For a global tea giant, a startling reality drove the need for transformation: most of their sales came from promotions.
Like many consumer goods companies, the tea giant had built their business model around discounts, premium store positioning, and promotional campaigns. But they were making multi-million-pound decisions on promotions with minimal visibility into how they might perform.
They forecasted numbers one or two weeks in advance. This short-term thinking meant the tea giant was constantly scrambling—adjusting production schedules, managing inventory levels, and missing revenue opportunities.
The problem wasn’t unique. Any consumer goods company running promotions faces the same fundamental challenge: How do you predict the sales lift from a promotion that hasn’t happened yet? Traditional statistical models could forecast based on historical sales patterns, but promotions create external variables that historical data alone cannot capture.
“If you have a holiday, there’s a sales peak, there’s a bump in demand. You have to produce more in those periods,” said Pivot’s analyst. “The problem Pivot was trying to solve is to understand, based on the future planned promotions, what will be the bump in the sales?”
The tea giant hadn’t ignored this challenge. They have been running this particular project for two years. They would make progress, and then park it.
The stop-start nature of their internal efforts highlighted a common enterprise challenge: complex machine learning projects need sustained focus and specialised expertise. That’s difficult to maintain alongside day-to-day operations.
When Pivot’s team joined the project, they brought hard-won experience from machine learning transformations at other global consumer giants.
Like any AI or machine learning project, data is key. A lot of data is not available. There are gaps. We worked on that as a starting point.
Instead of building models on theoretical or dummy data, Pivot insisted on working with live production data from the start. They prioritised practical results that planners could trust and use immediately.
Pivot used real-life production data for testing. “These were accurate numbers which they are planning in their day-to-day lives,” said Pivot’s analyst.
“By anticipating data challenges and building quality controls from the start, Pivot avoided the months-long delays that typically derail machine learning initiatives.
Pivot’s technical approach was equally pragmatic.
“The planners have been delighted to see the results, because they understand that it’s accurate.”
Where the data is good, machine learning tends to do better than normal statistical forecasting. But in some cases, regular statistical forecasting and moving averages can beat machine learning. We were doing a pick-and-mix approach. The new model incorporated variables that traditional forecasting missed: promotion stores, price information, type of promotion, holidays, and seasonal patterns. By understanding these relationships, the tea giant could now answer critical questions:
What happens if I decrease my price by 50 pence? Do I need to produce more? Do I need to have two extra shifts on Saturday and Sunday to fulfil demand?
The results have been compelling. The team had been predicting August sales using data up to May 2025, and found the machine learning system achieves exceptional accuracy in these three-month forecasts.
“The planners have been delighted to see the results, because they understand that it’s accurate,” he said.
For a business where promotional investments can run into millions of pounds across hundreds of stores, this level of predictive accuracy translates directly to bottom-line impact.
The system goes live in late October 2025, followed by extensive business user validation before full deployment in early 2026.
The measured rollout reflects Pivot’s commitment to sustainable implementation, rather than dramatic launches that fail to deliver lasting value.
The project demonstrates how machine learning, implemented pragmatically with deep business understanding, can transform fundamental operational capabilities. By focusing on business outcomes rather than technical sophistication, Pivot is delivering a solution that works in the real world.
