An order system went live at the worst possible week

The project before the failure

By the late 1990s Hershey Foods was running on a patchwork of legacy order-management and distribution systems it had accumulated over decades. The company decided to consolidate onto a modern enterprise platform — a decision consistent with what much of American manufacturing was doing under the pressure of Y2K remediation budgets and the broader ERP wave that SAP, Oracle and their implementation partners had been riding since the early part of the decade.

The programme Hershey assembled was substantial. It combined SAP's R/3 for enterprise resource planning with Manugistics for demand-planning and Siebel Systems for customer management — three separate enterprise packages that had to be integrated with each other and with Hershey's warehouse and logistics operations. The total investment was reported at around $112 million (1999 figures). That kind of budget was not unusual for a Fortune 500 manufacturer, and neither was the ambition: Hershey wanted a single, coherent view of inventory, orders and fulfilment across its North American business.

What was unusual was the timeline. Hershey's original plan had allowed roughly four years for the implementation. Somewhere in the middle of the project that window compressed to about two and a half years, driven in part by the desire to have the system stable before any Y2K disruption in January 2000. That compression is important because it is where the decision calculus changed. The team could hit the go-live date or it could conduct the kind of parallel-running, phased cutover and user acceptance testing that a system of this complexity required. It could not easily do both.

The go-live was set for July 1999.

What happened when the orders came in

July is not a neutral month in the confectionery business. Halloween is the single largest seasonal event in the American candy calendar — larger than Easter, larger than Valentine's Day by volume — and the retail supply chain for it begins filling in late summer. Retailers and wholesalers place their Halloween orders in July and August, expecting fulfilment through September so that product is on the shelf by October.

When Hershey's new systems went live, the order-processing and warehouse-management functions did not behave as designed. Orders that were received could not be fulfilled correctly: the system had difficulty processing the volume of incoming orders, shipments were delayed, and in some cases product that existed in finished-goods inventory was not being dispatched because the fulfilment logic could not work through the queue. The failure was not a single dramatic crash. It was a steady, grinding inability to translate demand into shipment at the rate the business required.

The consequences appeared first in Hershey's own reporting. The company disclosed that it was unable to fulfil orders on time, and the effect surfaced in the financial results for the third quarter of 1999: Hershey reported net sales significantly below what it had projected, and it attributed the shortfall directly to the ERP implementation problems. The retail impact was real and visible — shelf gaps at a moment when the company's products were in peak demand. Retail buyers who couldn't get reliable shipment dates began to source from competitors. Market share, once lost at Halloween, does not automatically return the following year.

The scale of the documented shortfall was striking. Hershey's third-quarter 1999 results showed net sales had dropped roughly 12 percent from the comparable prior-year period — an extraordinary figure for a consumer-goods company with established distribution. Analysts at the time estimated that unfulfilled orders accounted for a loss in the region of $150 million in sales during the peak season.

What the failure established

Hershey recovered. The systems were stabilised over the following quarters, and the company's underlying brands and market position were strong enough that it did not suffer lasting structural damage. That is an important distinction from cases like the FoxMeyer collapse, where the ERP implementation failure was entangled with the company's bankruptcy. Hershey's failure was costly and visible, but the company survived it, and by 2000 the systems were functioning.

What the Hershey case established, in the literature of ERP implementations, was a set of lessons that project managers and their clients still cite. First: go-live timing matters independently of technical readiness. A system that might have been nursed through its early instabilities in a quiet February became a crisis in July because there was no slack in the external calendar. The business cycle does not pause for software.

Second: scope compression is a risk that compounds. Reducing the implementation timeline from four years to two and a half years did not halve the amount of work — it moved the untested complexity forward into the live environment. Integration between three separate enterprise packages — SAP R/3, Manugistics, Siebel — was always the hardest part of the technical programme. That integration did not get simpler because the schedule shortened.

Hershey's original plan had allowed roughly four years for the implementation.

Third: the decision to avoid a phased cutover was consequential. A phased approach, in which specific geographies or product lines go live while others continue on legacy systems, creates a fallback. Hershey ran a big-bang cutover: everything switched at once, which meant there was no fallback when the order volume arrived. This was not an unusual choice for the period — phased cutovers are expensive and operationally complicated, and many programmes of this era made the same call — but the combination of big-bang cutover, compressed timeline and peak-season timing amplified every instability the system carried into production.

The case also illustrated something about where ERP risk is actually concentrated. The packages themselves — SAP's R/3, Manugistics, Siebel — were mature commercial software used successfully by other large manufacturers. The failure was not a product defect in any single package. It was a failure of integration, timing and testing volume, which is precisely what makes ERP risk hard to manage through vendor selection alone. Choosing better software would not have resolved a schedule compressed by a year and a half and a go-live date that sat at the edge of the highest-demand quarter.

Hershey's story is often taught alongside the Lidl and FoxMeyer cases as one of the canonical ERP failures on the record, and the comparison is instructive. FoxMeyer went bankrupt; Lidl wrote off years of work; Hershey saw third-quarter sales fall roughly 12 percent at Halloween. The common thread is not malice or incompetence in any simple sense — it is that very large software programmes are systems with their own dynamics, and when those dynamics collide with external pressure, the external pressure tends to win. The shelves were empty because candy was in the warehouse and the software could not find it. That sentence is what the project left behind.