top of page

DATA STORIES: Lessons Learned in the World of Data

Stephanie Rabin
4 days ago
7 min read

Updated: 2 days ago

Many who know me know I always talk about "telling a story". It sounds simple, but can be quite challenging. At the core of any success is data or rather data-centered decision-making or story telling.


This is what I've learned from many of my treks in the land of data stories.


Data Comes Out of the Woodwork

Most people think about data as that stored in a financial database or out on the IoT. But it can come from anywhere and everywhere.


Some very useful sources of alternative data I have used:

  • Survey output - Apart from your typical data sources, this is such a huge source of information I'll use to make clear decisions....You could survey customers, agents, employees, leadership, anything! But there's a world of things to consider for these:

    • Too long a survey, you lose people

    • Too much personal information at the wrong time, you scare people

    • Too many ways to interpret a question or too complex a question, people will make their own decisions on what you intend

    • Too small a group of respondents, you may not have enough reliable or generalizable feedback

    • All open ended questions, you need to content code the heck out of it


  • Interviews / Focus Groups - You absolutely CAN get data from interviews and focus groups.

    • More complex content coding, but if you ask your cohort all the same questions....their answers are data points!

    • It actually helps in an interview to have a few "easy" questions. Ones where you ask something like, "on a scale of 1-10, where does this land?" or "which word in this list represents your opinion". This both creates data for you and also softens the discussion.


  • Ad Hoc Team Assessments - You may have a collection of fun responses to free (or not so free) assessments. This provides incredibly powerful help with team collaboration or communication strategy.

    • What common themes are on your team? Do you have 20% "Woo's" in the CliftonStrengths?

    • What outliers do you have? Do you have that one strategic thinker that may not yet be tapped to do more complex work?


  • Statistics found in newspaper articles - For one project, I literally read articles around the globe to find instances of fire loss. Pulling enough of these together, I was able to build a statistical study for a clients' exposure to fire loss.


Have the End Goal in Mind

An anything, anywhere, all at once approach is a big trap many get stuck in. Data should be used to prove your point. Exploratory analytics are fantastic and pivotal. But, all analyses need that hypothesis to start or an end in sight.




United We Stand, Divided We Fall

Consolidating your information in one place sounds so simple; but I've seen it go so wrong.


For one company, they budgeted new systems implementations separately from data migration. So, after years of approving the new systems and NOT the data migration in the interest of money savings, they were pennywise and pound foolish. They ended up having 5 policy admin systems and 5 claims systems. You couldn't easily connect the 10 systems to get meaningful analytics. And every analysis was painstakingly checked.


If I could offer one huge piece of advice: keep it all together. Put the time and money into one single source of the truth. The implications in workforce inefficiencies and potentials for error are so large.


The KISS Principle

Data doesn’t have to be hard. The Keep It Simple approach is critical.


I had one client, a major health & wellness informational entity. They were providing a slew of phenomenal analytics on their sales movements to their CEO and getting no feedback in return.


What was wrong? It was hundreds of pages and tabs in a spreadsheet of insights. There was nothing to point to the bottom line? And the room for error was immense.


With just the introduction of a dashboard of key metrics, the CEO was able to visualize the winds, losses and trends for the company. And, the team was able to use this dashboard to find the inconsistencies in the analyses they did, BEFORE they gave it to the CEO. The information was all still there, but it then became the proof behind the dashboard.


Measure Twice, Cut Once

You need to plan your analysis, as it won't always be a snap of the figure. I like to have my "proof" already in a set of slides or mapped out in a spreadsheet. That way I know exactly what I need.


The one thing that gets missed most: linking data. I'll have clients send me their insurance claims information and their policy information for me to be able to look at loss ratios for various cuts of data. Sounds simple, right? But, if I can’t connect the losses to the policy information via, say a common set of years or a policy number, it's impossible to put those trends together.


This same issue happens for some of the voluminous regulatory or rating agency requests that come in. I've seen companies spend hours, days, or weeks on pulling information only to realize they forgot something which sets them back to step 1!


"Keep it Clean"


I think this will be an even bigger challenge as we expand our technological reach of data. You need a standard set of data checks on everything you do and at every step of the analytics process.


I was just on a call with an InsurTech sharing a new data information tool they have built. Phenomenal work. But, I found two big misses in the data just in a 40 minute call. Somehow the source they had was outdated. Could you imagine an underwriting or pricing decision being made on outdated information? Just one error could have catastrophic impact for a company emanating from a single policyholder suing for whatever decision was made on that small piece of information.


We can't always control the "cleanliness" of our data, but we do need to plan for it. One time, we were building a technical product to analyze reported financial data from a selection of companies, providing as-if scenarios for growth. We thought we had thought of everything. We even used ratios of data so that big and small companies could be included in the analysis. But...even that had a snafu. There were no errors in the reported information, so technically it was clean data. But, one of our selected companies had so little exposure in a specific state and line of business, the expense ratio was 10,000%. What did this mean? We needed to then plan for these anomalies in our analyses. And this was but one anomaly we had to seek out and plan for.


"Data is the New Oil"

Raw data is unrefined and valuable only when we process it.


Data is the new oil. It’s valuable, but if unrefined it cannot really be used. It has to be changed into gas, plastic, chemicals, etc. to create a valuable entity that drives profitable activity; so must data be broken down, analyzed for it to have value. --Clive Humby (2006)

This gets right back to the basics of data. We need common definitions and naming conventions for everything. Data should be linked together where and when it can.


There have been systems (sadly more than one) that I worked with that had a wealth of data to help us perform a multitude of business analytics. But, some of the data was stored with field names that made no sense. And in other cases, slightly different naming conventions were used which sometimes was a result of the data elements truly being different and in other cases they were actually the same.


So, now imagine how that wonderful crude oil of data can be used? Only a select few can access or even understand it. Or worse, you can spend months trying to trouble shoot why an analysis isn't making sense only to find the data definition was off.


Numbers don't lie. Really?


Data should be informative and, if done right, will prove what you already knew or, at least, point you in the direction you expected. When it doesn't....dig deeper to confirm you have it right.


This idiom is such a powerful and dangerous one. I had a strong data scientist run some numbers for me on an insurance consumer survey I ran. He came back with conviction that women and men used insurance agents differently according to the survey. This felt like liquid gold of brand new insights no one had. But, then I dug in. He made the ever so simple mistake of not "normalizing" the results. We had more women take the survey...more women responded than men on this pivotal survey question...just not proportionately more.


Black Box Warning

We all know we are in the midst of an industrial revolution as it pertains to AI. The benefits of which I can barely fathom.


But, as an actuary, I am no stranger to the danger of black boxes. When catastrophe models were introduced, we ran up against a huge wave of concern about the black box they were....you put your exposure data in and get a magical set of numbers out telling you your catastrophe risk. And then, a year or so later, that box changes and you have no way of looking in to see what was causing it.


Just today, I was hearing the same story around another wildfire tool that leverages the power of AI and the physics of wildfires themselves. Truly phenomenal insights. But we now have an even bigger black box that can process enormous amounts of information


Recently, with an insurance sector survey on the future of work I compiled, I asked AI to help me categorize some of the responses. Then I asked again. Then I asked again. It took a lot of iterations to get those categories right. And I still had to look into the results to see if the categorization made sense.


So, my only comment here is: be wary and ask a lot of questions.


Back to you...

So what have you seen in terms of your own data stories

Comments


bottom of page