A forecast rarely breaks in the model. It breaks before the model, when the numbers going in disagree.
If you run wholesale operations or sales for a fashion, retail-facing, or golf brand, you know the meeting. Sales brings the pre-book total. Ops brings what the ERP says is committed. Someone has a spreadsheet of reorders taken by phone, and someone else has EDI orders that never touched the portal. That's four versions of demand, one forecast, and an hour spent deciding which one is right.
This post covers why that happens and what to fix first. We couldn't find independent research on forecasting in wholesale specifically, including pre-book, at-once, and reorder, so the evidence below comes from general planning, spreadsheet, and supply chain research. Where a point is our own view, we say so. RepSpark is a vendor, and this is written from our seat.
For most brands, demand is recorded in several places at once: the ERP, a planning spreadsheet, the wholesale portal, rep notes and email, phone orders, EDI, marketplaces, and the sell-through reports retailers send back (when they send them). Each one records something different at a different time. None of them is wrong. They just aren't measuring the same thing.
That's the root of the problem. A forecast can only be as good as the demand signal it starts from, and a fragmented signal doesn't average out. It distorts.
Three bodies of research, none of them wholesale-specific, point the same way.
Planning tools struggle on fragmented data. McKinsey's April 2026 analysis of advanced planning and scheduling (APS) deployments, which draws on a survey of 80 companies fielded in November 2025, says about 65% of APS programs miss their expected ROI and names poor data management as one of the top five reasons. It adds that "even the most sophisticated planning engine will struggle" when master and transactional data are fragmented, incomplete, or inconsistently owned. McKinsey sells transformation work in this area, and the finding covers planning software projects broadly, not wholesale. Read the McKinsey article.
Spreadsheets carry errors. Raymond Panko's review of field audits, published through the European Spreadsheet Risks Interest Group, found errors in 94% of 85 audited spreadsheets. The audits were small, non-random studies. A separate audit of 50 spreadsheets by Powell, Baker, and Lawson in Information & Management found 47 had at least one error, with errors in 1.79% of 270,722 formula cells. These are operational spreadsheets in general, not forecast models, so treat them as a caution, not a measurement of yours. See Panko (2015).
Order data distorts demand. In a classic 1997 paper in Management Science, Lee, Padmanabhan, and Whang showed that the variance of orders can be larger than the variance of sales, and that the distortion tends to grow as you move upstream. A manufacturer who only sees order data "will be misled." One cause they identify is rationing and shortage gaming: when buyers expect a shortage, they order more than they need. A wholesale brand sees retailer orders, not retailer sales. This is the same position.
The five below are our framing, drawn from how wholesale teams describe their planning, not a published taxonomy.
Pre-book is your earliest read on a season. Fisher and Raman's 1996 study in Operations Research, a model built around one fashion skiwear company, found that responding to early sales improved profits by 60% over the firm's informal approach. It's an old, single-company result, but the logic holds: early orders carry information.
The catch is that a booked order isn't a shipped order. Pre-book orders get cut, moved, or cancelled as the season changes. If sales counts bookings and ops counts shipments, neither number is the forecast input. Track pre-book from booked to shipped, by account, for last season. The gap is your real signal.
Full Turn Custom Apparel's case study describes this shift: factory POs placed on actual pre-book demand plus a buffer, instead of against historical sales alone. If you're sorting out how pre-book and at-once inventory interact, see our post on managing at-once vs. pre-book inventory.
Sales records show what sold. They don't show what buyers tried to buy when the system said zero. Demand that was oversold, short-shipped, or never placed because stock showed unavailable drops out of the data.
Planners call the full number unconstrained demand and the recorded number constrained demand. If last season's best seller stocked out in week six, the sales history understates its demand, and a forecast built on it will under-buy again. This is our view of how to read the data; we found no independent measure of how much demand disappears this way in wholesale. The practical step is to flag stockout days by SKU, and treat sales on and after those days as a floor, not a total. Our overselling prevention guide covers how available-to-sell is calculated.
Phone orders, emailed orders, rep notes, marketplace orders, and EDI often land somewhere other than the main ordering flow. In the Full Turn case study, some EDI orders don't originate in RepSpark but still flow into the brand's ERP. That's a normal setup. It also means any forecast built from portal orders alone is missing a slice of demand.
The fix isn't to force every order through one door. It's to tag every order by source, so the forecast knows what's in it and what's missing. We cover how mistagged off-platform orders skew reorder metrics in What's Really Behind Declining Reorder Rates in Wholesale Apparel.
The ERP updates on one schedule. The portal is live. The planning spreadsheet is whatever someone pasted in on Tuesday. When the team opens all three, they're looking at three different moments and calling it one picture.
Compared with a manual flat-file process, an ERP sync updates inventory in seconds rather than a daily batch, and it takes manual touchpoints from two or three per order down to zero. If you want to know which of your connections are real, Is Your Wholesale Data Actually Talking to Your ERP? A Quick Audit walks through it.
"Booked" in the CRM, "committed" in the ERP, and "demand" in the planning spreadsheet usually mean different things. Sales counts a pre-book order the day the buyer submits it. Ops may not count it until it's allocated. Both are right for their purpose, and neither matches the other's total.
Write one page that defines each term: what counts, when it counts, and which system owns it. It costs an afternoon and removes more arguments than any dashboard.
In fashion and retail, the signal is noisy and the conditions keep shifting. The 2026 BoF-McKinsey State of Fashion report calls tariffs the number-one hurdle and found 46% of executives expect conditions to worsen, up from 39% the year before. When the environment moves that much, a forecast built once at the start of the season goes stale. At the retail level, IHL Group estimates inventory distortion, which is the combined cost of overstock and out-of-stock, at about $1.7 trillion a year, or 6.2% of global retail sales in 2026. IHL sells that research, and it covers retail broadly, not wholesale. It does show how costly a mismatch between inventory and demand gets.
In golf, the channels don't move together, so blended history misleads. Golf Datatech reported that in 2018 on-course apparel dollar sales fell 9% while off-course rose 6%. That data is old, but it's the clearest public example we found. More recently, Circana reported in October 2025 that golf specialty retail dollar sales were up 7% through August, with growth shared across on-course and off-course channels. A brand that sells into pro shops, resorts, and off-course specialty shops can see each channel reorder differently. Events add another layer: in 2025, RepSpark Event Microsites generated $13.9 million across 1,582 sites, an average of $8.8K per site. If event demand isn't tagged separately, a forecast built from regular pre-book and reorder history can't see it. That last point is our inference.
For golf-specific forecasting, see How to Use Wholesale Data to Forecast Golf Product Demand.
Before you build a better model, reconcile the inputs. The table maps where demand is recorded and what each source misses. Fill in the last column for your own team.
| Source | What it records | What it misses | Who owns it |
|---|---|---|---|
| ERP | Committed orders, inventory, shipments | Orders not yet entered, anything off-system | |
| Wholesale portal | Retailer-placed orders, live availability | EDI, phone, and rep-keyed orders elsewhere | |
| Planning spreadsheet | Whatever was last pasted in | Changes since the paste | |
| Rep notes and email | Intent, context, verbal commitments | Anything not entered | |
| EDI | National and large-account orders | Visibility in the portal | |
| Retailer sell-through | What sold at retail | Retailers that don't share it |
Then work through six steps, in order:
The last step is the one most teams skip. A classic study of 32 retailers, reported by Harvard Business School's Working Knowledge in 2000, found only nine analyzed their forecast accuracy. That was a long time ago and in retail, so don't over-read it. In RepSpark's own 2026 State of Wholesale survey of 26 lifestyle apparel brands, 52% forecast demand in manual spreadsheets and 24% don't forecast at all. That sample is small and the post calls it directional, so read it that way. The full numbers are in The B2B and ERP Integration Benchmark for Wholesale Brands.
Connecting the data won't forecast for you, and we don't claim it will. What it does is give your planning process one set of order, inventory, and account data to work from, so the argument in the meeting becomes "what do we do about it" instead of "which number is right."
That's the pattern in the results. L*Space's invalid orders fell from roughly 20% to near zero. Stance sales rep Sam Lovell says: "I save 260 hours per year by using RepSpark. I don't need to call customers back to provide them with tracking or proof of delivery." Peter Millar saw a 148% increase in reorder value as unassisted retail ordering overtook rep-entered orders. These are order-process results, not forecasting results, but clean, connected orders are the input any forecast depends on. For the wider fragmentation problem, see Why Enterprise Brands Struggle With a Fragmented B2B Wholesale Tech Stack.
RepSpark is built for lifestyle, golf, apparel, and outdoor wholesale. More than 250 brands run on it, including Titleist, 5.11 Tactical, Oakley, Ray-Ban, Peter Millar, and L*Space. RepSpark connects to 35+ integrations, including 20+ ERPs.
When a forecast misses, the instinct is to blame the model. More often the model did what it was told with inputs that disagreed. Reconcile where demand is recorded, define your terms once, flag what the data can't see, and then decide what's worth automating.
If you want to see how your order, inventory, and account data could line up in one place, schedule a demo.
Each source records demand differently: the ERP records committed orders, the portal records retailer-placed orders, spreadsheets record whatever was last pasted in, and EDI, phone, and email orders often sit outside the main flow. A forecast built on those inputs inherits their gaps and disagreements. A McKinsey analysis from April 2026 says even the most sophisticated planning engine will struggle when data is fragmented, incomplete, or inconsistently owned.
Typically pre-book orders, at-once orders, reorders, ERP inventory and shipment data, rep notes, EDI orders, marketplace orders, event orders, and retailer sell-through reports where retailers share them. Each has a different timing and coverage, so tag every order by source and define each term once.
Spreadsheets are manually updated and easy to get wrong. Raymond Panko's review of field audits found errors in 94% of 85 audited spreadsheets, and a 2008 audit of 50 spreadsheets found 47 had at least one error. Both cover operational spreadsheets in general, not forecast models, and the samples were small.
Constrained demand is what the sales record shows, which is limited by what was in stock and orderable. Unconstrained demand is what buyers would have ordered with unlimited stock. If a SKU stocks out, sales history understates its true demand. Flagging stockout days by SKU and treating sales on those days as a floor is one practical way to adjust.
Pre-book is an early and valuable signal. A 1996 study of one skiwear company found responding to early sales improved profits by 60% over an informal approach. But booked orders get cut, moved, or cancelled, so track pre-book from booked to shipped by account before relying on it, and add at-once and reorder data as the season runs.
Fashion demand swings with trends, tariffs, and a shifting economy: in the 2026 BoF-McKinsey State of Fashion report, 46% of executives expected conditions to worsen. In golf, channels can move in different directions: Golf Datatech reported on-course apparel down 9% and off-course up 6% in 2018, and Circana reported golf specialty retail up 7% through August 2025. Both reward forecasting by channel and account, not from blended history.
Compare the forecast to what was demanded, not only to what shipped, and break it out by account and SKU group. Many teams don't measure it at all: a 2000 study of 32 retailers found only nine analyzed their forecast accuracy, and in RepSpark's 2026 survey of 26 apparel brands, 24% said they don't forecast.
RepSpark gives your planning process connected order, inventory, and account data, with ERP integration across 35+ integrations including 20+ ERPs. RepSpark AI Order Insights flags expiring drafts, incomplete orders, and unusual order patterns from order activity. RepSpark doesn't publish a forecast accuracy figure or claim to replace your planning process.