Kimberly-Clark improves forecasting strategy with new demand planning software

When Kimberly-Clark wanted to boost its sales and inventory forecasting, it turned to demand planning software.

Consumer goods manufacturer Kimberly-Clark revealed recently that it plans to use new demand planning software to improve its sales and inventory forecasting. A fall 2009 pilot of Terra Technology's Multi-Enterprise Demand Sensing (MDS) application showed a 40% boost in accuracy, the companies said. 

With $19.1 billion in annual revenues and brands that include Kleenex and Scott, Kimberly-Clark maintains a demand signal repository (DSR) to consolidate data from retail point of sale (POS) systems, distribution centers, and syndicated services such as The Nielsen Companies and Information Resources Inc. (IRI), according to Jared Hanson, a demand senior specialist at the company. Terra's MDS software will run on top of the DSR behind the company firewall, outputting daily forecasts around midnight.

Selecting and implementing demand planning software

Kimberly-Clark chose the standard pilot, which costs $25,000. That is refundable if forecasts don't improve by at least 40%, according to Rob Byrne, Terra's chief executive officer. Terra ran data from two of Kimberly-Clark's infant and childcare lines through MDS, using 2008 figures to train the rules engine. "Then we ran this hands-off, with no interaction, for 2009," Byrne said.

The implementation began in February, and eventually, 15 to 20 demand planners will use MDS, according to Hanson. The software is virtually "no touch" and runs automatically after setup, he said.

MDS employs pattern recognition to identify which data streams have the most predictive value, according to Byrne. For a high-volume, low-priced item such as paper towels, for example, pallet shipments from a distribution center can be a reliable measure of demand, while POS figures have less relevance. On the other hand, for lipstick, a higher-priced item tracked by more stock-keeping units (SKUs), POS scanner data is necessary, he said.

But POS data has been over-hyped as a demand signal, in Byrne's view, and must be supplemented with other information, including company demand forecasts. "Maybe you sold 50,000 cases last week because Walmart had a promotion," he said. "It's just simplistic to say what's selling in the store is going to be ordered in the future."

The MDS forecasts will feed back into the SAP Advanced Planner and Optimizer (APO) software that Kimberly-Clark uses to plan production. Hanson expects most of the savings to come in distribution, primarily from reduced safety stocks.

Building a demand signal repository

Hanson said data management has been the biggest challenge in implementing MDS.

"A lot of the data streams that Terra is looking for, we have captured," he said. "They want very frequent refreshes of the data."

So Kimberly-Clark is working with retailers and third-party providers to standardize the information in their data streams.

Hanson said the company's IT department is working to cleanse the data going into the DSR. "They need to make sure the data they do collect is consistent across the board," he said. "We want to make sure there are no redundancies in the data."

The biggest problem has been trying to establish linkages between external and internal data, which often employ different schemes for matching UPC codes to product stock codes, for example.

Kimberly-Clark is also taking a hard look at its DSR, which is mostly homegrown and has been through several versions in three years, Hanson said. "There's more of an enterprise effort under way to determine if our current infrastructure is adequate. The processing power to massage and manipulate that amount of data is pretty large-scale."

What lessons has Hanson learned from the MDS project?

"Creating and understanding metrics for success is absolutely key," he said. "The data validation requirements are significant. That's a step that [must not] be underestimated."

Dig Deeper on Demand management