What CPFR is
CPFR, collaborative planning, forecasting and replenishment, is a framework in which a retailer and a supplier agree on one shared demand forecast for a defined set of items, commit to a joint business plan behind it, and then work an exception list when reality moves away from the plan. It is not software and it is not a data feed. It is a written agreement about whose number wins and what happens when the two sides disagree.
I ran the supplier half of one of these for three years, feeding a weekly Walmart Retail Link pull and a KeHE Connect extract into the same spreadsheet the retailer's replenishment analyst was looking at. The framework worked exactly as far as the exception meeting kept happening, and no further.
Where CPFR came from
The framework was organized by the Voluntary Interindustry Commerce Standards (VICS) Association, which ran the original pilot with Walmart and Warner-Lambert in 1995. That pilot is the number everyone still quotes: a 30% reduction in inventory and a 3% increase in in-stock performance, according to SPS Commerce's write-up of the model. VICS published a nine-step process model as the implementation guideline, and VICS itself merged into GS1 US in 2012, which is why the standards lineage now runs through GS1 rather than a standalone body.
Thirty years on, the nine steps are rarely run as nine steps. What survived is the shape.
The four pieces that actually carry a CPFR program
The front-end agreement. Which items, which stores or DCs, which horizon, what counts as an exception, and who owns resolving one. If this document does not name a percentage threshold and a person, the program is a standing meeting rather than a process.
The joint business plan. The promotional calendar, the item changes, the distribution moves and the seasonal shifts, agreed in advance. This is the piece that supplies the information a statistical forecast cannot derive from history, and it is the piece with the most commercial sensitivity, which is why it is the piece that gets thinned out first.
The shared forecast. One number, not two reconciled numbers. In practice most programs run two forecasts and a variance report, which is a useful thing but is not CPFR.
The exception-resolution loop. A weekly comparison of forecast against actual sell-through, filtered to the items that broke the threshold, with a named owner and a decision. The whole framework earns its cost here or nowhere.
Exception resolution, with the numbers
Northbay Provisions is an illustrative brand selling a 12 oz salsa into 640 stores of a regional grocery chain. The front-end agreement sets the exception threshold at plus or minus 15% against the retailer's weekly forecast, measured in cases of sell-through.
| Week | Retailer forecast | Supplier forecast | Actual POS | Variance vs retailer | Exception |
|---|---|---|---|---|---|
| 1 | 1,850 | 1,900 | 1,790 | -3.2% | no |
| 2 | 1,880 | 1,900 | 1,845 | -1.9% | no |
| 3 | 1,900 | 2,600 | 2,540 | +33.7% | yes |
| 4 | 1,920 | 2,600 | 2,610 | +35.9% | yes |
| 5 | 1,950 | 1,950 | 1,505 | -22.8% | yes |
| 6 | 1,960 | 1,960 | 1,610 | -17.9% | yes |
Actual sell-through across the six weeks is 11,900 cases. Weight the absolute errors by volume and the retailer's forecast misses by 18.7%, while the supplier's misses by 8.7%. The whole ten-point gap sits in weeks 3 and 4, because the supplier funded a feature ad and the retailer's statistical model had no way to see it. That is the single clearest argument for CPFR that exists: the information was in the building, on the other side of a wall.
Weeks 5 and 6 are the more interesting rows. Both sides miss by the same amount, 445 and 350 cases, because neither forecast carried the post-promotion dip. Nobody owns the dip, so 795 cases of salsa shipped into DCs a week before there was demand for them. A shared forecast fixes the promotional peak on the first cycle. The trough takes a policy, and most programs never write one.
If you want the arithmetic behind weighted error rather than the plain average, measuring forecast accuracy walks through WMAPE and why a simple MAPE flatters a low-volume week.
Why most CPFR programs stall
The joint business plan gets thinner every quarter. Promotional calendars are commercially sensitive and staffing turns over. Two years in, the shared plan is a list of dates with no depth, and the shared forecast quietly reverts to two forecasts.
Exceptions arrive without an owner. A threshold that flags 40 items a week and routes them to nobody generates a report, not a resolution. The programs that hold up flag five items and name a person.
The two sides measure different things. The retailer forecasts store demand; the supplier forecasts shipments. When one is talking sell-through and the other is talking sell-in, the variance report is partly an artifact of the unit, which is the same confusion consumption versus shipment data describes.
Nobody funds the boring half. CPFR is a weekly meeting with an agenda. It survives when someone's job description contains it, and it stops when that person moves.
Where Scout fits
Scout is the shared demand-signal layer underneath a program like this, not the machinery that runs it. It ingests the retailer POS and distributor movement feeds, puts sell-through, distribution and promoted lift into one weekly number both sides can look at, and flags where actuals broke the plan. It is not an S&OP or ERP system, not an EDI gateway, and not a replenishment engine. It does not raise or transmit a purchase order and it does not carry a supplier EDI connection. The plan, the order and the commitment stay in your planning systems and in the front-end agreement.
The short version
- CPFR is a retailer-supplier framework built on a front-end agreement, a joint business plan, one shared forecast, and a weekly exception-resolution loop.
- It came out of the VICS Association's 1995 Walmart and Warner-Lambert pilot, which reported a 30% inventory reduction and 3% better in-stock. VICS merged into GS1 US in 2012.
- The value is concentrated in promotional weeks: in the worked example the supplier's number beat the retailer's by ten points of weighted error, entirely because it carried a feature ad the retailer's model could not see.
- Programs stall when the joint business plan thins out, exceptions have no named owner, or the two sides are forecasting sell-through and sell-in and calling it the same number.