Ask buyers what they want in a protein bar and they will say more protein, less sugar, better taste and a lower price. All true, and no help. Every real product is a set of trade-offs, and a line-up has room for only a few flavors. Two methods are built for exactly these choices.
The method
A choice-based conjoint shows each buyer a series of screens. Each screen has three products that differ in features and price, plus the option to buy none, and the buyer picks one, about twelve times. Those choices reveal how much each feature is worth, measured against price, without asking the question directly.
TURF, short for total unduplicated reach and frequency, works from a simpler rating of each flavor or variety. It searches every possible line-up for the few items that together give the most buyers something they want.
The question it answers
What should the product be, and how many versions should we make? Conjoint answers the first: which features, which level of each, and what a change does to share against named competitors. TURF answers the second: which three or four flavors, colors or sizes reach the most people with the least overlap.
How we run it
We start with the decisions the team actually faces and turn them into features and levels, usually four to six features with two to four levels each. Price levels span the real shelf and are treated as test levels. A typical study has 600 to 1,000 category buyers, each completing 10 to 14 choice tasks. We estimate each buyer's own values with hierarchical Bayes, so results can be cut by user type, region or channel. The same buyers then rate every candidate flavor, which keeps the product and the line-up tied to the same people.
What you get
Feature importance and part-worths, a guide in cents to what each upgrade is worth, and a simulator your team can keep using after the readout to test any combination against competitors. For the line-up: reach for any set you choose, the best set at each size, and the point where one more flavor stops paying for itself. A single recommendation ties both together.
Where it goes wrong
Too many features turn the choice tasks into a memory test, and buyers start choosing on price alone. Price levels outside what people have seen produce answers nobody believes. Share of preference gets read as a sales forecast, though it assumes every buyer knows and can find every product. And line-ups get built from the most popular items, which often appeal to the same people. Three favorites can reach fewer buyers than two favorites plus one item with a different following.
The sample below follows a fictional snack brand, Copperleaf, choosing a new protein bar and its launch flavors with 800 bar buyers. Try a choice task, see what matters most, build a bar in the simulator and pick a flavor line-up to watch its reach change. Every brand, price level and number is invented.
Illustrative only. All brands, products, price levels and figures are fictional. Open the sample full screen.