This is the Electee Review. Today, why the forecast for a new product isn't a number. It's a set of decisions. The article makes a blunt claim. Forecasting demand for a new product works best as a way to learn and decide, not as a search for one perfect figure. The supplier wants a purchase order and marketing has booked the campaign. Nobody knows whether customers will buy steadily, rush in, or wait. Order cautiously and you lose sales. Order aggressively and you tie up cash and stock you'll have to discount. A confident spreadsheet target doesn't settle that trade-off. Start with a skeptical point about the data itself. Sales you observe are not the same as real demand. Customers can only buy what is in stock in channels they can reach. A late marketplace listing or a delayed email campaign changes the launch, so weak sales may not mean weak interest. And a wait list shows interest. It is not an order book. On comparable products, the article cites research showing that averaging several similar past launches beats relying on the single closest match. The gain levels off after about five or six. Experts should give their estimates separately, Delphi style. So the most senior person doesn't set the opening number. Models? Start simple. The BAST diffusion model separates buying driven by advertising from buying driven by word of mouth. But its settings are assumptions, not values to borrow from another category. Studies cited show that combining several models improves accuracy. The article is careful though. These are results from specific cases, not universal promises. Machine learning reorganizes the evidence you have. It doesn't create a launch history that doesn't exist. Here's the number that should worry planners. In one study of durable goods, a first-year forecast error of 0% came with a monthly mean absolute percentage error of 26.5%. The total looked perfect. The timing didn't. That gap is where stockouts and late reorders happen. So the forecast has to feed an explicit inventory policy. The losses are not equal. A stockout kills momentum, while extra stock means markdowns. Stage the first order, set reorder triggers, review weekly in the opening weeks, and name who approves exceptions. And when no comparable product can be trusted, as with products that create a new category, drop the false match. Use explicit scenarios and clearly labeled assumptions instead. The takeaway you don't need to eliminate uncertainty. You need to make it visible and decide in advance how you'll respond. That's the review.