In the world of usage-based insurance (UBI), Progressive is an exception: Its massive amount of driving data – 110 terabytes covering some two million vehicles, 1.5 billion separate trips and more than 10 billion miles driven – allows it to quickly test new rating factors and to do so with a great degree of accuracy.
"We continue to test new ideas all the time,” says David Pratt, Progressive's general manager of usage-based insurance. “Our research team will come up with a theory for what would predict safe driving. With our big data set, we can test those quickly."
Few other UBI providers are as fortunate. Octo Telematics, which launched its first UBI product in 2002, also has tons of data: 194 billion kilometers of driving performance from two million drivers globally. But most other companies are making educated guesses based on more limited data combined with more traditional ratings factors.
As Nino Tarantino, CEO of Octo Telematics North America, says, "Today, there is no data standard and no clear understanding of which data and how much is required."
For example, Octo works with seven insurers in the U.S. and Canada, and each one asks for a different data set to be collected – everything from hard braking and acceleration to how many trips a driver takes and how long the trips are, as well as when and where they occur.
Actuarial guidelines often cite 100,000 earned car years as the threshold for credibility for a model, says Dwight Hakim, vice president of telematics, Verisk Insurance Solutions, a provider of underwriting data and tools for UBI. An earned car year is equivalent to one car insured for one year.
In traditional motor vehicle insurance, the number of earned car years is used to show state regulators that an insurer’s pricing decision is based on plenty of evidence. This helps reassure regulators, and helps agents selling the insurance.
"Credibility is particularly important when insurers are constructing a rate plan that might increase premiums," Hakim says. "Regulators need to see a model with high credibility if that model might result in rate increases. Assuming the insurer has a good financial position overall, modest rate decreases are easier to justify."
While using earned car years – or the equivalent in telematics driving data – may be critical when an insurer asks regulators to approve a price increase, it may not be strictly necessary to see how different UBI rating characteristics perform, Hakim says.
Trial first, price later
That is what many insurers in need of beginning to test or launch UBI programs are betting on, to avoid the need for a large data set.
For starters, insurers can assume that good drivers self-select for UBI programs. "Chances are [that] the people who try it are more likely to be safe drivers,” says Thomas Hallauer, research and marketing director for telematics consultancy Ptolemus Consulting Group. “So you know you can offer some kind of discount anyway. You also know they will probably stick longer to your contract."
Another strategy, Hallauer says, is for UBI insurers to collect an initial data set through trials, and to then revise their ratings as more data comes in. Coming up on one full year of offering UBI in the U.S., American Family Insurance used just this approach.
"We feel like we got enough to launch, but as we see the data we know we need to refine it," says Pete Frey, personal lines UBI program and product manager, American Family Insurance.
Combining telematics data with traditional rating factors, such as age and location of residence, is yet another strategy. The Hartford is one of many U.S. insurance companies to do just that, saying it makes for finer segmentation for its consumers.
"Almost all programs in the U.S. augment telematics data with other carrier-specific rating factors,” Hakim says. “Assessing the degree of overlap among rating factors to avoid double-counting takes a significant amount of work, but carriers are implementing telematics because they know doing so will help them stay competitive and win market share."
But the benefit of adding more data must equal the cost, Progressive’s Pratt warns. While he agrees that UBI rating will continue to evolve and become more sophisticated, "if it costs a lot to get the information, it's maybe not worth it," he says.
Consumer acceptance is another sticking point. "Everything we use has to be something the customer thinks is fine," Pratt says. "We have to be able to explain to people why it makes sense, why it's actually fair that we do it that way."
Pooled data
Finally, insurers can get larger data sets from third-party telematics data providers, such as Verisk Insurance Solutions, Towers Watson and Octo Telematics.
Verisk offers what it calls “Driving Behavior Database for Modelers,” which makes available to statistical modeling applications: data from telematics devices; exposure, premium and loss information on insured drivers; and third-party data, including weather conditions, road type and traffic flow.
Towers Watson has a pooled data offering that collects telematics data and claims, policy, vehicle and driver information. And it uses this pooled data to score drivers and to provide those scores to insurers. Also available from Towers Watson are models that take into consideration third-party information like maps and weather, road type, population density, weather and angle of the sun.
Octo’s Insight Centre collects global, real-time data from its installed base of Clearbox telematics devices, which customers can then interrogate to inform their UBI offerings.
However, there are caveats when it comes to using pooled data.
Hakim, for example, warns that while amassing large quantities of data is critical, not all data is equally valuable. Its utility depends on its accuracy and completeness; how frequently it's sampled; and the source – whether OBD2 outputs, GPS or accelerometers in cell phones.
"Knowing how each device works and the manner in which the different technologies interact is key,” he says. “The complexities of reading the car’s diagnostic information, appending accelerometer data and then transmitting [it] wirelessly require a deep bench of experience.”
American Family Insurance’s Frey makes a similar point. The availability of aggregated, third-party data is not the issue, he says. While there are plenty of companies offering data to insurers, "carriers are just trying to figure out how to use it,” he says. “The data is almost overwhelming."
The profit question
One of the big, unanswered questions about usage-based insurance is whether insurance carriers will ultimately be more profitable than – or even as profitable as – providers of traditional motor vehicle insurance.
"It's an accepted thing that putting in a UBI program for starters is a cost,” according to Frey, of American Family Insurance. “It is about a longer-term strategy. Ideally, companies hope to achieve more growth."
Among those costs that must be taken into account are the cost of telematics hardware – if an insurer is going the route of using its own devices – and the cost of the back-end infrastructure. Adding UBI on top of an already-existing infrastructure, as major insurers must do, is extremely costly, according to Hallauer, at Ptolemus. In this respect, UBI upstarts that rely on scalable cloud solutions have an advantage, Hallauer adds.
Still, the hope is UBI will ultimately pay off.
Enabling more sophisticated pricing is one benefit. "You want to make sure you have the right rates for the right drivers,” Frey says. “One of the biggest goals is trying to create fair rates eventually."
And insurance discounts are only part of the equation, Hallauer adds. “The pricing decision is not a discount decision; it's how do you change the offering to make it enticing?" he says.
According to Hallauer, discount-based UBI incentives will eventually evolve into a more service-oriented approach that may range from life-saving services, such as calling an ambulance after a crash, to more prosaic ones, such as allowing drivers to get a driving score.
Ultimately, he says, UBI can make the customer feel comfortable in having a stronger link with the insurer.
Value beyond discounts
Frey says American Family Insurance is looking in this direction. While it hasn't pinpointed what services it might offer, it's considering safety services, making drivers aware of their driving habits and stolen vehicle recovery.
Octo’s Nino Tarantino notes that the value-added services approach is more common in Europe, where some insurance companies offer personal safety and security services, instead of discounts.
The environment is different in North America, he says, because of Progressive. The early mover's decision to base its UBI offering on a good-driver discount established the tone for consumer expectations and shaped the market.
According to Pratt, at Progressive, the company tries to set its prices so the profit margin is the same for people who sign up for Snapshot, its UBI product, as it is for people who opt for traditional insurance.
Progressive's theory is that UBI customers will stay with the company longer – and so far that's been the case. Therefore, the lifetime revenue per customer should be higher, even though the margin is the same as for traditional insurance customers.
Progressive may also save some money through helping people drive better. The company noticed that the driving of UBI customers improved when, a couple of years ago, it added the option of turning on audio feedback in the device so that, for example, it beeps when someone brakes hard.
"We see this training effect within the first few weeks people have the device plugged in,” Pratt says. “They learn to avoid hard braking, and we have evidence it helps people avoid accidents. That could be a big change in the industry, trying to actually reduce your risk."
But the big win is likely in the life of the customer relationship. Pratt adds. “We are not trying to make ourselves more profitable with this. We are trying to attract and keep good customers for a long time."
In fact, Tarantino thinks that most usage-based insurance programs cannot succeed if they're based only on discounts because of all the additional costs associated with them. He says Octo's experience with 2.4 million insured drivers in Europe shows that "when the benefits of pricing and determining risk are combined with the benefits of the understanding the moment of loss for the insurance company – when there is a crash, that is – it will be successful."
There is never enough data
So is there any such thing as enough data?
Pratt doesn't think so. Progressive’s Snapshot considers mileage, time of day and hard braking. The company recently began a pilot using GPS-enabled devices to examine whether highway versus city street driving can contribute to predicting future losses.
Progressive has found that the measurement of how someone drives is indeed a better predictor of risk than driving record, age, gender or any of the traditional rating factors. Still, "the models can still get a lot better," Pratt says.
Say Progressive wants to find out if people who are low-mileage drivers are safer drivers. The company can segment out those drivers from its customer base and see which ultimately did have accidents and which didn't. It could go further and segment the low-mileage drivers by age to determine whether low mileage is a good predictor for all age groups.
Even with 10 billion miles driven, there may not always be enough drivers in the database to provide a meaningful segment to test a theory, Pratt says.
This article has been produced in the lead-up to the Insurance Telematics USA conference and exhibition, which will take place in Chicago on Sept. 3-4. This conference will tackle the steps needed to take UBI programs to the mass market. Find out more
here.