Inference ID

By Kevin Ashton

When it comes to supply chain visibility, a little RFID can go a long way.


I recently met with a logistics company—let’s call it Acme—that’s considering using RFID to track shipments in its supply chain. The company’s execs explained they were committed to “six nines”—or 99.9999 percent—accuracy. If they were going to track shipments in their supply chain, they needed to know where everything was, all the time. To achieve supply-chain visibility with RFID would be expensive, if it were even possible. Some shipments would be difficult to tag with low-cost, passive RFID, and shipments sometimes traveled to places where it would be difficult to install RFID readers.

The discussion that followed was one I have had many times before. Why, I asked, were there only two choices—tag everything or nothing at all? This “do all or do nothing” approach is common in early RFID planning, and it nearly always reaches the same conclusion: Do nothing. It’s seldom practical to tag everything, and it makes little business sense to adopt RFID if there’s no value unless everything is tagged.

But if the goal is visibility, there is another approach. I suggested to Acme that they consider how the human eye operates. While extraordinarily useful, it turns out the human eye is far from 100 percent reliable as a data-capture device. For example, it can’t see through things, and it doesn’t perform well in poor light. Yet none of this matters very much, because the human brain, the computational system that processes the imperfect data gathered by the eye, uses a sophisticated system of inference—basically, educated guessing—to fill in all the gaps. If an object passes behind something, the brain expects it to appear on the other side. If an object moves into an area of shadow, the brain makes adjustments—it doesn’t assume the object has suddenly changed color. If one like object appears smaller than another, the brain infers it is farther away. And so on.

Acme and other end users considering RFID can adopt the same strategy. Instead of waiting for perfect data from a perfect data-capture device, they could tag a representative sample of shipments and use inference to fill in the gaps. It takes surprisingly few tagged shipments to paint a valuable picture of supply-chain dynamics; a small percentage of tagged objects, randomly distributed across the supply chain, can yield valuable information about supply-chain velocity, and identify bottlenecks and common exceptions.

This is not a new approach—it’s standard inferential statistics. Opinion polls don’t ask everybody a question—they take a representative sample to infer the views of an entire population. Marine biologists don’t tag every fish in the sea—they track only a few.

If Acme RFID-tagged a sample of its shipments, that would allow inference about a whole range of things. The accuracy wouldn’t be 100 percent, and some details about individual events would be missed. But the most important thing about visibility—the big picture—would be captured, and at relatively little cost.

Kevin Ashton was cofounder and executive director of the Auto-ID Center.