This step-by-step guide will help you create a vision you can share with team members, partners, and bosses. Get buy-in to build AI / machine learning products, and align on key analyses to tackle this upcoming year. All to help you evolve away from ad-hoc data pulls.
Let’s get started.
1- We aim to ____
Let’s first acknowledge the organization’s strategic objectives. They’ll serve as a north star to guide our roadmap ideation and brainstorm sessions.
The more SMART the objective, the better. It offers less room for interpretation, and it allows more people to work towards a common goal.
For example, a vague or non-specific goal of "reducing our carbon footprint" could be a 0.1% reduction relative to last year, yet could also be a 50% reduction relative to the year 1800, or anything in-between. It leaves too much room for misinterpretation, making strategic alignment difficult to achieve.
On the other hand, a goal "to limit global warming to 2 degrees Celsius relative to pre-industrial years, by achieving carbon neutrality in the year 2150," leaves much less room for debate (See Paris Agreement).
A specific goal for business teams may be to "hit 50% growth in sales year over year, via upsells and repeat purchases."
2- Today we ____
I’m a big fan of the ideal design process exposed by Ackoff (affiliate link), whereby we first start by painting a picture of the current state.
One way to get started is to draw a process diagram. It can show a supply chain journey, a customer experience, or any process. For example, here’s a simplified e-commerce buying process:
Once we’ve visualized the process, we can layer on the current state of Analytics work and products supporting this process. I recommend classifying analytics work into the following 3 categories:
- Visibility to gather data and show what’s happening or happened: e.g. datasets, ETLs, data warehousing, dashboards, ad-hoc data reports, business reviews, etc. (foundation analytics)
- Forecasts to answer "what if": e.g. annual sales forecast, AB tests or counterfactuals of product changes, simulation of process changes, etc. (Needs visibility first)
- Optimization to support decisions and improve a KPI: e.g. maximize customer lifetime value, minimize the cost of shipping, etc. (Need visibility, and maybe forecasts first)
The buying process diagram could now look like this:
It’s often tempting skip to optimization or forecasting work, apply auto-ML to a data file, but don’t. Start by getting everyone visibility into the current state, so that we all get a sense of the problem and how bad it is. Otherwise we could be spending weeks and months developing a machine learning product toward a problem nobody cares about.
3- Ideally we should ____
Let’s now fill the gap between the [current state of analytics] and the [strategic objective] with the [future state of analytics]. Let’s brainstorm what analytics work will help the organization hit its goals.
I personally like to start by focusing on tangible problems before allowing our imagination to run wild. Specifically, I look at the current state and ask:
- "Why do we do this and for who?" to acknowledge whether we’re doing something for end-customers, internal stakeholders, or someone else, and expose why it’s "important." This helps 1) identify things we may not need to do as they don’t add value, and 2) prioritize work that impacts the end customers.
- "If nothing changes, how will this fail in the next 10 years?" to surface key deficiencies that need to be addressed. This often sparks ideas on what needs to improve.
It’s helpful to involve key partner teams at this phase – people that will either collaborate with us on the build or invest resources in our projects. For two reasons:
- It allows us to brainstorm changes to processes and products, and move beyond analytics.
- Collaboration allows teams to align their priorities during the execution phase of the Strategy.
Involving different types of analytics professionals is also key. We tend to bias toward solutions we’re familiar with, so involving people from all areas of analytics will uncover many more ideas: data engineering, business intelligence, data analysts/scientists, and even product designers/managers for product analytics ideas. Together we’re more likely to imagine a holistic future covering all aspects of analytics: Visibility, Forecasts, and Optimizations.
It’s finally time to imagine the future state now that we have the right people and the right problems. Start from a blank slate.
There should be a very messy board at the end, but with plenty of inspiring analytics initiatives. Something like, but much better than this:
4- Fill the gap
The last step to our workshop is to create a tangible plan. To start moving away from the current state, and start building the ideal state.
In other words, 1) create the initiatives, and 2) prioritize the initiatives.
On creating initiatives, here’s an intro blog post on the basics or writing up agile initiatives (themes, epics, stories). With regard to the "Acceptance Criteria" section of a story, functionality and requirements are likely only relevant to product work like dashboards or data pipelines. For analyses and reports, I find "Questions to answer" to be more relevant.
On prioritizing initiatives, it’s obvious we want to invest in the most valuable ones first. There are however many other dimensions to consider:
- Value vs. Urgency: What about low-value and high urgency work? Or vice-versa? There’s likely no "right" answer, but it can help to quantify the impact an initiative will have, estimate effort, and get feedback from partners. So start with an opportunity analysis before going too deep.
- Is it realistic? The lack of key resources (tools, money, people) or support from leadership or partners needs to be identified. They’re prerequisites. We’d only start any initiatives once prerequisites have been checked off.
- Is it motivating? A good strategy or roadmap needs to feel purposeful and motivating to the folks executing it. It’s the whole point of the workshop – to allow analytics teams themselves to shape their plan. Ranking initiatives by how much the team cares about them will help weigh against value and urgency.
So there we have it: 4 steps to develop an analytics roadmap.
To my fellow analytics people, I hope you can use this as a reference to shape a long-term plan. Move away from ad-hoc data pulls and toward optimization and automation work.
For more Data guides, see using data to drive product design and presenting data stories like a Pixar filmmaker.
Help me write more. Follow me if you enjoyed reading!
