The Census Bureau published advance retail sales for September at 8:30 this morning. Total retail and food services sales were $745.0 billion, up 0.8 percent from August and up 6.3 percent from September 2025. Nonstore retailers, the category that carries most of ecommerce, came in at $124.3 billion seasonally adjusted. That is up 0.3 percent on the month and up 8.2 percent on the year. (Census Bureau release CB26-165, October 15, 2026)
Eight percent annual growth six weeks before Black Friday is a good number. Check the dollar figure against your notes before you use it anywhere, because it is not the figure this series carried a month ago.
The whole history was rewritten on September 28
Census republished the monthly retail series on September 28. Not adjusted estimates were revised back to January 2018. Seasonally adjusted estimates were revised back to January 2015. Quarterly ecommerce sales were revised from the first quarter of 2018 through the second quarter of 2026. End-of-month inventory estimates moved as well. The work benchmarks the monthly survey to the 2024 Annual Integrated Economic Survey and swaps previously reported data for corrected data. (Census, 2026 annual revision, explanation of revisions)
August is the most recent month published on both bases. Here is what happened to it.
| August 2026, seasonally adjusted | As published Sept 16 | As published now |
|---|---|---|
| Total retail and food services | $773.9 billion | $739.3 billion |
| Nonstore retailers | $141.3 billion | $123.9 billion |
Nonstore sales lost $17.4 billion of one month, 12.3 percent of the line. Total retail lost $34.6 billion, 4.5 percent. (CB26-153, September 16, 2026, against the current series published through FRED)
Year to date makes the point harder. Through August, not adjusted nonstore sales were published at $1.057 trillion, up 10.3 percent against 2025. Through September they read $1.046 trillion, up 8.0 percent. A month of sales got added and the cumulative total went down.
We quoted the old number here. Our September 16 post put August online sales at $141.3 billion. On the current series that month is $123.9 billion. The post was right about the trend and wrong about the dollars. Watch for that in your own spreadsheet.
Growth rates held up better than levels
The year-ago months were cut too, which is why percent changes moved less. August 2025 nonstore sales, not adjusted, went from $121.7 billion to $108.9 billion. The 2026 comparison and the 2025 base both came down by roughly the same share.
So the percent changes in today's release are usable. The dollar levels in any release you saved before September 28 are not. Growth did get trimmed, from 10.3 percent year to date down to 8.0 percent, so a forecast built on double-digit online growth is now running about two points hot.
Rebase your numbers before you touch Q4 staffing
This takes about ten minutes.
- Delete any saved Census dollar figure from a spreadsheet or a deck. Pull the series fresh if you need a level.
- Rerun any comparison where you divided your own revenue by a Census total. Online share of all retail sales is 16.7 percent this September against 16.4 percent last September, on the revised series. If your deck claims a market share, it moved.
- Rebuild the Q4 volume forecast off growth rates and your own order history. Your orders were never revised.
- Stress it both ways. Ask what breaks at 30 percent above plan.
What this means for your brand
The number that should change your October decisions is 0.3 percent, not 8.2 percent. September was close to flat against August. That was before October Prime Day on the 6th and 7th, which lands in next month's data on November 17. Today's release also comes from a subsample of roughly 4,800 firms, so the September figure will move again.
Here is the position worth taking. Plan Q4 parcel volume off the year-over-year growth rate, around 8 percent for online, applied to your own order history. Then buy capacity for being wrong by a third in either direction. Underbuilding costs you customers in December. Overbuilding costs you money in February, which is the cheaper mistake.
That asymmetry is the argument for capacity you do not have to own. Our ecommerce fulfillment absorbs the swing, with orders processed within 24 hours whether the day brings 300 or 3,000. No seasonal hiring in your name, and no lease on space you sit on after New Year. If inventory is the constraint, warehouse storage scales by the pallet, not the building.
Canal Winchester does part of the work on cost. Sixty percent of the US and Canadian population sits within 600 miles, and 90 percent of customers get their order in two days by ground. Short zones cost less per parcel and hold up better when carrier networks tighten in December.
If you are not sure your current setup survives a November that runs a third above plan, send us your last 12 months of order data and we will show you where it breaks.
FAQ
Did online sales actually shrink, or did the measurement change?
The measurement changed. Nothing retroactively unsold itself in August. Census benchmarked the monthly survey to the 2024 Annual Integrated Economic Survey, corrected previously reported data, relinked the old survey sample to the current one, and updated its seasonal adjustment models. Both 2025 and 2026 came down, so the growth story is roughly intact while the levels are much lower.
Which figure belongs in my Q4 forecast?
A growth rate. Take nonstore growth of 8.2 percent year over year as the category rate, then apply your own channel mix and promotional calendar on top. Pull a Census level only when you need market size for a deck, and pull it fresh. Our FAQ covers how we plan capacity with brands heading into peak.