How can you use your data to boost your sales in insurance and safely cut off unprofitable clients?

We’re now deeply involved in an AI-based dynamic pricing solution that enables our client – a major international insurance company – to free up to 30% of its underwriters' capacity. This solution compares insurance policies and creates customized quotes for brokers within the predefined bounds of profitability. This project also aims to enhance the attractiveness of policies while ensuring acceptable profitability margins. 

Similar solutions already exist in the market, but aren’t mainstream yet. Which means it’s a great opportunity for you to build an exclusive product of your own to gain competitive advantage using your own company’s data and those of your competitors (publicly available, of course).

Insurance pricing is super complex and takes into account a multitude of factors, such as 

• risk profile of the insured
• current market conditions
• competitive landscape 

and more to determine the optimal price for a policy.

Dynamic pricing software is like a super-smart assistant that helps insurers set prices for insurance policies in real time. It uses AI to analyze all of the relevant data and determine the optimal price for a policy.

Here are some of the very obvious benefits of using dynamic pricing software in insurance:

• Improved profitability: Dynamic pricing software can help insurers optimize their pricing strategies.
• More competitive rates for customers: Insurers can make offerings that are cheaper than their competitor’s without badly affecting marginality.
• Increased efficiency in the underwriting process: Save time and money by streamlining the underwriting process, as underwriters won’t have to do the bidding and paperwork for an unprofitable client.
• Reduced risk of adverse selection: Reduce the risk of adverse selection, which occurs when customers with higher-than-average risk are more likely to purchase insurance than customers with lower-than-average risk.

Here's the link to Earnix dynamic pricing engine and Akur8 dynamic pricing platform in case you're curious about what their ready-made solutions can do.

For those who decided to stay with us, here's a short overview of the top features (to our liking) from the two above platforms.

So far, it all sounds pretty optimistic, but we’re in the business, right? There must be challenges out there! There are, and you’re about to find what they are in the next section.

What assumptions are you going to make when working with your data?

1. Your task is to optimize prices for a business that can adjust their pricing dynamically. This business offers goods or services and has the freedom to set prices that fluctuate based on market conditions.
2. Despite potential business-driven limitations, the fundamental assumption is that there's sufficient data for learning and improvement.
3. The sales team has diligently tracked their sales, noting not only the items sold and the prices they fetched but also instances where they attempted but failed to sell an item at a specific price.

Other: pre-existing pricing mechanisms, human involvement in price determination, etc.

How are you going to collect those pricing data?

Through talking to your people

Business knowledge entails engaging with personnel, primarily the sales team, to gain insights into their pricing strategies, whether or not such strategies are readily discernible from the data. Additionally, explore their perspectives on crucial factors for increasing prices to maximize profit in specific sectors.

Through existing pricing solutions

Some of your people may have formulated a theoretical price and integrated it into a system that generates price recommendations. While some of these legacy systems should be replaced, others may require integration alongside sales data.

Through market research

Equally important are market research and competitive information. It is essential to understand not only the business's offerings but also its position within the broader market and how its pricing compares to that of competitors.

What price optimization path are you going to take?

When thinking over your price optimization solution, you’re likely to come up with three price optimization scenarios: 

• revenue optimization
• profit optimization, and 
• market share optimization. 

1. Revenue optimization aims to increase revenue while maintaining a stable market share, potentially involving price adjustments. 
2. Profit optimization involves increasing total profit by focusing on customers with higher profit margins, even if it means losing some market share. 
3. Market share optimization involves lowering prices to capture market segments that were previously unattainable due to high prices.

For your solution, you need to decide which scenario you’re going to focus on or whether it is going to be all three scenarios. Thus, those scenarios determine your initial functionality set.

Each scenario implies usage of multiple forecasting models, not simply one model per one scenario. The model development, in its simplest form, means collecting and analyzing your historical data on sales and applying market research and competitive information on top of that.

How are you going to analyze your company’s historical data?

Price optimization strategy begins with analyzing your own data. 

Data standardization

Frequently, businesses confront pricing challenges due to the absence of a cohesive strategy. Legacy solutions or operational research approaches may be employed, hindering effective pricing. To address this issue, the initial step involves standardizing pricing through the application of machine learning algorithms to historical data. This helps establish standard prices for individual products or the entire business.

By the way, data mapping is another effective way of seeing what stands behind your sales data. Read this article by our Delivery Director, Andrey Zhilitsky, to see how we’re mapping terabytes of insurance data daily for our client using the platform we built from scratch for them.

Data optimization

When you’re done with standardizing your data, let’s move to optimizing your decisions. Now you can leverage insights gained from both successful and unsuccessful decisions, as you have your all data brought to order in front of you. 

This phase commonly entails optimizing revenue through price adjustments driven by factors like customer behavior and competitive dynamics.

How do you understand you’ve got a working solution, not some unfeasible theory?

To prove the validity of a model beyond theoretical concepts, it's essential to demonstrate its practical effectiveness. A typical first step is to reproduce your sales team best practices, establishing a foundation of trust to the model. By modeling your current processes and identifying areas for improvement, it becomes possible to optimize the model further and align it with your business goals.

Next step is to deploy real-time market data to enrich your model. This step may lead you away for further optimization, and that's okay. Model data, compare them to the expected results. Repeat till the model accuracy is about 70% (which is considered good by the industry standards) or above. Continue with Real time visualization using dashboards, charts or graphs of your preference.

By the way, in case you're in doubt which tools to use for BI dashboard development, check out these great articles from my colleagues -- from Alex Shumsky on Bubble and from Ivan Sokolov on Advanced dashboards built using PowerBI, Tableau and QlickView.

Things no one talks about: special clients and discounts

Optimizing price strategies involves not only such factors as revenue, market share, but also – yeah – customer relationships. Businesses may have varying objectives, and it's crucial to tailor the optimization approach accordingly. For example, a company may prioritize a focused customer base with higher prices over a broader market share. Or providing discounts to a very important customer to retain them during a period of business turbulence.

Writing client-specific code might not always be practical in such cases. Nonetheless, the principles and concepts behind the correction factors in price optimization are relatively simple. During the live testing phase, a correction factor can be calculated by analyzing the discrepancies between actual prices and model predictions.

Just before you go exploring the features you’d like to add in your dynamic pricing software, here are a few more implications of dynamic pricing software development.

Data Requirements:

Training AI models requires a significant amount of data. Most companies opt to use their data and develop data models behind their firewalls to tailor Gen AI models to their specific needs.

Pre-Trained Models:

While building foundation models from scratch is effort-intensive, utilizing pre-trained models and incorporating them into your environment is a common approach.

IT Infrastructure Needs:

To support the software, a robust IT infrastructure is necessary. Insurance infrastructure, in particular, presents challenges due to the need to unify and integrate data on a large scale. ML Ops, which ensures the training, deployment, maintenance, and monitoring of models over time, is a critical aspect to consider.

Bias Potential:

AI models carry the potential for bias. It's essential to be aware of this when using generative AI models, which are trained on a vast corpus of data, often the Internet. These models can provide general information but may not always be accurate or reliable.

Data Quality and Talent:

Training models on complete, up-to-date, and accurate data is crucial for reliable results. Having a strong team of experts working on data management and a proper knowledge management strategy is even more important than having the technology in place.

Despite these challenges, dynamic pricing software is becoming increasingly popular in retail, real estate and very soon – I believe – is going to dominate even such a complex area as insurance. As AI technology continues to develop, dynamic pricing software is expected to become even more sophisticated and widely used.

So, here’s your short plan of action if you decide to start with your own dynamic pricing solution:

Phase 1

During the 1st phase, after a thorough data analysis is done, structure all data from quotes, consolidate all variables involved in the financial calculation formulas into a comprehensive dataset. 

Phase 2

The 2nd phase implies training a neural network-powered AI bot on internal computations to ensure the precision of the calculations. This will help you analyze your policies against competitors, rank them based on customer appeal, and propose revised customized quotes which are sometimes even more profitable than the initial ones.

Phase 3

In order to see how the solution works for you, select a single insurance type for the MVP (say, General Liability).

Assign a dedicated Data/Business Analyst to ensure a deep understanding of the cost structure and the pricing mechanisms of insurance policies. This dedicated analyst will work closely with the development team and will also be responsible for validating the accuracy of the results and figures.

The tech team may include only 3 roles: the architect /technical lead, tech PM/BA and a full-stack developer. The insurance Data/Business Analyst is crucial to fill in the domain gaps and test the results credibility.

Phase 4

As soon as MVP is ready and proves valuable, scale it to other business lines/units wherever your decision making is hampered, takes too many variables into account, and time to reach a winning decision.

Pricing strategy is undergoing historic changes right now. Intelligent usage- and event-based dynamic pricing models preventing unnecessary expenses for users and double coverage for carriers will consolidate their positions among insureds and insurers alike. 

Such is the case of California-based Jumpstart, a parametric insurtech utilizing publicly available data on earthquakes to provide affordable event-based insurance. Powered by USGS data, the insurtech offers $10K to $20K payout based on the earthquake intensity. No further questions asked. Low monthly fee, transparent eligibility criteria, no need for loss adjusters and extra fast claim settlement cycle make it an ideal modern insurtech player. A lot of fascinating stuff is going on under Jumpstart’s hood, and it gives me thrills to think what, except for a large set of publicly available data, has made this unique and competitive business model possible. The guys keep it secret, but I guess I know the answer.

AI is still in its infancy, but already helps businesses gain a competitive advantage. A healthy mix of third-party solutions and in-house products well tested on your internal data is what I see as a big next step for AI development for better insurance pricing strategies and business performance assessment.