You can only put so many flavors on the shelf, so many features in a bundle or so many claims in a campaign. Picking the items people like most seems sensible, but popular items tend to appeal to the same people, and the set ends up reaching fewer customers than it could. TURF analysis solves that problem by looking at the combination rather than the items one at a time. It sits alongside the methods in our guide to customer segmentation analysis, because both ask the same question: who are you reaching, and who are you missing?
Key takeaways
- TURF analysis picks the combination of items that reaches the most people at least once, not the items with the highest individual scores.
- Reach counts unduplicated people; frequency counts how many items in the set each of them accepts.
- The input is a yes/no acceptance judgment for every item and respondent, usually from top-two-box purchase intent or MaxDiff.
- Exhaustive search checks every combination and finds the true best; greedy search adds one item at a time and is faster but not guaranteed.
- Read TURF results by segment and with weights before you commit to a line-up.
What is TURF analysis used for?
TURF analysis answers line-up questions where you can offer a limited number of options and want as many people as possible to find at least one they would buy. Typical uses include:
- Product line and portfolio decisions: which variants to keep when a range is pruned, or which to add.
- Flavor ranges and menus: which flavors launch first, which dishes go on a seasonal menu.
- Message sets: which three claims to rotate in a campaign so that most of the audience finds one that lands.
- Feature bundles: which features go into an entry-level package.
It is the wrong tool for measuring how much people prefer one item over another, or which attributes drive a choice.
What inputs does a TURF analysis need?
TURF needs one yes/no judgment per respondent per item: would this person accept it? You set the threshold that turns survey answers into that judgment, and it shapes the result.
- Top-two-box ratings. Ask purchase intent for each item on a five-point scale and count “definitely” or “probably would buy” as accepted.
- MaxDiff scores. Turn each respondent’s item scores from a MaxDiff analysis into accept or reject, for example by counting each person’s top few items as accepted, or by adding an anchor question that asks which items they would actually buy.
- Direct selection. Ask “Which of these would you buy?” and accept whatever is ticked.
The result is a matrix of 1s and 0s: one row per respondent, one column per item.
What is the difference between reach and frequency?
Reach is the number, or share, of respondents who accept at least one item in the combination. Each person counts once, however many items they like; that is the “unduplicated” in the name.
Frequency is the total number of acceptances within the combination: if one person likes two of the three items, they add 2. Dividing frequency by reach gives the average number of acceptable items per reached person. Reach decides the ranking; frequency breaks ties and tells you how much choice reached people have.
TURF analysis example: which three flavors should you launch?
The numbers below are illustrative, kept small so the arithmetic is visible. An ice-cream maker can launch three of five new flavors. Eight respondents rated each flavor on purchase intent, and 1 means top-two-box.
| Respondent | Vanilla | Salted caramel | Strawberry | Mango | Pistachio |
|---|---|---|---|---|---|
| R1 | 1 | 1 | 1 | 0 | 0 |
| R2 | 1 | 1 | 0 | 1 | 0 |
| R3 | 1 | 1 | 0 | 0 | 1 |
| R4 | 0 | 0 | 1 | 1 | 0 |
| R5 | 0 | 0 | 0 | 0 | 1 |
| R6 | 1 | 0 | 0 | 0 | 0 |
| R7 | 0 | 0 | 0 | 1 | 0 |
| R8 | 1 | 1 | 1 | 0 | 0 |
| Single-item reach | 5 (62.5%) | 4 (50%) | 3 (37.5%) | 3 (37.5%) | 2 (25%) |
Step 1: Rank the single items
Vanilla is the most popular flavor (5 of 8), then salted caramel (4 of 8). Picking by popularity, you would launch those two first.
Step 2: Find the best pair
Now check all 10 possible pairs. The top five:
| Pair | Respondents reached | Reach | Frequency |
|---|---|---|---|
| Vanilla + Mango | R1, R2, R3, R4, R6, R7, R8 | 7 of 8 (87.5%) | 8 |
| Vanilla + Strawberry | R1, R2, R3, R4, R6, R8 | 6 of 8 (75%) | 8 |
| Vanilla + Pistachio | R1, R2, R3, R5, R6, R8 | 6 of 8 (75%) | 7 |
| Salted caramel + Mango | R1, R2, R3, R4, R7, R8 | 6 of 8 (75%) | 7 |
| Vanilla + Salted caramel | R1, R2, R3, R6, R8 | 5 of 8 (62.5%) | 9 |
The other five pairs each reach 5 of 8. Now look at the two most popular flavors together. All 4 salted caramel fans already like vanilla, so unduplicated reach is 5 + 4 − 4 = 5: salted caramel adds nobody. Its frequency of 9 (9 ÷ 5 = 1.8 flavors per reached person) is the highest of any pair, the signature of two items that please the same people.
Vanilla + Mango wins: 5 + 3 − 1 = 7, because only R2 likes both. Frequency is 5 + 3 = 8, or 8 ÷ 7 = 1.14 per reached person.
Step 3: Find the best set of three
Checking all 10 sets of three gives one clear winner:
| Set of three | Reach | Frequency | Per reached person |
|---|---|---|---|
| Vanilla + Mango + Pistachio | 8 of 8 (100%) | 10 | 1.25 |
| Vanilla + Salted caramel + Mango | 7 of 8 (87.5%) | 12 | 1.71 |
| Vanilla + Strawberry + Mango | 7 of 8 (87.5%) | 11 | 1.57 |
| Vanilla + Salted caramel + Strawberry | 6 of 8 (75%) | 12 | 2.00 |
The other six sets reach 6 or 7. Pistachio, the least popular flavor, completes the best set because it is the only flavor R5 accepts: reach is 7 + 1 = 8 and frequency 8 + 2 = 10. The three most popular flavors (vanilla, salted caramel and strawberry or mango) would reach only 6 or 7. The reach curve for the best sets runs 62.5%, 87.5%, 100%.
Greedy or exhaustive: how do you search the combinations?
Exhaustive search calculates reach for every possible combination and takes the best. It guarantees the true optimum, but the count grows fast. Five items give 10 sets of three. Twenty items give 15,504 sets of five, thirty items give 142,506 sets of five, and forty items give 3,838,380 sets of six.
Greedy (stepwise) search starts with the highest-reach item, then adds whichever item brings the most new people, and repeats. In the example it picks vanilla (5), adds mango (+2, to 7), then pistachio (+1, to 8), matching the exhaustive answer. That is not guaranteed: greedy search never revisits its first pick, so it can miss a set built from less popular items that together cover more people. Use exhaustive search when the item count allows, and treat greedy results as a shortlist otherwise.
How do weighting and segment-level TURF change the answer?
If your sample over-represents some groups, weight it before running TURF analysis. Weighted reach is the sum of the weights of reached respondents divided by the sum of all weights, so a reached respondent with weight 2.0 counts twice.
The best line-up for the whole market is also not necessarily best for each segment. Run TURF within the segments that matter, whether defined by purchase behavior or found through cluster analysis, and compare the winning sets. Before you do, check the acceptance rates themselves with crosstabs and banner tables with weighting and significance testing, so you know which segment differences are large enough to act on.
What are the limitations of TURF analysis?
- The threshold drives the answer. Move from top-two-box to top-box and the matrix, and possibly the winning set, changes. Test the result under both.
- It ignores cannibalization. Reach counts people, not purchases. A set can reach everyone and still sell little more if the new items mostly replace existing ones.
- It ignores intensity. A person who “probably would buy” counts the same as an enthusiast.
- It reflects stated intent. Acceptance in a survey is not shelf behavior, and items are judged one by one, not as a set on display.
What are the common mistakes?
- Launching the top items by individual score. As the example shows, the two most popular flavors can reach the same people.
- Choosing the threshold after seeing the results. Fix the acceptance rule before you run the analysis, or you can steer it toward any answer.
- Calling a winner on a one-point gap. Reach is a sample estimate. Two sets a point or two apart are effectively tied; use frequency, cost or strategy to choose.
- Running TURF on an unweighted, skewed sample. The winning set will fit the over-represented group.
- Treating 100% reach as the finish line. Once everyone is reached, another item adds only frequency. Weigh that against the cost of making and stocking it.
How do you put the results to work?
Treat the winning set from a TURF analysis as a hypothesis. Rerun it under a stricter threshold and within your key segments, then check it against current sales. Once the line-up is in market, a key driver analysis shows what keeps each segment buying, and seeing what drives satisfaction and loyalty by segment helps you spot when a group the set was meant to reach starts drifting away.