Think of unstructured data like a ball of tangled wool. You can teach a machine to sew anything, but it can only do so once you unravel the wool and straighten the yarn. The SOV problem is somewhat similar. The ideal solution to Ugly Data requires the equivalent of machines untangling balls of tangled threads.
It's not as easy as solving a Rubik's cube. So what would it take to create this truly comprehensive yet seemingly impossible solution to the SOV problem?
Well, that's precisely what I've been toiling away at for the past several months. I can now finally say with a fair degree of certainty that while it's no easy task, any solution worth its salt should, at the very least, factor in ALL of the following insights to be effective. Exclude any one component, and your solution will not be able to live up to the promises that made you invest in it in the first place.
Keep It Simple
If you have to spend more time training your users than you've spent training your software, you're bound to end up with everyone and everything feeling more than a little bugged.
Insurance policy terms are complex—catastrophe models, even more so. The challenge is to balance both perfectly. Rely on a less risky and uncomplicated solution to understand and use.
The solution needed must be so intuitive that it should allow anyone to clean SOVs in seconds without the need to master (or migrate to) a new platform or invest hours in training how to use new software platforms or waiting for results.
Do it all in Excel
Most SOVs come in an Excel format. No other tool is as widely accepted nor nearly as well suited to tabulate details about a finite number of insurable assets. An excel-native tool not only reduces the aforementioned learning curve and risk of transcribing data from one format to another but also continues to offer all the powerful functions, macros, and APIs that Excel will allow. The ideal solution must aid you in enhancing Excel’s functions to filter, replace or delete redundant data in the blink of an eye while learning and remembering each action to effectively predict how to treat such data when it reencounters it.
Leverage the true power of AI & ML
The use of AI and ML as a solid solution to inconsistent and incomplete data should now be just as apparent as it is intuitive.
While this is far from a novel approach, it is prudent to note that many companies have already spent millions of dollars attempting this. Despite promising otherwise, most fail to provide a reliable and comprehensive solution.
A human can train a classification model to identify which sets of categories each observation in an SOV belongs to. But accurately sorting data is only step one, and even here, it is not uncommon to see errors and inconsistencies creep into this aspect of existing SOV scrubbers. The real value of AI & ML lies in augmenting and correcting incomplete and inaccurate data.
It is extremely easy to miss errors on a so-called 'clean' SOV, regardless of whether this is being done at a manual or automated level. So how do you know if every field has been completed and validated as required? This is precisely where the intelligence and sophistication of the cleaning process come in.
Odd as it sounds, it's easier to teach a program what it should know than to teach it how to recognize and highlight what it does not know.
A data solution that can host multiple third-party APIs to augment SOV data while simultaneously allocating a confidence score to the augmentation will allow the user to identify and verify information instantly. This creates a flywheel effect as users are perpetually teaching the software what to do the next time it encounters an anomaly of a similar nature. The aim is not to replace humans but to make our jobs easier.
Crowdsource information
Insurance brokers often send the same SOV to a handful of underwriters at different insurance companies. Each underwriter then sends that SOV to an offshore team to be cleansed and modeled. A comprehensive solution must avoid reinventing the wheel. Take the intelligence learned from one user and share it across the crowd to benefit all.
For example, if a building's construction description is provided as "Wood Frme," no user should ever have to map that construction type to "Wood Frame" once one has "taught" the tool its meaning.
Avoid the Telephone Game
Here's an illustration we're all intimately familiar with.
Software is art - a team sport at that. An Underwriter talks to a Project Manager. The Project Manager talks to a Product Manager. The Product Manager talks to the Team Lead, who then talks to a Front-end team and a Back-end team, among others. Errors typically accumulate in the retellings, so the collective artistic expressions of the software engineers differ significantly from those of the Underwriter, especially when the Underwriter sees the resulting software.
What is really needed is a data solution (team) that can traverse the worlds of underwriting and software coding seamlessly to breathe life into the catastrophe modeler's and underwriter's vision.
When evaluating a data solution, see how many of these requirement boxes it checks because the exception of even one of these could corrupt your source data and have a domino effect on the rest of your risk analysis.
In conclusion, and at the risk of sounding whimsical, the product needs to be intelligent and agile, one that trains non-stop to achieve perfect results every time it runs; in some ways, the Usain Bolt of the software world.
That is exactly what we have created with SOV Wizard, our proprietary SOV solution that helps you clean SOVs transparently and accurately in seconds, without ever leaving Excel.
Click here for a brief preview of the software, or DM me for a quick conversation and a free demo.
Copyright © 2022 Wizard Analytics All rights reserved.
댓글