Know less than the platform knows
An insurer selling micro-policies through a shopping platform is often offered far more data about each buyer than the risk requires. It should take only what the insured event needs, price on that, and keep a full record of every policy, however cheap.
Some of the largest volumes in Chinese property and casualty insurance come from policies most customers barely notice. Return-shipping cover, delivery-delay cover and screen-damage cover are offered at the checkout of e-commerce and travel platforms for a yuan or two, and they are bought in the hundreds of millions each year. I spent four years building the systems behind such policies at Yunzhu Online Property Insurance in Hangzhou, and Hesui still sells a smaller online book of similar products.
The engineering is demanding. Issuance must answer within the platform's time limit or the sale fails, and festival days can bring ten times an ordinary day's traffic. But the hardest decisions in this business are not about capacity. They are about data. An insurer embedded in a platform's checkout can be offered almost everything the platform knows about the buyer. My argument is that it should decline most of it.
The offer of a richer feed
In 2019 return-shipping cover was Yunzhu's largest line: about 870 million policies at an average premium below one yuan. The loss ratio was 71 per cent and rising, because more shoppers were returning more goods. A large platform partner offered us a feed of each buyer's return history, so that we could price cover separately for frequent and infrequent returners. Our pricing team estimated an improvement of about 4.5 points in the loss ratio. Competitors were reported to be pricing this way already.
I argued against accepting the feed, and after two months of analysis our chief executive agreed. The argument had three parts.
The first was legal and ethical. The Cybersecurity Law already required personal information to be collected lawfully, properly and only as far as necessary, and a national personal-information protection law was then being drafted. A buyer's whole return history, across every merchant and category, was more than we needed to insure one parcel's return. The Personal Information Protection Law, in force since November 2021, has since made the principles of necessity and minimisation explicit, and it asks that automated decisions be transparent and fair and not impose unreasonable differences in trading terms such as price. Had we accepted the feed, we would have been unwinding buyer-level pricing within two years.
The second was explanation. A shopper charged 1.80 yuan when a neighbour paid 0.60 had no way to see why. Many frequent returners were not abusing anything. They bought clothing from merchants whose sizes varied, and returned what did not fit. Pricing on their history would have charged them for the merchants' inconsistency.
The third was dependence. A feed we could not verify, from a partner that could change or withdraw it at will, would have become the core of our pricing. Our model would have been only as good as the partner's data and only as durable as the partner's goodwill.
Price the risk you insure
The alternative was to price on what actually drove the risk: the merchant, the product category and the characteristics of the order. Merchants whose goods were frequently returned, or frequently complained about, carried higher rates. Clothing and shoes carried higher rates than books or kitchen goods. Large or heavy parcels cost more to return and were priced accordingly.
This approach cost us. The loss ratio rose to 74.8 per cent in 2020, 3.8 points higher, and in two clothing categories we lost share to a competitor pricing on buyer history. It also showed us a risk the buyer feed would have hidden. Some of the worst losses came from merchant collusion: related accounts buying and returning goods to collect the shipping allowance. Merchant-level detection recovered RMB 23 million in 2020. A model focused on buyers would have looked in the wrong place.
There was a commercial argument as well, which I underplayed at the time. Merchants respond to their own rates. Once merchants could see that frequent returns raised the price of cover on their goods, several corrected their size charts and product photographs, and their return rates fell. A price attached to a buyer changes nothing about the goods. A price attached to a merchant gives the party able to reduce the risk a reason to do so.
The general principle is that an insurer should price the risk it insures, using the smallest set of data that describes that risk. For a return-shipping policy, that is the transaction and the merchant, not the person. For an accident policy sold to farm-machinery operators, it is the machine, the season and the work, not the operator's browsing history. Whenever an insurer finds itself reaching for personal data beyond the insured event, it should ask whether it is pricing the risk or the customer.
A record for every small promise
The other discipline is to treat a one-yuan policy as an insurance contract rather than as a checkout feature. When I arrived at Yunzhu in 2017, some partners held the only copy of the policy record. If a customer asked us what they had bought, we sometimes had to ask the platform. We changed that. Every policy, however small, received a stored wording version, a record the customer could retrieve without the partner's help and a claims route that worked without a telephone call. We automated 92 per cent of return-shipping claims from carrier tracking data, so that the policy paid without the customer having to ask.
At Hesui our online book is small, about RMB 600 million of premium in 2025, mostly accident cover for farm-machinery operators, produce-in-transit cover and return-shipping cover sold through two platforms. It follows the same rules: no data beyond what the insured event requires, prices that can be explained by the policy's own characteristics, and a complete record held by us. Under the Data Security Law we classify each data set we receive from partners, and every year we review partner feeds and delete fields we no longer use.
Some colleagues across the industry regard this as an expensive scruple. I think it is a durable position. Platforms change their terms, regulators sharpen their expectations, and customers increasingly ask why they were charged what they were charged. An insurer that knows less than the platform knows, and can explain everything it does know, will be able to answer all three.