see more and more businesses looking for a change; the number of AIPM and DPM hires is steadily increasing. I now observe teams changing their mentality in time. They transitioned from a purely service-oriented mindset to one that encompasses a product-oriented approach. This means actively engaging with stakeholders, understanding the pain points, and being involved in the journey right from problem identification to solution implementation and beyond.
It's about ensuring that the AI solutions developed are not just desired but are integral to driving business value and enhancing the user experience.
Companies are looking for value, not just solutions.
It’s about delivering an AI PRODUCT!
JBK 🕊️
Do you feel like the Data PM (DPM) role will grow in parellel to the AI PM role? There's a fair amount of DPMs wearing AI PM hats. Or data scientists and engineers filling that role without the title. Especially the DPM role.