A single solution resulting from a TURF analysis should never be delivered to the end customer; instead, a set of candidate solutions should be considered and analyzed together with the customer, their data, and their internal constraints, in co-construction.
The TURF analysis (Total Unduplicated Reach and Frequency) is a data analysis method that identifies the combination of products, features, or messages that maximizes the impact of a portfolio on a target audience. It is widely used to optimize product ranges, media plans, or marketing offers, relying on data from subscription intent, purchase intent, multiple-choice questions, or conjoint/MaxDiff analyses.
Usual Practices for TURF Analysis Reporting
The classic TURF analysis reporting consists of presenting:
- The optimal combination for each portfolio size (for example, the best assortment of 3, 4, 5 products out of 20 to be tested).
- The incremental reach brought by each additional element to the combination.
- A table or a chart showing the progression of the portfolio’s performance according to its size (number of offers included).
This reporting generally relies on an iterative algorithm: first, the element with the most subscription or purchase intent is selected, then elements that bring the most additional coverage are successively added until the desired portfolio size is reached. This process is easy to understand and explain, which explains its popularity among research agencies and customers.
Limitation of Classic Reporting: The Risk of a Local Optimum
However, this reporting has a major risk: the iterative algorithm used does not guarantee reaching the global optimum, but only a local optimum. In other words, the best combination found by this method is not necessarily the one that truly maximizes the total impact of the portfolio on the target audience.
Definition:
A local optimum is a solution to an optimization problem that is the best possible solution within a restricted region of the search space. That is, in a given neighborhood, but is not necessarily the best global solution. In other words, in this neighborhood, no other solution is better, but elsewhere in the search space there may exist a superior solution, called the global optimum.
The local optimum is a central concept in optimization, as many algorithms can get us “stuck” there, without reaching the global optimum of the problem.
Illustration: Forward TURF vs. Backward TURF
- Forward TURF: The portfolio is built by adding at each step the element that brings the most new customers. This method can quickly get stuck in a suboptimal solution, as some elements, though weak individually, may create strong synergy with others in the portfolio and offer better coverage when combined.
- Backward TURF: Conversely, elements with the least impact on coverage are successively removed to keep only a limited number of products. This approach can lead to another combination, sometimes more effective, but it also depends on the order of removal and may miss the global optimum. Although the backward TURF approach is technically functional, we do not recommend it as it is too influenced by the first iterations because of offers that are strongly rejected by respondents.
These two approaches do not necessarily yield the same optimal solutions in the end.
In both cases, the solution found depends heavily on the order in which elements are selected or removed, which exposes you to the risk of missing the true optimal combination.
The Only Solution: Explore All Combinations
To ensure you find the ideal solution(s), it is necessary to compute the impact for all possible combinations of elements, for each desired portfolio size. This exhaustive approach, although more computationally intensive, is the only way to ensure that the selected combinations truly maximize coverage in the target audience.
Computation Time: A Major Challenge for Exhaustiveness
When you want to compute all possible combinations in a TURF analysis, the issue of computation time quickly becomes central, especially as the number of offers increases. Take the example of a 20-product portfolio: the number of possible combinations to form assortments of size (k) is given by the combination formula:, where (n) is the total number of products.
-
For a single assortment of 10 products out of 20, there are already = 184 756 combinations to evaluate.
-
If you want to explore all assortment sizes (from 1 to 20 products), the total number of combinations to test is . That is exactly the size of an Excel sheet, including the names of the TURF output variables in the column headers.
This computation volume is considerable: for each combination, you need to compute the portfolio’s impact, which means reviewing all the data in the database. Even with powerful tools, processing can take several minutes to several hours depending on the size of the dataset and available computing power. Moreover, if the number of items to test exceeds 20, you cannot deliver all results in a single Excel data sheet.
“TURF analysis, when conducted exhaustively, requires computing power that grows exponentially with the number of products to combine.”
That is why, in practice, it is often necessary to optimize algorithms (parallelization, heuristics, pre-filtering of low-performing products) or to limit the analysis to realistic portfolio sizes for the customer. But it is essential to keep in mind that only an exhaustive approach guarantees not to miss the true optimal combination.
Towards Enhanced Reporting: Suggesting a Set of Candidate Solutions
In practice, it is rare that only one solution is relevant for the customer. Indeed, the final choice must take into account internal constraints (margin, production capacity, portfolio coherence, etc.) that are not integrated into the TURF algorithm. It is therefore preferable to:
- Present several candidate combinations to customers, corresponding to the best portfolios for different sizes.
- Allow the customer to evaluate these solutions knowing their own internal criteria (quality, profitability, feasibility, brand strategy, etc.).
“TURF analysis must deliver several solutions to customers, allowing them to knowingly choose the combination best suited to their objectives and internal constraints.”
The Stat & More Approach for Studying Portfolios with a Large Choice of Offers (>20)
To conduct a TURF analysis with more than 20 offers, Stat & More offers a structured approach, combining statistical robustness and computational efficiency.
✓ Pre-filtering of offers: Initial selection based on objective criteria (minimum preference share, strategic relevance, elimination of offers rarely chosen). This reduction in the number of products or services to a relevant subset greatly reduces the computation load while preserving the representativeness of the analysis.✓ Use of heuristic algorithms like Modified greedy methods: Progressive construction of combinations by maximizing at each step the incremental impact of the portfolio, while testing several starting points to limit the risk of local optimum and at several portfolio depths.
✓ Exhaustive computation on subsets: For the most strategic portfolio sizes (e.g., 5 to 8 products), exhaustive exploration of all possible combinations on the filtered subset, to guarantee identification of the global optimum for these critical sizes. Partitioning algorithms can be used to form homogeneous groups of offers. From these groups, a flagship product or service is extracted from each group. This set then forms a coherent group to test and inject into the TURF algorithm.
✓ Parallelization and software optimization: Development of internal tools leveraging parallelization (distributed computation over several cores or servers) to speed up processing of large volumes of combinations. Automation of calculations and integration into functions allowing reproducibility of calculations, to ensure reliability and documentation of each calculation step.
In the end, we present several candidate solutions for each portfolio size, with associated metrics (reach, frequency, cannibalization where applicable).
“For large datasets, the use of hybrid methods (pre-filtering, heuristics, targeted exhaustive computation, software optimization) is essential to ensure TURF analysis is both robust and operational.”
Stat & More thus relies on methodological and technical expertise to offer its customers TURF analyses adapted to the complexity of their product or service portfolios, even when these far exceed 20 offers. We support our end customers in the final decision-making. We integrate their constraints (margin, logistics, brand image, production constraints, etc.) and combine them with the statistical results of the TURF analysis. This is the ultimate achievement in customer support, allowing us to make, together, the best decision regarding which product or service portfolio to bring to market. This is a complete, 360° TURF analysis that leaves no room for debate.
In this sense, TURF analysis is a co-construction process, working as a team with end customers to identify the optimal solution(s) in the customer’s environment.
Conclusion
There are indeed methods and algorithmic techniques that allow for identifying optimal portfolios even when the number of offers tested is relatively large. TURF analysis should not be limited to classic iterative reporting, which exposes you to the risk of a local optimum. To truly serve the interests of the customer commissioning a TURF analysis, it is essential to compute all possible combinations and deliver a set of candidate solutions, to be evaluated according to the customer’s internal data (price positioning, margin, logistics constraints, brand image, business strategy, etc.). Only then can TURF analysis fulfill its promise: help build the most effective and context-appropriate portfolio or offer for the customer.
To learn more about the TURF methodology and its applications, see our previous article on the Stat & More blog.
REFERENCES :
✔ (2024, November 24) Stat & More, Benoit Riou : MaxDiff et TURF https://statandmore.com/en/blog/sam-maxdiff_turf/
✔ (2022, July 25). Contributeurs aux projets Wikipedia : TURF analysis https://en.wikipedia.org/wiki/TURF_analysis
✔ (2025, May 15). Contributeurs aux projets Wikipedia : Cannibalization (marketing) https://en.wikipedia.org/wiki/Cannibalization_(marketing)
✔ Laramore, J. & SAS Institute. (n.d.). Optimizing Product Assortment with Total Unduplicated Reach and Frequency Analysis in SAS/OR. Paper SAS2981-2019. https://support.sas.com/resources/papers/proceedings19/2981-2019.pdf
✔ Rossi, A. (2023, August 11). Chapter 2 TURF | R Tools for Market Research. https://bookdown.org/rossialessio095/R_Market_Research/turf.html
✔ Serra, D. & Department of Economics and Business, UPF. (n.d.). A new model for designing a product line using TURF analysis. In Research Papers in Economics [Journal-article]. https://core.ac.uk/download/pdf/6431385.pdf
✔ (2025, May 18) Contributeurs aux projets Wikipedia. Algorithme glouton https://fr.wikipedia.org/wiki/Algorithme_glouton
✔ (2025a, January 2) Contributeurs aux projets Wikipedia. Partitionnement de données https://fr.wikipedia.org/wiki/Partitionnement_de_donn%C3%A9es