When the Montreal-based payments processor Terra Payments began about four years ago as SureFire Commerce, a service provider was used for their IT application. Senior vice-president and CTO Danny Chazonoff was unsatisfied with the service and in late 1999 decided to build all their IT capabilities in-house. Despite many contrary opinions, Chazonoff persevered with them building their own data centre and eventually even their own automated system to predict the fraud risk of every transaction in real time. In October 2000, they processed their first transaction with own technology. Three years later, this past fall they processed their 50 millionth transaction on a sophisticated version of that home-grown platform.
Chazonoff explains that the payments his company processes are high-risk in the sense that the merchant and the consumer are not physically located in the same place and the merchant doesn’t see the customer’s face, card or signature. This includes Internet sales and mail order telephone order (MOTO).
The company facilitates the payments for people who respond to a TV infomercial’s urging to call a 1-800 phone number for steak knives for three easy payments of $19.99, for example. “The merchant will take the information over the phone and then through our infrastructure process get either authorization or decline information about the transaction with the consumer,” says Chazonoff. The company ensures that millions of international payment transactions a month are not fraudulent.
It does so with an automated risk management system they created themselves. The system also performs a reconciliation so merchants are paid on time and accurately. It also provides reporting information to merchants and partners. Further, its reporting capabilities reveal Terra Payments’ exposure to determine the company’s security clauses and rolling reserve to offset charge backs.
The good news when Terra Payments successfully built their own data centre and developed their own risk management technology was that they had full control to ensure the highest level of support. The bad news was that doing so required a lot of people. The company ballooned to about 300 people very quickly and was dangerously close to becoming too heavy for the revenue. Once most of the infrastructure was in place with the data centre and the basic techno-logy, Terra Payments launched a massive four-phase IT restructuring that reduced the staff count by two thirds.
“I was running a race and the race was: can I deliver the applications that are going to save our business faster than the people are let go?” Chazonoff recalls.
Fortunately, he won the race – and earned the company a Canadian Information Productivity Award for efficiency and operational improvements, presented last November. More importantly, since April 2003, the company has been showing a profit.
Among the few thousand merchant customers are online pharmacies, Dreammates online dating company and CrazyApe online seller of DVDs, CDs and video games. Terra Payments is also the exclusive payment processor for payroll processing, online billing and payments with Intuit Canada’s QuickBooks and Quicken in Canada and the UK. The company’s technology office is near Hull, Que. It has sales offices in the U.K. and the U.S.
The 52-member team of fraud investigators and analysts has been cut down to 11 people who reportedly manage more transactions at a higher level of quality.
Terra Payments created this turnaround by automating manual processes. They began by improving the technology to process transactions over the Internet. “The old technology took 20 to 30 seconds to get the yes or no authorization for the transaction to clear,” Chazanoff recalls. “Now it has been cut to two and half to three seconds and includes some basic risk management functionality.”
Next they enhanced their engine and application to accept online cheques which is a popular payment option among U.S. consumers and a likely growth area for the company.
They also automated reconciling merchants’ accounts. And, they boosted their risk management techniques to try to minimize fraud processed through the system.
They will assign a rule to run with a certain frequency – every 15 minutes, every hour or in some cases only once a day. They begin with it in test mode and watch how effective it is at detecting fraud with minimal false positives before modifying it for production mode. For example, a rule might run through all the databases to find relationships between bad transactions and transactions that otherwise look good. “It’s sort of a tumbling array technique that we use to make all these associations and say ‘hey, you know what? Here’s a fraudster coming in using many different aliases, using many different credit cards and email addresses but we’re going to catch him every time. If we don’t catch him in real time, then we’ll catch him shortly thereafter. That’s why we run it every 15 minutes or half hour or every hour.”
Terra Payments authorizes transactions in real-time but only settles transactions at the end of the day. About 60 people in the risk management group manually looked at all the transactions processed to assess if they were good or not. “We realized this was not scalable and as we were growing in number of transactions per merchant, it didn’t make sense to hire more people to do this mundane, not very accurate method to catch fraud.” Further, bad transactions that came in on weekends or during morning hours were checked too late to be stopped when staff returned.
Now, the company’s risk management group numbers only 11 and two or three of those people are focused on assessing the risk of potential merchants defrauding a consumer.
“The technology helps us differentiate ourselves from other payment processors,” Chazonoff stresses. “We believe that the technology is key in trying to be faster, more accurate and less dependent on people.”
Off duty data centre staff rely on Computer Associates’ Unicentre TNG for remote managing. Chazonoff says they are also looking at CA’s Neugent technology to further enhance their risk management applications.
On Jan. 20, Terra Payments and Montreal-based Optimal Robotics Corp. announced a merger of the two companies, scheduled to take effect next April. Optimal Robotics is a North American provider of self-checkout systems to retailers and, through Optimal Services Group, provides depot and field services to retail, financial services and other third-party accounts. As a wholly owned subsidiary, the Terra Payments name will change to Optimal Payments. The management will stay in place and, one can safely bet, so will the winning technology strategy.
Terra Payments’ risk management involves a four-fold approach
Fraud management — A huge database contains hundreds of thousands of credit card and e-mail addresses that they have processed and which subsequently initiated a charge back, i.e., the consumer calls their credit card company and asks to have that charge reversed. If he or she tries to order again, the order gets blocked and the transaction isn’t processed.
Transaction scoring — Built with the help of California-based Fair Isaac Corp., a neural network looks for trends and patterns and makes obvious associations. All transactions are fed to this neural network, including those that have come back with a charge back. The neural network tries to make associations, working on what the charge backs have in common – maybe the time, the amount, the frequency – to come up with an estimate of how likely a current transaction is that it will go bad. Chazonoff swears it takes literally one second to give a score which ranges between zero and 999. “Depending on the score, we decide whether we’re going to accept or decline that transaction. Obviously the more we grow our databases, the more accurate and fine-tuned this neural network becomes.”
Merchant rating — Terra Payments adjusts that neural score according to a merchant’s low or high charge back history. “For a good merchant giving you a lot of volume and with a low charge back level, you are going to be a lot less strict on the scores. As merchants become better at managing charge backs, Terra will be less aggressive with respect to ratings and scores assigned to the particular merchant.”
Rules, rules, rules — A decision rules engine enables them to create rules on the fly as needed. “If there’s a lot of fraud coming from Taiwan, for example, we can put in a rule to decline a transaction greater than $50 and the country of origin is Taiwan. Transactions over $1,000 and between 12 midnight and 5 a.m. have a greater likelihood of being fraud, so we put in a special rule in production. We may see that in the case of only one merchant there is a high fraud rate between 3 a..m. and 4 a.m., so we will put a specific rule on that merchant at that time frame.”
