Turns out that inventory tracking is much harder than you would think. Last year, a report in DC Velocity surveyed 177 executives at retailers, and found:
Today, cycle count accuracy levels at DCs that use warehouse management software in conjunction with automatic identification technology exceed 99.9 percent. Accuracy at the stores, however, appears to be falling far short of that mark. Only 30 percent of respondents reported that their store inventory accuracy level was 98 percent or higher. Another 32 percent said store inventory accuracy rates fell between 95 and 97.9 percent, while 15 percent characterized their accuracy rates as between 90 and 94.9 percent. At the low end of the spectrum, 17 percent said it was below 90 percent and, surprisingly, 6 percent did not measure inventory accuracy at the store.The report goes on to say that integrating accurate point-of-sale data should help fix this, but I'm not convinced that alone will do the job. Inventory inaccuracy and shrink at the retail level has multiple causes (e.g. customers picking up items and then putting them down elsewhere after changing their minds), not all of which can be captured by POS.
For those with interest in probability, we can take this calculation a little further. Suppose inventory accuracy is 96.5% (the median from the survey above), and only 20% of the inventory inaccuracy errors actually are related to stockouts. Suppose also that each customer orders 25 items per order (this is more common in the grocery sector; obviously for apparel it would be much smaller). Then the likelihood of an error within any order is approximately 16%. The probability of a bad order in a day, with 5 ecommerce customers per day is 58%. Changing any of the numbers above even a little bit easily bumps up the probability of a bad order in a day to well over 90%. Omnichannel is challenging!