How Trendhim Co-Developed AI Forecasting with Kleene.ai and Cut Inventory 20%

Book call

Download PDF

Industry

Retail

employees

51-100

website

https://www.trendhim.co.uk/

About

Trendhim is a Danish direct-to-consumer brand selling men's accessories online to customers in 30+ countries. It designs its own 12,000+ products in-house, which means a long supply chain – 90 days at the fastest, often more than 180 – where forecasting accuracy directly determines whether the right product is on the shelf when a customer wants it.

50%+ less replenishment resourcing

20% lower inventory after 12 consecutive months of month-over-month decline

Lower out-of-stock rate despite running leaner inventory

Challenges

Trendhim's previous forecasting system had become a bottleneck. It would over-forecast some products and under-forecast others, and the logic behind it was opaque enough that the team couldn't trust it.

Worse, it didn't handle stockouts correctly. When a product was out of stock, the system logged the period as zero sales rather than treating it as missing data – so every stockout taught the model that demand was lower than it actually was, and the next forecast would under-order again. A self-reinforcing loop in the wrong direction.

"We had too many cases where we went out of stock and, when we went back to check the forecast, the miss wasn't subtle – it was off by a factor of 10 or more. Once you've seen that a few times, you lose trust in the system entirely and start double-checking everything by hand. That's the worst place to be – a forecasting tool that creates more work than it removes."

The only workaround was more man power: human review layered on top of every forecast. But the problem was structural, not a tuning issue – with the out-of-stock flaw baked into the model, no amount of human override could fix it. That's when it became clear the tool needed replacing, not patching.

Solution

Kleene.ai already ran Trendhim's data warehouse, so rather than buy a closed forecasting product, the two teams co-developed a new AI-driven forecasting application built natively on the existing data stack – shaped around Trendhim's operational reality rather than a generic SaaS.

The model retrains weekly off the warehouse and runs inside the buying team's weekly rhythm, with a feature set built to fit:

Product groupings with category-specific safety stock and per-supplier replenishment periods
MOQ optimization and exclude-period logic, so promotions don't poison the next cycle
Built-in supplier vacation handling for events like Chinese New Year, previously done manually

Since switch-over, Trendhim has seen 12 consecutive months of month-over-month inventory decline – now roughly 20% below baseline – while replenishment resourcing is down more than 50%. And despite the leaner inventory, out-of-stock is lower than before.

The most valuable thing has been getting back to a place where we trust the forecast. When you trust the system, you stop second-guessing every output, you stop manually overriding, and the team can focus on the strategic side of buying rather than the firefighting side.