TURF Analysis
Choose the set that reaches the most people, not the set with the highest individual scores. The three most popular items often appeal to the very same customers.
Overview & learning objectives
You run a small ice-cream company and you have just won space in a grocery chain's freezers — but only for three flavors. You survey 320 shoppers and the results look easy: chocolate, vanilla and cookie dough score highest. So you launch those three.
Six months later sales are disappointing, and here is why. Most of the people who would buy cookie dough were already going to buy chocolate. You did not add customers with that third slot; you gave the same customers another thing to pick. Meanwhile mango sorbet scored mid-table, but the people who wanted it wanted nothing else on your list — and you left all of them on the table.
TURF (Total Unduplicated Reach and Frequency) is how you avoid that. Instead of ranking items one at a time, it scores every possible combination and asks a different question: how many different people does this set reach? An item earns its place by bringing someone new, not by being popular.
🎯 What You’ll Learn
- Why the top three items are usually the wrong three. Popularity is measured one item at a time; reach is a property of the set.
- What “reach” actually counts. The share of people who would take at least one thing from your set. Nobody is counted twice, however many items they like.
- That you have to define “would buy” before you can answer. Does a 4 out of 5 count, or only a 5? Change that and the winning set can change with it, which is why this tool makes you choose.
- Where adding items stops being worth it. Reach curves flatten. At some point a new flavor only sells more to people you already had.
- That the best set is often not unique. Several combinations frequently tie within a rounding error, and then the tie is broken by cost, logistics or brand fit — not by the data.
🤔 Try this:
- Load Frost & Fold Creamery below and leave the portfolio size at 3.
- Compare the two headline numbers: reach of the best set against reach of the most popular. The gap is what picking by popularity would have cost you, in percentage points of the market.
- Now look at Top portfolios. Is the winning set made of the three highest-scoring flavors? It almost never is — and the item that replaced a favourite is the one that brought new people.
Step 1 Choose your data
Each case is a real portfolio decision with a fixed number of slots. Start with Frost & Fold Creamery if this is your first time — twelve flavors, three freezer slots, and a clear overlap trap. Or upload your own survey: one row per respondent, one column per item.
Use a case study
Or upload your own
One row per respondent, one column per item. Coming from MaxDiff? Upload
maxdiff_respondent_scores.csv as-is.
Drop a CSV here
Three shapes are accepted and detected automatically: MaxDiff scores
(score_ columns), rating scales, or 0/1 acceptance. Optional
weight and segment columns are picked up if present.
Is this right?
Detection reads the values, not the column names, so it can be wrong about an unusual file. Changing this changes what counts as “reached”.
Step 2 How many can you offer?
Two decisions, and the second one matters more than it looks. Portfolio size is how many slots you actually have. What counts as reached is where you draw the line between “would buy” and “merely does not hate it” — and moving that line can change which set wins. Decide it on business grounds before you look at the answer, not after.
The number of items you can actually launch. Shelf space, budget or contract terms usually decide this before the analysis starts.
This is a modelling decision, not a detail. Settle it before you look at the answer.
Constraints — lock items in or rule them out
A contract, a supply limit or a brand rule can decide part of the set before the data does. Reach can only fall when you force a choice; that drop is what the constraint costs.
Step 3 Read the answer
Start with the box below, which says in one sentence which set to launch and what it buys you over the obvious choice. Everything after it is the evidence: which items earn their slot, how much is left to gain, and whether a different judgement call would have given you a different answer.
Most popular vs. widest reach
Two sets of the same size, side by side. Popular is what you get by ranking items and taking the top few. Widest reach is the set TURF chose. The height of each bar is the share of people who would take at least one item from that set — so the difference between the bars is the customers that picking by popularity would have missed.
📊 How to Read This Chart
- A big gap means the popular items overlap heavily: they are competing for the same people. This is the classic TURF finding.
- Almost no gap is a real and useful result too. It means the popular items already reach distinct audiences, and ranking would have served you fine. Report that rather than inventing a story.
- The bars count people, not purchases. Somebody who would take four of your items still counts once.
What to watch
Top portfolios
The best combinations, ranked by reach. Look at the top few rows together, not just the first one. If several sets land within a point of each other they are effectively tied, and the choice between them belongs to cost, supply or brand fit rather than to this table.
| # | Items | Reach | Frequency | Notes |
|---|
Frequency is the average number of these items a reached respondent would take. It is averaged over reached respondents only, so it never falls below 1.
Would a different rule change the answer?
You had to decide what counts as “would buy”, and that was a judgement call. This re-runs everything under the neighbouring rules. If the same set wins throughout, your answer is robust and you can say so. If the winner moves, your recommendation depends on a line you drew — and that belongs in the write-up rather than hidden in a footnote.
| Rule | Best set under that rule | Reach | Same answer? |
|---|
What each item adds
Each bar is the new people that item brought in, on top of everything already chosen before it. A tall first bar and short ones after it is normal: the first item has the whole market to win, and every later item can only pick up whoever is left.
Items are added in the order that grows reach fastest, so each bar is what that item contributed on top of the ones before it, not its reach on its own.
📊 How to Read This Chart
- A near-zero bar is an item paying for shelf space without bringing anybody new. It is the first candidate to cut.
- Order matters. An item looks weak here partly because of what came before it. Swap the order and the same item could look strong, which is why you judge the set, not the ranking.
- This is the chart to use when someone asks “why is that in the lineup?” — the bar is the answer.
How many do you need?
Reach as you allow yourself more slots. The elbow is the decision. Before it, each extra item is still winning new customers; after it, you are mostly paying to sell more to people you already reach.
Where the line flattens, extra items stop buying new people and start buying frequency from people you already had.
📊 How to Read This Chart
- Find where the line bends and compare it with the number of slots you actually have. If the bend is below your budget, you can launch fewer items and lose almost nothing.
- A line still climbing at the right-hand edge means your market is more varied than your shortlist — more items would still be finding new people.
- The curve can never go down. Adding an item cannot take a customer away, so a flat stretch means “no new people”, not “worse”.
Who overlaps with whom
This is the chart that explains why the popular set loses. A dark cell means the two items are chasing the same people, so putting both in your lineup buys you very little. A pale cell means they appeal to different crowds — which is exactly the pairing that grows reach.
Each cell reads: of the people who accept the row item, what share also accept the column item. It is deliberately asymmetric, because a small item can sit entirely inside a large one without the reverse being true.
📊 How to Read This Chart
- Read along a row: “of everyone who likes this item, what share also likes that one?”
- The grid is not symmetric, and that is the point. A niche item can be almost entirely contained inside a popular one — 90% one way, 15% the other. Only the first number means “this item is redundant”.
- Look for a pale row. An item that overlaps with nothing has an audience all of its own, and it will usually appear in the winning set even with an unremarkable score.
Does it work for everyone?
The same chosen set, scored separately inside each group in your data. An average hides this: a set can reach 80% overall and still reach barely half of the one segment you were trying to win. A large negative gap is a reason to revisit the recommendation, even though the total looks healthy.
| Group | n | Reach of the chosen set | Gap vs overall |
|---|
A set can look healthy in total and still miss the group you most need. Size is not the same as importance.
Methods and results
Managerial interpretation
Take it with you
Exploratory questions
All of these are answerable with the controls above. The last two are the ones worth arguing about in class.
- How much does popularity cost you? With Frost & Fold at three slots, compare the reach of the best set against the reach of the most popular. Multiply that gap by a market of 50,000 shoppers. How many people is it?
- Which item is carrying the set? In What each item adds, find the item with the smallest bar. If you dropped it for the next-best candidate, how much reach would you lose — and would anyone notice?
- When should you stop? Use How many do you need? to find the elbow. If each extra flavor costs $40,000 a year in production, where would you stop, and what reach would you be giving up?
- Does the answer survive your judgement call? Change what counts as reached from a 4-or-5 to a 5-only. Does the winning set change? What would you tell a client who asks which rule is “correct”?
- Is a tie really a tie? Find two portfolios within half a point of each other in Top portfolios. On what grounds would you choose between them, given the data cannot?
- Who are you leaving out? Load CivicPulse and look at Does it work for everyone? A set can reach most of the population and still miss a group badly. When is the highest-reach set the wrong recommendation?