How Product Teams Get More Value From Data
We’re pleased as punch that Sharon Chou has joined Fortune’s Path as a data scientist to help our clients get more value from their data.
Sharon Chou, PhD and artist
Sharon is a data consultant who has designed and implemented machine learning models, captured decision intelligence, and programmed automation systems for SaaS, tech support, online education, healthcare, tax advisory, and human resources.
As a data scientist, Sharon helps clients boost customer satisfaction, heighten engagement, streamline workflow efficiency, and get more value from their data.
Her past lives include quantum physics researcher, circuit designer, Twitter poet, abstract painter, and recorder/ harmonica player in a rock band that got acquired. (We didn't know rock bands could get acquired, ed.)
Sharon holds a PhD in electrical engineering from Stanford University, and Master’s and Bachelor’s degrees in engineering from the Massachusetts Institute of Technology. Welcome, Sharon!
Data by itself are raw materials, like minerals we mine out of the ground. A lot of the valuable bits are hidden inside ordinary rocks. Knock a hundred stones open, and if you’re lucky, one turns out to be an amethyst geode.
Is it possible to tell which rocks contain precious minerals before breaking them all? Perhaps, if the miner has enough historical data about where the previous hits were located, the rocks’ shape, textures, colors, etc., for both hits and duds.
To create effective predictive models, companies — and especially product and marketing teams — need datasets that demonstrate what to do and what not to do based on previous experience. These datasets are the training data that teach machine learning models how to take the past to inform future outcomes.
Here’s how product teams can get more value from their data.
Get Started with Predictive Modeling
To get going, it’s important to ask:
How easily and elaborately can your teams describe your best revenue-generating clients/services/products? How are they different from the less revenue-generating?
How were the leads generated and by whom? Were any prior relationships leveraged?
How conscientiously are business development teams tracking their interactions with current prospects? Is that information in a centralized data source?
Was the same due diligence done for past prospects for both successful conversions and those who dropped out of the funnel?
Collecting enough relevant data about past and current customers is a prerequisite for driving meaningful actions from data-derived insights. Without it, opportunities are lost. And big opportunities don’t always mean the buzziest technology. Generative AI applications are certainly important, but they’re only a subset of methods to get value out of data.
The Gartner hype cycle just about nails the current journey of AI.
Improve Your Data Prospecting Pipeline
Working with data is like panning for gold. You have to sort through a lot of junk to find the nuggets. Here’re some examples for how to improve your data prospecting pipeline to enable more timely and accurate insights from data.
Streamline Data Collection
Does your team know where to find relevant customer data, and can they do it quickly?
A client’s customer service managers were not documenting their interactions with prospects on the internal software platform. They found the platform too onerous to use, so they kept notes on their local computers instead.
In-depth interviews were conducted to understand how the CSMs do their jobs, e.g., what tools they like to use and why. Then we proposed an improved workflow design to the CSMs, product managers, and in-house developers.
After a few iterations of Agile cycles, the CSMs documented their client interactions more completely, which enabled better customer insights via advanced analytics using long-overdue incoming data.
Improve Data Accuracy
Have you found data inconsistencies or errors that really affected the business?
A custom apparel company was struggling to determine how revenue dropped suddenly last year. After scouring their SQL database entries, we discovered data labeling errors: all the T-shirts that were recorded as a size Large were actually XL. That translated to $7+ million in reclaimed revenue.
In addition, they now obtained the insight that the plus-size sector is larger than they previously knew. To avoid a repeat of this same issue, the client collaboratively designed data review protocols to be implemented on a regular basis.
A favorite Twitter poem Sharon found that demonstrates how easy it is to insert inconsistencies into your data.
Data Visualization
Do you find it hard to draw useful conclusions from your data?
An eye clinic had been using Excel files to track patient visits, and asked for help with upgrading to interactive dashboards. The revamped visualization helped clinic staff discover that revenue from diabetic retinopathy and cataract operations increased dramatically over the last few years.
But the increase was largely from retinopathy operations, not from cataract surgeries — as they previously assumed.
Whether they move towards more diabetic-specific patient services or recruit more patients for cataract surgeries, interactive dashboards make it much easier to analyze data and discover longitudinal time-series trends that aren’t obvious from static reports.
Predictive Modeling with Machine Learning
Ready to leverage predictive modeling, but unsure where to start?
A SaaS company offered software platforms that forecast energy usage and bills for commercial customers and recommended custom energy and bill-saving measures. Utility companies provided energy usage data from smart electric meters with hourly measurements, but lacked the ability to provide forecasting services to those without smart meters.
The challenge for the data science team was to take the very sparse, monthly electric usage data from the non-smart meters to help predict future energy usage. They created a machine learning model trained on data from historical weather, business types, and building operations. The machine learning algorithm was based on 1990’s technology, but it worked!
Natural Language Processing
Tackling any unstructured data like free-form text?
A customer service team was trying to improve customer satisfaction for tech support. The data scientist dissected more than 10K chats between service agents and customers, examining features such as reaction time between exchanges, word length distribution, etc.
The result? Customer satisfaction was highly correlated with the service agents’ punctuation usage, whether or not the issue was actually resolved. Does that mean if the agent uses lots of punctuation, the customer will be satisfied? Not necessarily - remember, correlation doesn’t equal causation. Perhaps the agents who use more punctuation are perceived as more professional? It’s worth digging into with the agents and their trainers as they set new best practice guidelines.
Need Help with Your Data?
These are just a few of the ways to extract more value out of data.
Let us know how the data science team at Fortune’s Path can help you get the most out of data.
We help SaaS and health tech leaders unleash the power of their data with expert product management and competitive intel.
Efficiencies, revenue, purpose — not necessarily in that order.