Utilizing Natural Language Processing in the Health Insurance Industry
Whether or not you think you do, you actually do need health insurance
No matter how healthy you are, you will always be prone to catching a cold, falling victim to the flu, spraining your ankle, etc… As a result, you will have to visit a doctor at any given moment. The expenses associated with visiting a doctor can be high, and because of this, it’s wise to invest in health insurance, which covers, at the very least, a portion of a person’s medical expenses. Most companies will provide health insurance plans for employees. For everyone else, they have to find a health insurance provider that gives them the most value for as little money as possible.
Whether or not you think they do, health insurance companies need people who need health insurance
On the flip side, if you are a health insurance company, your goal is to maintain fair ownership of the available market for health insurance. With only a finite number of people at any given time who are eligible for your health insurance plan, you need to make sure you are optimizing your bids in such a way that you give yourself the best chance to maintain the highest amount of market share you can. However, with the ever-changing bidding landscape, along with the constantly changing competition, it can be difficult to scale your program from year to year to ensure your bids are optimal every single day. This is because every day, you need to:
- Make sure you are advertising to the people who are the best fit for your health insurance plan
- Make sure your keywords are up to date, in terms of semantics and bids
Although the first bullet point above can be hard to address at first, it can be handled through proper audiences and demographic bid adjustments. The second bullet point is a bit harder to fulfill, given not only do you need to maintain optimal bids on your current set of keywords, but you need to regularly expand your keyword list to expand your program. On the surface, this can be a daunting task since you need to guess the starting bids on these keywords with zero historical data.
The future will happen, and it’s getting closer. Fast.
2020 is rolling around, and you need to start thinking about expanding your keywords to keep up with search demand. There are likely a lot of keywords you won’t think of at first, however, you do know you will need to generate keywords specific to the year 2020. A sample of keywords you will create might include:
[best health insurance for 2020]
+2020 +health +insurance
+why +is +health +insurance +so +expensive +in +2020
While a bit arduous, the easy part of ramping up for 2020 is figuring out which 2020-specific keywords to add to your accounts (this can be done in a multitude of ways, with the easiest way being to take your top-performing 2019 keywords and replaying ‘2019’ with ‘2020’). The hard part comes in estimating what the starting bids should be.
Considerations for setting bids on new keywords
With bid landscape tools available, you might be able to figure out which starting bid gets you a certain amount of clicks. However, you need to understand the value of the keywords to get a sense of which starting CPC will help you meet your overall business goal; whether that’s maximizing profit, hitting a certain ROAS, CPA, etc… This is where natural language processing (NLP) comes in, which will help you estimate the value of a brand new keyword, and help you jumpstart your bids and get you ahead of the competitive landscape curve.
NLP will help you enjoy the future faster than ever
To estimate a keyword’s worth (whether that’s revenue per click, conversion rate, etc…), you will need to use its historical data to calculate its value. For brand new keywords with zero historical data, you will need some sort of data aggregation method to use outside data to estimate the value of these keywords. One such aggregation method is NLP, a method used to calculate the similarity score between two strings of characters. There are lots of different flavors of NLP (TF-IDF, Cosine, etc… ), and you’ll have to find the correct method for your particular use case.
Once you figure out the correct NLP method, you can use this to find semantically similar keywords to the brand new keywords that you can use in order to estimate the values of the new keywords.
Example: [2020 health insurance]
The following chart shows a handful of keywords, and their similarity scores (on a scale of 0.00 – 1.00) relative to [2020 health insurance]:
The keyword ‘2019 health insurance’ has a relatively high similarity score to [2020 health insurance], while ‘health insurance for me’, has a relatively lower similarity score.
Factors to take into consideration when utilizing NLP
NLP can be powerful, but only if used on the correct data. When determining which higher-volume keywords to compare to keywords without data, you need to:
Only use high-volume keywords for the NLP calculation
The higher volume a keyword is, the more representative its known value will be. Because of this, you only want to base NLP on keywords you are VERY certain have accurate known values. You don’t get this with low-volume keywords so you wouldn’t want to use low-volume keywords to predict the bid for a brand new keyword.
Make sure the keyword you are estimating the bid for is grouped in an apples to apples manner with the keywords it will use for NLP
You need to make sure the keywords your using to estimate the brand new keyword are a fair representation of how that keyword SHOULD have its bid calculated For example, if your campaigns are segmented by county, the high-volume keyword [2019 health insurance near me] might behave differently depending on which county this is being searched, i.e. it might be a more expensive keyword in San Francisco relative to Fresno.
Natural Language Processing was built for Health Insurance paid search marketers who are trying to get an edge
In the paid search marketplace, health insurance is a tough field to compete in. With tools such as NLP at your disposal, though, you can give yourself a clear edge by getting the most value out of your new keywords before your competitors do.