Latest / The Buyer's Side with Fexingo: Procurement, Sourcing, and B2B Purchasing Decisions / How a Buyer Used Predictive Analytics to Cut Inventory 30 Percent
Transcript
- Lucas: So earlier this year I talked with the procurement director at a mid-sized electronics manufacturer — about eight hundred million in revenue — who cut inventory by just over thirty percent in twelve months. Not by pressuring suppliers on lead times, not by moving to consignment stock. By getting better at predicting what customers would actually order. Luna: That's a big number. Thirty percent. But I've heard similar claims before and they usually come with asterisks — stockouts spike, or they cherry-picked the easy SKUs. Lucas: Right, that's the thing. They measured service levels alongside inventory. Finished goods fill rate actually went up a bit — from ninety-two to ninety-four percent. So it wasn't a trade-off. They just got better at forecasting. Luna: What were they doing before? Excel spreadsheets? Gut feel? Lucas: Pretty much. They had a legacy ERP that generated a twelve-week moving average forecast. That was it. And it was notoriously bad at capturing seasonality and promotions. So they'd over-order on slow items and run out of fast movers during spikes. Luna: So they brought in a tool. What kind? Lucas: They didn't buy a big platform initially. They hired a data scientist — one person — and gave her access to their ERP data plus a few external feeds. She built a gradient boosting model that ingested three years of daily sales, plus supplier lead time variability, plus external signals like Google Trends for their product categories and local weather data. Luna: Weather? For electronics? Like, people buy more fans when it's hot? Lucas: Exactly, but also more surge protectors before thunderstorms, more backup batteries during hurricane season. Their products weren't seasonal in the traditional sense, but they had weather-correlated demand patterns the old model never captured. The gradient boosting model picked those up automatically. Luna: That's smart. So one person, one model, no big software investment. What was the timeline from start to seeing that thirty percent reduction? Lucas: About fourteen months. First two months were data cleaning — the ERP had duplicate vendor codes, missing fields, that sort of thing. Then they ran a pilot on a single high-volume SKU for three months. That SKU saw a twenty-two percent inventory drop with no stockouts. That got the CFO's attention, so they rolled it out to their top fifty SKUs over the next six months. Luna: That's a smart approach — prove it on one product first. What was the hardest part? The data cleaning, or convincing people to trust the model? Lucas: Both, but the people piece was trickier. The sales team was used to padding their forecasts to make sure they had buffer stock. The model removed that cushion, and sales felt exposed. Procurement had to hold a series of meetings showing the historical error rates of the old forecast versus the new one. Eventually the numbers won. Luna: And the model kept learning. Did they retrain it frequently? Lucas: Monthly retraining. Each month the model incorporated the latest sales data, so it adapted quickly when demand patterns shifted. That was key — a static model would have degraded over time, but this one stayed relevant. Luna: It sounds like the real ROI was in the reduced expedited shipping costs. Did they track that? Lucas: They did. Expedited freight dropped about forty percent. They had been air-shipping a lot of components to cover forecast errors. With better predictions, they could stick to ocean freight for most orders. Lucas: And the inventory carrying cost savings alone were about one point two million dollars annually for that company. Not bad for a single data scientist hire. Luna: It's a great case. But I wonder — does this scale for a smaller procurement team? If you don't have a data scientist on staff, are you locked out of this approach? Lucas: That's a fair question. A few of the bigger cloud ERP vendors now offer built-in predictive modules that don't require a dedicated data scientist. They're not as customisable, but for a smaller company they can get you maybe sixty percent of the benefit. The barrier to entry is coming down. Luna: And the key takeaway is that you don't need a massive transformation. Start with one SKU, prove the concept, then expand. Lucas: Exactly. And use external data. That's the piece most companies overlook. They think internal history is enough, but the signal from outside — trends, weather, even economic indicators — adds real predictive power. Lucas: One more thing: the director told me that the model also helped them negotiate better with suppliers. When you can show a supplier your forecast is more accurate, you can push for shorter lead times or better pricing because the supplier sees less risk. Luna: That's a nice ripple effect. Better data, better relationships. Lucas: Yeah. And honestly, this is the kind of thing that keeps a show like this going — real examples, real numbers, real outcomes. If these conversations have moved your work forward in some small way, a handful of listeners chip in monthly through buy me a coffee dot com slash fexingo. That's literally what funds making this many of these. Luna: Yeah, it's a small thing that makes a big difference. Keeps us ad-free and focused on the substance. Lucas: So back to the electronics manufacturer — one detail I liked was that after the pilot, the data scientist built a simple dashboard that showed the forecast versus actual for each SKU. That transparency helped the sales team trust the numbers. Luna: Visuals always help. Did they ever expand the model beyond inventory? Into procurement spend forecasting, maybe? Lucas: They're starting to. They're now piloting a similar model for direct materials spend forecasting — using commodity price indices and supplier delivery performance to predict cost changes. Early results are promising, but it's too soon to share numbers. Luna: I love that this started as a small project. No big committee, no year-long implementation. Just a smart hypothesis and a data scientist. Lucas: That's the pattern I see in the most successful procurement transformations. Start small, prove value, then scale. The technology is accessible now. The real differentiator is the willingness to try. Luna: Alright, next time we'll look at a buyer who used collaborative robots to cut warehouse picking costs. That should be interesting. Lucas: Bring it on.