TURF analysis: choose the range that reaches more buyers

Your best-selling flavour may deserve its place in the range. The next most popular flavour may appeal to almost exactly the same buyers. If you can launch only two, choosing the top two individual scores can leave people without an option they would consider.

TURF analysis helps answer the range question: which combination covers the most people under a defined rule? The useful result names a feasible set, shows whom each addition reaches, and explains what could change the recommendation. A single winning percentage leaves too much of that decision hidden.

This guide works through a complete, fictional sparkling-water range. The individually strongest pair reaches six of twelve people. A feasible alternative reaches nine. Another pair reaches ten but cannot be supplied at launch. Those differences show why the response records and business constraints belong beside the result.

What is TURF analysis in market research?

TURF stands for Total Unduplicated Reach and Frequency. It compares sets of alternatives while counting each person only once in the set's reach. In product research, the alternatives might be flavours or services, with a fixed number of places available in the range. Displayr's TURF guide describes this portfolio task and the use of respondent-by-alternative records.

The definition of “reached” matters. Here, a person is reached when they would consider at least one selected flavour at the stated offer. Someone who would consider both selected flavours still counts as one person. This is coverage of stated consideration in the example; it is not a forecast of transactions.

The method suits a team asking, “Which two offers belong together?” If you need to determine which benefits deserve development, start with our MaxDiff guide. That priority question and the range question can use related evidence, but their recommendations need different calculations.

Set the offer and the constraints before comparing ranges

Imagine Rill, a fictional sparkling-water brand, has two launch slots. Each candidate is one 330 mL can offered at USD 1.80 before tax. The can format, displayed price and shelf context are held constant; the flavour label changes. The candidates are A Lime, B Lemon, C Berry and D Ginger.

Lime must remain because it is the approved incumbent. Ginger cannot enter the launch range because its flavouring supply has not been confirmed for the launch volume. These are explicit business conditions. They do not become evidence that people dislike Ginger.

The decision is therefore which available flavour should join Lime. We will still show every pair so the team can see what the constraints exclude and when it should revisit them. A supplier who returns only the unconstrained maximum would leave this launch decision unresolved.

Keep the person-by-flavour record

The following twelve records are manually authored teaching inputs. No survey, Sapien simulation, taste test or transaction was conducted. Each row represents a distinct fictional adult category shopper; all responses are complete and all weights are one.

A 1 means the person is recorded as willing to consider that flavour at the offer above. A 0 means they are explicitly recorded as unwilling to consider it. Read each row's four values in A / B / C / D order: Lime, Lemon, Berry and Ginger respectively.

Lime is considered by six people, Lemon by five, Berry by four and Ginger by four. Individual totals would put Lime and Lemon first. But P01 through P05 already appear in Lime's audience, so Lemon brings Lime no additional person.

Berry shares P06 with Lime and adds P07, P08 and P09. The smaller individual total makes a larger contribution to this particular range. You need the joint records to see that: four separate item totals cannot tell you which people overlap.

Calculate unique reach for all six pairs

For each row, check whether at least one selected column contains a 1. Count that row once if it does. Divide the number of reached people by the fixed eligible base of twelve.

For Lime and Berry, Lime covers P01 through P06. Berry adds P07 through P09, while P06 is already covered. The result is nine distinct people out of twelve, or 75%.

The recommendation under the stated launch conditions is keep Lime and add Berry. It reaches nine people, compared with six for the only other feasible pair. The unconstrained maximum is Lime and Ginger, but that pair is not currently available to the business.

P12 remains in the base even though no flavour reaches them. Removing that row would turn the same nine covered people into 9/11, or 81.8%, making the range look stronger without adding anyone. The eligible population should be defined before inspecting which offers it accepts.

Read frequency without turning responses into purchases

In this illustration, frequency counts accepted options within the selected range. We also show the mean number of accepted options per reached person. Neither measure records how often anyone shops or how many cans they buy.

Lime and Lemon collect eleven positive entries from six people. They do not reach eleven people. Five of those six people accept both flavours, which gives the range depth among the same audience. Lime and Berry spread ten positive entries across nine people.

That distinction matters when reading a supplier's report. If the objective is to give more eligible people at least one acceptable option, compare unique reach. If having several acceptable options for the same person matters, explain that additional objective and examine it separately. Repeated willingness to consider an offer does not establish repeat purchase.

Show what would change the recommendation

If Ginger's supply is confirmed and Lime must still remain, Lime and Ginger becomes the highest-reaching pair in these records. It covers P10 and P11, who are outside Lime and Berry's range, but loses P09. The net gain is one person, from nine to ten. That is a change in the composition of coverage as well as its total.

If Lime no longer has to remain but Ginger is still unavailable, Lime and Berry and Lemon and Berry tie at nine. They cover the same people, P01 through P09. Reach alone cannot select a unique winner. Operational suitability or another agreed objective could resolve the tie, provided the report keeps that reasoning separate from the reach calculation.

Twelve invented records make these relationships easy to inspect. They provide no estimate of sampling uncertainty or population demand. In a commissioned study, ask how stable the leading sets are under the sampling or modeling approach, alternative reasonable reach rules, and the buyer groups that matter to the launch. Plan the evidence around the consequence of choosing the wrong range rather than borrowing a universal sample-size rule.

Agree which definition of reach the study uses

The all-or-nothing consideration rule above is one approach. Sawtooth's TURF documentation also describes probability-weighted, first-choice and threshold approaches for its MaxDiff analysis. A probability-based result should be interpreted using its model; it is not the same calculation as counting the rows in our binary table.

Ask the analyst to retain the question, response scale, any threshold or utility transformation, and the rule used to calculate reach. If a rating becomes a yes/no entry, agree the cutoff before selecting the winning set. A change from “would consider” to “first choice” changes what the analysis answers.

Keep missing answers distinct from explicit rejection. The illustration has no missing cells. Real records may require a pre-agreed treatment of incomplete responses or a suitable model, with its assumptions and resulting base disclosed. Silently replacing every blank with zero changes the meaning of the data. If justified audience weights are used, report the weights and calculate coverage against the corresponding eligible weighted base.

Commission the evidence behind the range decision

A useful engagement returns more than a sorted list. For Rill, the team needs to inspect the complete offers shown, the eligible audience, the joint response data or auditable modeled inputs, the reach rule, and the two launch constraints. The readout should include the leading feasible sets and enough detail to explain why the apparent maximum was excluded.

The response source belongs with that record. Human stated consideration, modeled consideration and purchase history describe different observations. If the analysis uses synthetic respondents, examine how the audience is grounded and which human reference evidence supports this task. Our guide to synthetic and human respondents explains how to choose and combine those evidence sources.

For a larger range, confirm whether the analyst evaluates every feasible combination or uses a search procedure, and what that permits them to claim about the returned set. Retain close alternatives and ties so a later supply change can be evaluated without hiding the original decision rule.

Interpretation should then explain the people the addition brings into the range. In Rill's example, Berry adds three people whom Lime misses. The fictional table cannot explain their reasons. A real study could investigate the occasions or expectations behind that difference, while preserving the initial responses rather than changing them after seeing the preferred result.

Decide what still needs another kind of test

TURF coverage does not establish whether Berry takes purchases from Lime, brings shoppers from a competing brand, or creates a new occasion. When the decision depends on what a range change replaces in a real shelf encounter, our shopper research guide follows that choice and substitution question.

If the offers differ in size, features and price, the business may instead need a conjoint study to investigate those trade-offs. Keeping the price fixed in Rill's illustration avoids answering a different product-configuration question with the same reach table.

Taste and physical use need suitable evidence with the product. Supply readiness needs confirmation from operations. A commercial rollout should retain availability, distribution, promotion and actual sales records so the team can investigate trial and repeat purchase. None of those outcomes is measured by the nine positive rows in the recommended pair.

Bring Sapien the range decision

Bring the candidate offers, intended buyers, current range and conditions that could rule an option in or out. Sapien builds data-grounded simulated audiences, compares alternatives and helps teams interpret differences across groups. That work can help investigate the reactions and reasons behind a product-range decision.

If unduplicated coverage is the required output, agree the joint response records, reach calculation and analysis deliverable as part of the study design. Keep human product experience and commercial validation in scope where the decision needs them. The useful next step is a research design tied to the range you can actually launch.

Book a demo to explore your product-range question with Sapien.