Data science is great, and I’ve supported ground-breaking projects. However, the media is full of tales of failed projects and even the “winter (or autumn) of AI”. Data lakes can be powerful when done well, but some only become shallow puddles. The reality is that the difference between success and failure isn’t whether you hire data scientists or which emerging technology you embrace, it’s about how you think about your data projects. Talking with an independent expert and making sure that you are set up to achieve realistic outcomes with business value can help. After all, you wouldn’t build a house without consulting an experienced architect … and preferably one not tied to a building firm!

It’s tempting to get underway quickly. There is no denying the initial exhilaration when a data project starts, and the first results look promising. You want to set your company alight with innovation and contemporary solutions.

A tale of two data scientists aka a cautionary tale:

To get going you hired two data scientists, after a sharp intake of breath about the day rate. Perhaps you’re keen to run a proof of concept or build actionable insights and even though you’re hesitant about the spending commitment you want your business to be moving forward. You’ve been pitched a leading data visualisation tool that’s right up there on the upper right of the Gartner quadrant.

You ask the data scientists to deliver predictive modelling around your customers’ needs as your company wants to pre-empt those needs with compelling offerings. “3 months and we’ll have a POC for you”. You’re uneasy, as 3 months is a long time, and the spending commitment is close to £100k. But they’re the experts. You convince the CFO to authorise the spend. A month in and you ask how it’s going. “Great, we’re already able to predict a customer’s needs in 70% of interactions”. You’re impressed. They tell you things are taking longer than they like because they lack the necessary infrastructure. You secure an additional £50k spend and are even more encouraged to hear a month later that the predictions are now made in 85% of interactions.

You’re keen to demonstrate this newfound capability in the wider business, but the data scientists seem reticent. Slightly concerned, you ask how accurate these predictions are. “Well, we are confident that a prediction is right in 55% of cases … but the models are improving and with cleaner and better annotated data, we’d see better outcomes, and we’re only two months into the POC”. A data engineer is needed to build a data pipeline to process and cleanse the data and data annotation is required at an additional cost of £75k. Nervous, you speak with one of your analysts about the situation. Her statistical analysis reveals that there really aren’t any strong signals of customer intent within the data that is being processed. But the data scientists have delivered what they promised – a POC – and their model can predict customer needs …. 55% of the time. So, do you keep throwing money at it or stop?

This scenario is not atypical. In this case the subject was data science but could equally have referenced a data visualisation or a data warehouse project. What could have been done differently? A data review would have guided you in asking the probing questions right from the get-go. Did you need a proof of concept – after all its common knowledge that customer needs can be anticipated, so what were you trying to prove? Were you to trying to prove a value proposition (POV), in order to ascertain whether the company had a viable and sustainable business case for predicting customer needs? Did you want to understand if your data was suitable for the task at hand? In which case pertinent questions to the data analysts and data engineers before setting the data scientists to work might have pre-empted unnecessary and unexpected costs. And what were the measures of success?

Data initiatives often fail because meaningful measures of success were not defined and agreed. In the scenario above, the data scientists (rightly) feel they have delivered what they were asked to do. The issue was that they weren’t measured on the right deliverables. As a stakeholder you wanted accurate predictions while the data scientists thought they were being asked to build a predictive model. Did you have the right structure and skills in place to oversee the delivery of a data project? In many cases, highly skilled executives will benefit from having an independent critical friend / a mentor alongside them, who will pose the awkward questions and bring data experience to bear in driving out the best possible outcomes.

How we work:

As a data leader with Ammonite I work with leadership teams at these early stages. I operate as a mentor and an independent critical friend bringing decades of experience to help you shape your data programme and unleash innovation. Working closely with business leaders I help frame the conversation by carrying out a rapid but probing assessment of your capabilities. Having an honest conversation, I explore the routes to get the best outcomes for your company.

We don’t avoid the difficult conversations but always stay focused on the realistic goals you have set and build together an iterative plan of action that will see you realise early value while staying true to a strategic mission. And sometimes that means ‘not right now’ – perhaps now isn’t the time to invest in a market leading data visualisation tool when you have yet to prove the commercial value of visualisation. Or perhaps your first efforts need to focus on getting the right data assembled and using that data to build exciting prototypes without risking your customers’ data privacy or security.

Beyond this first phase we offer as much or as little ongoing support as you need, whether that be a lighter touch steering and mentoring role or a more involved delivery role that aims to bring you to self-sufficiency over time.

Author: Avi Marco, part of our Data Leaders article series.

Get in touch to discuss our services or this piece in more detail.