Pakketo.

Pakketo launched in 2016 with a focus on furniture and home decoration — functional design, good quality, and accessible prices, with a catalog that keeps growing. By 2019 the brand had established itself as a serious online destination for furnishing both indoor and outdoor spaces across Greece.

The brief for Black Friday 2019 was sharp: +80% revenue within a specific target ROAS. The obvious path — tease the sale early to build demand — carries a well-known risk in categories like furniture, where purchases are rarely impulsive. If existing shoppers catch the signal that Black Friday prices are coming, they stop buying at full price in the weeks leading up to the event, and the brand cannibalizes its own pre-BF revenue to fund the BF peak. The goal had to be met without pre-tipping the discount.

That constraint turned Black Friday 2019 into a forecasting problem as much as a performance problem. Pakketo needed to know, in advance, how users would behave during the BF week — volume, conversion rate, cross-category interaction, and the impact of the unique pricing structure — so that budgets, bids, and inventory could be positioned for maximum return. A pure execution approach, without that forecast, would leave money on the table or burn it in mis-priced campaigns.

We built a custom forecasting report in R Studio using Facebook Prophet as the underlying time-series model. Prophet was the specific choice because it handles the two dynamics that matter most during Black Friday: it supports holiday effects (sudden demand spikes tied to known calendar events) and accepts user-provided parameters for factors that are not visible in historical data — such as the specific discount depth planned for BF 2019 or the inventory positioning decisions made internally.

The model ingested two years of Pakketo sales data broken down by category, device, and audience segment, then projected expected demand patterns across the full Black Friday week at the day-and-category level. The output was not a single number but a distribution: probability-weighted scenarios for how revenue, traffic, and conversion rate would move hour by hour across the key BF windows.

Those forecasts drove every operational decision. Budget pacing was set against the projected demand curve rather than a flat daily spend. Bid strategies were pre-tuned to the expected CPC inflation. Inventory and ad creative were positioned around the categories forecast to overperform. And the pre-BF communications stayed silent on pricing — the brand ran its normal advertising, and the forecast confirmed there was no need to pre-warn users to hit the target.