AI product strategy is responsible for turning a business’s vision for AI into a roadmap that leads to products/services with bottom line impacts. Product strategy is an obvious need, but traditional approaches don’t work on their own. Data science and machine learning projects require new gates and processes, or they go off the rails.
I use five questions to start the process of building an AI strategy. What I’m going to outline in this post is how those questions are different for AI projects than for traditional technical projects. I will also discuss examples from my experience about how different answers can push the product strategy onto completely different execution paths. I’ll start with a question that aligns goals with realistic initiatives.
What goals can data science and machine learning support?
Step 1 in turning a company’s vision for AI based products into reality is determining what goals data science and machine learning can support. AI is often oversold as a cure for everything and that’s not a good view of the technology. Not every business goal will be supported by data science or machine learning. It’s a knowledge intensive task to parse between what’s practical or sensible and what isn’t.
I worked with a mid-sized hospital. Their highest priority business goal was to improve patient outcomes. While that doesn’t sound like a data science or machine learning problem, it is. Their goal is an example of an optimization problem. AI handles these very well.
Another example comes from a retail client. Their goal was to improve product margins. Again, this doesn’t sound like a data science or machine learning problem but it’s also an optimization problem. Optimizing pricing strategy with AI is a proven approach for increasing margins.
Business problems aren’t phrased as technical challenges. It’s the AI product strategy that teases this relationship out. The product strategy becomes the translator that lets senior leaders know what the technology is capable of supporting. It also translates business needs into a technical problem that data science and machine learning teams can solve.
What is the performance gain over the next best alternative?
Is there a benefit to using data science or machine learning? The knee jerk answer is, ‘Well of course there is.’ However, that’s not always the case. Many in place solutions perform as well as AI based solutions. Often, the gain in accuracy or performance doesn’t justify the cost.
This is an important gate to put in place. AI based systems need to sing for their supper. If there’s no clear benefit to be had, there’s no point in putting expensive resources on the project.
I’ll ask data scientists about their past projects and my final question is, ‘What was the business impact of that project?’ Most don’t have an answer. It’s beyond the scope of their deliverables. They are accountable for accuracy and performance. Someone needs to be looking at the business impacts of those improvements up front. Otherwise companies find themselves building AI systems because they can, not because they should.
Do we build internally or buy externally?
This is a traditional product strategy question. Buy or build? With data science and machine learning based products, this question requires both technical knowledge and business acumen to answer. Ask a data scientist, they’ll always say, ‘Build.’ However, that’s not always the right answer.
In the case of my retail client, their pricing strategy is a key component of their business model. Pricing strategy is also something that isn’t available from off the shelf software, but it can be supported with open source tools. Both reasons drove a recommendation to build internally leveraging a suite a tools to accelerate delivery time.
For my hospital client, they had the data and in place resources to build several solutions to improve patient outcomes. However, there were also off the shelf solutions for many of their use cases. We chose a hybrid approach. They built for two use cases where their data gave them a significant advantage over off the shelf solutions. They bought where off the shelf solutions would accelerate their implementations.
Every business case is different and it’s the job of the AI product strategy to decide which ones should be built and which ones are better bought.
Who do we need to hire or how do we select the right product?
Does the business have the right data science and machine learning talent to build the solution? This is a question that few businesses can answer. An incomplete product strategy leaves gaps in understanding around what skills are needed.
That leads to generalist hiring. If the business doesn’t have a clear idea of what they’re building, they must hire people who can build anything the business might need. Generalists with these broad skillsets are expensive and difficult to find.
The AI product strategy brings certainty to hiring. The specification of what’s being built to support the business goals feeds into specialist hiring. Specialists are easier to find (taking half the time to hire) and build teams around.
The need for contractors versus permanent hires is also clearer because the product strategy differentiates between one off projects and ongoing initiatives. For my hospital client, contractors were the right way to go for the two projects they built internally. My retail client required permanent hires to support their long-term project.
The other side of the coin is selecting the right product if the business chooses to buy. The same idea applies here as well. A business that doesn’t have a strong concept of what they need to buy ends up buying a solution that can be stretched to handle whatever the business might need from it. This leads to purchasing a solution that goes well beyond the actual business needs increasing the costs unnecessarily.
Hiring and purchasing both need to draw a straight line to the business needs and product roadmap specifics. That connection allows the business to hire and buy exactly what the business needs instead of generally what the business might need.
How do we measure the improvement?
With the second question, ‘What’s the performance gain over the next best alternative?’ the business made some predictions about project results. The product strategy holds each initiative accountable for delivering those results, on time, and on budget.
Early data science and machine learning projects often miss one or more deliverables. Many businesses accept this as a new normal for these types of projects. That’s a misconception. AI projects should have a cadence and certainty.
The complexity of these projects and uncertainty of results is an offshoot of partial AI product strategy. Business goals have concrete deliverables. When data science and machine learning are tied to those goals through a product strategy, results of each initiative can be measured as bottom line impacts. The final goal of the AI product strategy is to bring each project full circle.
Nothing can be learned or improved without this straight-line connection. Data science and machine learning projects, like all other technical projects, are always going to be an effort at continuous improvement. How can the business deliver better results? How can time to insight be reduced? How can users be better served? Answering these questions is only possible if the business has the tools to hold projects accountable for solid deliverables.