What is TURF Analysis and When to Use It?
A brand or category manager is choosing which flavors, SKUs, or ad claims to shortlist from a longer list. TURF analysis is the tool built for that decision. TURF, short for Total Unduplicated Reach and Frequency, ranks a fixed candidate list by unduplicated reach. The ranking assumes every item is chosen independently of the others. Use it when the decision comes down to reach or budget. Stop relying on it once the real question becomes which item steals sales from another. TURF has no mechanism to see that.
- TURF ranks candidate items (flavors, SKUs, ad claims) by unduplicated reach, treating each item as if it were chosen on its own.
- It's a good fit for a first-pass shortlist from a fixed candidate list under a budget or reach target.
- It's a poor fit once trade-offs, pricing, or cannibalization decide the outcome, because independence is an assumption TURF cannot verify.
- Discrete choice modeling (McFadden discrete choice, Mixed Logit, ICLV) replaces TURF's independence guess with an estimate of incremental lift versus cannibalization.
- Subconscious runs randomized experiments analyzed with these models and reports a confidence interval that covers the effect within the simulated population, not a bare rank.
## What is TURF analysis, and where did it come from?
TURF is a reach-maximization method that started in 1960s media planning, where it answered a simple question: which set of ad placements reaches the most unique viewers without wasting frequency on people already reached. It migrated into product line, flavor and SKU research, and messaging testing, where it still answers a narrow, useful question: given a fixed list of candidates, which subset maximizes the unduplicated share of respondents who'd pick at least one? Sawtooth Software, Qualtrics, quantilope, and Decision Analyst all ship or support dedicated TURF modules today, and it's routinely run right after a MaxDiff study to shortlist flavors or claims for the next stage of testing.
## When does TURF actually answer the question you're asking?
TURF works when the decision is genuinely just "which items to include" and the items are close substitutes drawn from a fixed list. That covers a lot of real shortlisting work: narrowing twelve flavor concepts to four for a limited shelf set, picking which five ad claims to test further, or choosing which SKUs cover the most households under a fixed number of facings. In that framing, TURF is cheap to run, easy to explain to a merchandising committee, and fast. The moment the list stops being close substitutes, or the decision needs to account for price, positioning, or which item takes share from which, TURF is being asked a question it wasn't built to answer.
## What does TURF assume, and where does that assumption break?
TURF assumes every item in the candidate set is picked independently of the others, and that assumption is also its most cited limitation. TRC's whitepaper on TURF implementation frames the standard method as implicitly treating candidate items as independent choices, and proposes hybrid "pseudo-TURF" approaches that layer choice-model utilities on top of TURF's reach logic to recover some of the complementarity and trade-off information the base method misses ([TRC Insights](https://trcmarketresearch.com/whitepaper/turf-new-methods-for-implementation/)). Decision Analyst's TURF whitepaper names a second assumption buyers often miss: TURF assumes 100 percent distribution and 100 percent awareness for every item in the candidate set, conditions that are rarely true in an actual retail launch ([Decision Analyst](https://www.decisionanalyst.com/whitepapers/turfanalysis/)). A third gap sits underneath both: TURF's reach number comes from stated "would you buy this" survey data, with no in-market or experimental check on whether that stated intent held up. Worth noting: TRC and Decision Analyst both sell or support TURF work. This isn't outside criticism. It's the industry's own methodology vendors documenting where their tool runs out of road.
Put together, those three assumptions mean TURF's ranking can look identical whether an added item pulls in a genuinely new buyer or just pulls a buyer away from an item already on the list. That's not a rounding error in the output. It's the difference between a ranked list and an answer to what launching Item C does to Item A's sales.
## TURF vs. discrete choice modeling: two different questions
These two methods answer different questions, and confusing them is the most common way a TURF study gets over-trusted. Quirk's notes that TURF can become computationally intractable once products carry multiple attributes, and that discrete choice modeling, while more expensive to run, overcomes most of TURF's limitations and produces more accurate predictions of in-market outcomes ([Quirk's](https://www.quirks.com/articles/discrete-choice-modeling-for-product-portfolio-optimization)). Quali-Fi puts the distinction plainly: TURF does not capture product trade-offs or the reasoning behind selection, while conjoint and discrete choice modeling make those trade-offs and that reasoning explicit ([Quali-Fi](https://quali-fi.com/learn/turf-vs-conjoint)).
| Dimension | TURF analysis | Discrete choice modeling |
|---|---|---|
| Question it answers | Which subset of a fixed list reaches the most people | Why people choose, and which action shifts that choice |
| Core assumption | Items are chosen independently of each other | Choices trade off against explicit attributes, including price |
| What breaks it | Cannibalization, partial distribution or awareness, no live validation | Requires more design and analysis effort up front |
| Output | A ranked shortlist by unduplicated reach | Preference shares, cannibalization estimates, confidence intervals |
| Best for | A first-pass shortlist from a fixed candidate list under a reach or budget constraint | Any decision where trade-offs, pricing, or cannibalization decide the outcome |
## Which randomized experiment replaces TURF's independence assumption?
A randomized experiment analyzed with discrete choice models replaces TURF's independence assumption, so cannibalization gets measured instead of assumed away. The experiment runs on a simulation of the market; the resulting choices are analyzed with McFadden discrete choice, Mixed Logit, and ICLV rather than counted into a reach score. That distinction matters: DCE, Mixed Logit, and ICLV are estimators, not causal methods on their own. The causal claim comes from the randomized manipulation built into the experiment design, not from the estimator that reads the results afterward.
Where a plain multinomial logit is used to read out preference share or substitution patterns, it carries the IIA assumption, that a change to one item's attributes shifts every competing item's share by the same proportion. The symptom to watch for: two candidate items that draw from different respondent segments, a mainstream flavor and a niche one with a small but loyal following, for example. A plain logit spreads the substitution evenly across the field and understates how much the niche item cannibalizes its closest match. That's the signal to ask whether Mixed Logit was used instead. Mixed Logit relaxes the IIA assumption when respondent-level taste variation matters, which is one reason it's used alongside McFadden discrete choice rather than instead of it.
Accuracy is checked against real human studies, not asserted. Simulated studies reproduce the direction and outcome of the original human study in 93 percent of cases, a figure we call replication accuracy, per the [validation paper](https://go.subconscious.ai/paper); results are visible on the [leaderboard](/leaderboard), and the full methodology sits in the [methods and validation hub](/blog/methods-and-validation). Two limits are worth stating plainly. First, 93 percent replication accuracy is a validation-set result, not a guarantee for a new, unseen market. Second, some published human studies used for validation could theoretically overlap with a model's training data; the replication protocol is designed to catch that, but the possibility isn't eliminated by assertion.
## Is TURF analysis still worth using?
Yes, for the one job it was built for: a fast, cheap first pass at shortlisting from a fixed list of close substitutes under a reach or budget constraint. It stops being worth trusting the moment the decision needs to know what launching one item does to another's sales, what price does to the shortlist, or whether the "reach" number would survive contact with actual distribution and awareness levels. At that point, a randomized experiment analyzed with discrete choice models answers a question TURF was never built to ask. TURF's shortlist shows who it reaches. It doesn't show which action actually drives the outcome: adding an item, dropping one, or moving price.
## What to do next
Before running or trusting another TURF study, check whether the decision is genuinely a fixed-list shortlist (TURF is fine) or a trade-off decision involving price, cannibalization, or "why" (it isn't, and TURF's output will look more confident than it is). If it's the latter, ask the vendor running the study whether the independence assumption was tested or just assumed, and ask what validation, if any, backs the reach number. If you want to see how a causal read on the same kind of decision is validated, the [leaderboard](/leaderboard) has the replication results by study. When you're ready to talk through a specific portfolio or messaging decision, [the team is here](/meet).Summary of changes: rebuilt the intro's two overloaded sentences into short declaratives (buyer/decision split; "use it / stop / no mechanism" split); fixed the 93% sentence's grammar and definition placement; scoped the confidence interval to the simulated population in the bullet list; added a one-line disclosure that the TURF-assumption sources are vendor/practitioner literature, framed as a credibility point rather than hidden; recast the "which action" list without comma-as-dash; retitled the Subconscious section as a direct question in the owned lexicon; and added a concrete segment-based symptom that tells a reader when to demand Mixed Logit over a plain logit. No other sections were restructured.