Evaluation study / Evaluation & assurance
Overview
A growing bar wanted to replace manual forecasts with AI. It used Knotest to check whether the system could improve stock planning with its own sales data.
Method
Tests covered promotions, holiday demand, and weather changes. Before the pilot began, the team agreed on metrics, success criteria, and conditions for rollout.
Findings
An eight-week A/B study tested six top-selling Category A beers, rotating every two weeks. It recorded a 53.7% improvement in stockout rate, alongside improvements in overstock, holding costs, and planning time.
Scope & limitations
The evaluation ran for eight weeks across six Category A beers, with two-week rotations. The available summary does not report per-product baseline values, sample counts, or uncertainty estimates. The finding describes this pilot setting.
Data & evidence
This page preserves the published project summary. Raw data, individual test results, and reproducible code are not supplied with this record. Contact HephaKnot with questions about the evidence or method.
Original project account
For a Hong Kong darts bar where beer makes up 70 per cent of revenue, inventory planning used to be a game of experience — and luck.
“You check last week’s sales, see if there’s a match or holiday coming up, and make a call,” said manager Chan. “Sometimes you get it right. When you’re wrong, you really feel it — empty shelves when people are ready to buy.”
As the business grew, the bar wanted to move from manual forecasting to AI, but they needed proof before making the switch. They turned to Knotest, a platform developed by HephaKnot.
“We wanted hard evidence,” Chan said.
First, they ran a model performance test on Knotest, comparing the AI’s Accuracy, F1 and AUROC against their historical manual predictions. The model passed. Then came an eight-week A/B Business Impact Assessment on six top Cat A beers, with Group A using manual forecasting and Group B the AI system, rotating every two weeks.
The results spoke for themselves:
- Stockout Rate: −53.7%
- Overstock Rate: −37.1%
- Inventory Turnover: +25.0%
- Monthly Holding Costs: −22.5%
- Planning Time: −50.0%
- Stockout Complaints: −58.1%
“Once you see numbers like that, it’s not really a debate,” Chan said. “You know what you’re getting. Now we’re not just guessing. We know what’s coming.”