Some Lessons on Fail-Proofing Tax Automation Platforms
Over the course of Thanksgiving and Black Friday, Americans spent more than $4.45 billion online – the 25 percent year-over-year increase was good news for ecommerce retailers yet bad news for some of the IT systems they depend on.
As in past years, merchants encountered some technical difficulties including a connectivity issue with a highly reputable tax automation platform.
At a high level, failure events such as this illustrate that tax software is a complex and delicate engine with many parts that must work together. Tax automation software must make accurate sales tax calculations based on data that is partial and often imprecise.
First, the system identifies the exact location of the buyer. To accomplish this, tax platforms use an Address Verification Service (AVS). Recognizing the customers may enter the same address in many different ways, AVSs automatically clean up and standardize the address. That way, the system can pin the buyer to one of roughly 10,000 different U.S. tax jurisdictions and determine what rates and rules apply. Systems that rely only on a zip code or zip+4 cannot make this determination with consistent reliability.
Next, the system must determine the contents of each transaction by parsing the sales record. What is each item? What does it cost? How is that item taxed in this particular jurisdiction? This process is tricky because the system must be able to distinguish line items, the shipping amount, the pre-tax total, price per unit, service and maintenance fees, etc., and make sense of all the information. The tax platform must also take into account tax thresholds, holidays and other factors that can change the rate. This all has to be completed in milliseconds.
You can think of the tax calculation process as a chain reaction. If any link in the chain fails or produces incorrect information, the subsequent links will also fail at their jobs. If the AVS is unresponsive, for example, the tax engine will wait and wait, and sales tax will never be calculated. If the system misinterprets the line items and their value, it will calculate the wrong tax. Each step must be flawless. Failure can either bring the transaction to a halt or cause the collection of incorrect taxes.
The risk of breakdowns often depend on how a tax platform is architected. For instance, some platforms license or buy multiple technologies and hitch them together like the reindeer on Santa’s sleigh. The integrations are initially robust, but software updates and other changes to the system can unintentionally degrade these connections. To run with the Santa metaphor, if Dasher’s harness rips, he’ll collide with the other eight reindeer and stop the whole sleigh. That means no presents (online purchases) for anyone.
Conversely, some tax automation companies try to eliminate integrations and build everything in-house. From an entrepreneurship perspective, this can be a significant burden but it means there will be less potential failure points and therefore less things that can go wrong. Think about a set of scuba diving gear as an analogy. Would you rather have one solid rubber tube connecting your regulator (breathing piece) to your air tank? Or, would you like to stake your life on a tube that consists of 10 separate sections of rubber glued together?
Here’s the universal dilemma facing sales tax software companies: do we combine multiple components, or do we build a single, unified system? The integration approach will get a solution up and running quicker and less expensively, but as the system scales to greater traffic and adapts to new code, the integrations will become the weakest points. The unified system will take longer to develop, but it will scale and perform better and more accurately in the long run.
The lesson to take home is that ostensibly simple, useful technology tends to have a ton of complexity under the hood. To reduce the likelihood of breakdowns and inaccurate calculations, merchants should look for tax systems that are unified in their design and tested both for accuracy and scalability.