This is a summary of a CAO conference I just attended in Miami. For additional context, read the first post that explains the audience profile.
Two themes emerged from the CAO conference: Things you know you should do but maybe don’t, and new topics on the horizon.
Let’s start with things we know but don’t do (e.g., eat right and exercise but for data analytics). These topics made up about 80% of the event for me which was initially disappointing until I recognized the value between what we say we know and what we actually do. Think about what you could do better from this first list.
Seven Things You Should Already Know (but Probably Don’t Do)
The list is not in any particular order; if you need clarity on anything, please reach out.
- No one wants to know how the sausage is made. Your data scientists need to be able to tell a story without explaining what a p-value is or the difference between a prop and an evar. No jargon. Ever. I actually test my big presentations in front of my nine-year-old. If he gets it then I’m good.
- Data Miners vs. Subject Matter Experts (SMEs). Data miners tend to interpret the data without enough SME context. This means they may not know what’s reasonable. The best case is either training SMEs to become data miners or embedding your data miners in the business. There is a third option, partnering, but this is a much weaker choice.
- Buy, Sell, or Hold. Just tell your audience what to do. I’m reminded of George Carlin’s skit about what women want. Imagine telling your spouse his or her daily score. Honey, today you’re a 50 but last week you were a 38. What does that mean? What should I do? Don’t play games. Just take a stand and tell your stakeholders what to do.
- Single Source of Truth. Both in data sources and in formulas. People will rework the data to display what they think is correct (and what makes them look good) until they cannot. Therefore, you need a tech to pull the data together, business SMEs to refine it, and ideally, the CFO’s office to bless any connect to cost or revenue if you’re going to get real impact.
- Start with Business Value. Where value is lower costs through optimization or efficiency, increase growth by finding rich new types of customers or reduce risk through stronger predictive modeling. Problems worth solving trace their roots back to these fundamental business drivers. But understand, this is an iterative process of learning and trying based on data.
- Democratize the Data. Carefully. Giving stakeholders access to data with multiple sources of the truth is like spilling a bag of sugar in a sandbox of toddlers. No one will be happy with the end result. Instead, bake your data into business-ready cookies and then monitor how much they eat. Consider teaching some more advanced skills and certifying higher levels of access.
- Message Your Data before someone else does. This means creating a data dictionary or user guide to explain the subtle nuances known about the data such as the best dates to use, fields that might go together, what’s rolled up into what. This is particularly relevant for companies who are required to share data publically. Get ahead of the story and protect the data.
Now, here is the 20% of the event that was fresh. These three insightful ideas represent those golden nuggets that make all the difference.
Three Insightful Nuggets:
- Gaining Budget
The ability to gain budget when your department is seen as a cost center is a challenge. One speaker said it is a three-plus year process that tracks to the build-up of value. He outlined it this way:
- Demonstrate value in specific projects (ROI to get more funds for team)
- Create scalable infrastructure that can be reused
- Demonstrate value across multiple use cases by leveraging infrastructure
- Create new businesses (new P&L owners, data monetized, billion dollar value)
Another speaker led a cross-functional COE first to examine what was needed to deliver value to the organization. When asked about redeploying existing teams he said, “We addressed this specifically in the ask. Can we bring these people into this function? No, those people already used for stuff. And we’ll atrophy their knowledge which we need. Leave them alone. All of our ask was for net new resources. And we got it.”
- Measuring ROI
Everyone wanted to know how to measure the ROI of their data analytics team. This question is often muddied by mixing the ROI of project outcomes vs. the ROI of operational purchases. The latter should be seen as a sunk cost. It is necessary just to get in the game. For project ROI, it gets back to revenue, cost, risk, balance sheet, human capital, and your contribution to this. Quantify the upside.
However, that is not easy if your analytics team is decentralized. The common thought was organizations often do not give the data analytics person enough real estate to operate. One speaker said, “If the CEO were the buyer you would get access to every data warehouse in the company.” Rarely do we have universal access. This, in turn, gates how long the journey takes as well as how easily you can connect to ROI.
So another way to look at ROI was to model impact instead. For example, run discrete event simulations to optimize pre/post then attach impact assessment. It might not be revenue, it might be cost optimization. As the bar of analytics continues to rise, the team must be less about shiny dashboards, and more about impact.
- The Rise of the Data Product Manager
The CAO teams that were cranking out the best results were veraciously building on top of their data platforms. Apps, tools, products of all sizes and shapes to both answer worthy business questions (based on use case) and expedite the easy adoption of data. The best teams actually bring a product manager on to do this. Separate product management teams can also help you protect longer-term “build work” from urgent short-term “project work”.
Here are a few examples of these product portfolios from the healthcare space: mortality reduction, capacity management, re-admission prediction, cost analytics, service line analytics, healthcare utilization explorer, clinical variation (how are we treating similar patients differently), risk segmentation, and throughput simulation (to move efficiently in physical spaces).
One Last Tip
Finally, I saved the most valuable tip for last to thank you for reading all the way to the end. This comes from a speaker who was also a neuroscientist. He said, “We are human which means we make emotional decisions first, then justify them with logic.”
Therefore, do not underestimate the value of people liking you first in order to actually believe the data and take action on it. If you need an example of this, just consider the classic case of Semmelweis (link to YT video I just recorded). We think we are rational about data but we are not. We emotionally decide what to believe.
To get your message across, to have a real impact in an organization, the feeling must be, “Allison’s got my back. I may not understand this data completely but I trust her and her judgment.” So to win, you must be seen as a brilliant advisor, not the nerd in the corner. I think we try to capture this with “storytelling” but what it really means is we all need a lot more EQ to go with the IQ.