• by David Atlas , June 28, 2006

Last week I talked about the problems with complaint systems as they exist in the world of e-mail marketing.  Complaint systems are extremely important to e-mail marketers; if complaint rates become too high, a loss of e-mail privileges can result, up to being completely banned from sending messages to certain receiving domains. 

Existing complaint rates have three primary problems. They are not accurate, often taking into account consumers who press the "This is Spam" button repeatedly, for example, as if each click were a separate complaint. There is a lack of granularity in the system, both in terms of complaints not being separated out by sending entity but lumped together based on IP address, as well as a lack of granularity in terms of options for consumers to report their dissatisfaction. Finally, there is a lack of security in the system.

What would an ideal complaint system look like? How would it work?

Sender-Based, Not IP-Based. Complaints could be based on senders--not their IP addresses. When a consumer decides to complain, the complaint needs to accrue to the reputation of the sender, the person who (in this consumer's mind) done him wrong, not the IP address that that sender happens to be using. Speeding tickets are issued to drivers, not the car the driver is operating. Complaint rates needs to work in the same way. Moving from scoring IP addresses to scoring domains is not enough; we need to be able to score the reputation of sending entities.

No Group Punishment. Indeed, we can take this one step further: There should be the ability to target complaints not just at the human senders behind an IP address, but we should be able to sub-group these groups of human senders in a meaningful way. Beyond just the ability to complain about BigBank, we want a system that lets consumers differentiate their complaints among all of BigBank's many senders--the S&L group, the credit card group, etc. When your teenage son starts accumulating speeding tickets, it affects his driving record, not the family's. It should work the same way with e-mailing reputations.

One Man, One Vote. Complaints need to be tied per individual, per message. There is no reason for the trigger-happy consumer who clicks "This is Spam" ten times to register ten complaints. They should be counted as one.

Normalizing Complaints. Taking this a bit further, it is possible to have a system that actually tracks how often people complain. While still keeping the complainant's identity anonymous, it would identify if complaints are coming from people who tend to complain constantly. If the system knew how often a complainant complained, we could actually produce a complaint rate, say, that excluded the top 10 percent most complaining customers and the bottom 10 percent of least-complaining customers. In the same way that the scores of many judged athletic contests start by throwing out the highest and lowest scores, doing this might further improve the real accuracy of complaint rates.

Granularizing Complaints. Perhaps one of the most frustrating things for consumers and e-mailers alike is the inability to differentiate among reports of spam, requests to unsubscribe from a given list, requests not to be contacted again at all by a certain sender, etc. A complaint system that afforded customers a pick-list of choices beyond "Report as Spam" would enable them to accurately categorize the nature of their dissatisfaction and specify their requested remedy. This would lead not only to more accurate complaint rates, but, ultimately, to better e-mail list hygiene, all of which would lead to better e-mail marketing results.

Secure and Trusted Systems. A complaint system that was cryptographically secured would be a trusted system. This trust flows bidirectionally. Consumers need to trust that when they complain, they are doing so to an authorized third party who will deal with the offending sender appropriately, and that the mechanism of complaint itself is not a trick to confirm the validity of their e-mail address. Similarly, senders need to make sure that the system cannot be gamed, and that complaints are legitimate.

Differentiate Among Different Types of Messages. Another type of granularity that would be helpful would be to differentiate among different types of messages. Knowing that a given sender has a complaint rate of x for marketing messages but y for transaction messages would provide a much truer (and more efficient) system for affording or suspending e-mailing privileges.

Real Numbers. Complaint rates need to be tied accurately to the actual number of e-mails sent, per domain. By having a system of complaint rates that uses the actual number of e-mails sent in the denominator, the reputation data will be both more consistent and more accurate.

A complaint rate that remedies the problems of granularity, accuracy, and security is absolutely possible. In fact, done right, a reputation system based on this kind of complaint mechanism would result in overall greater feedback reliability and specificity for marketers. The results would be not only more accurate reputation data, but, ultimately, more equitable allocation of e-mailing privileges.

Benefits would flow back to consumers as well. With better lists come a higher overall percentage of relevant e-mail to recipients. The experience of scanning and responding to inbound messages from volume senders would over time prove more rewarding. And knowing someone was really listening to their complaints would actually result in more engaged consumers. More engaged customers, more equitable allocation of e-mailing privileges, greater reliability and specificity of consumer feedback to volume senders: complaint systems, done right, have the potential to benefit everyone.