Contour scores what twenty experience factors are worth to one kind of person in one market. Every score is the sum of four terms that can be read separately, so the number can always be explained to whoever asks where it came from.
Segment and market preference model · the panel, the coefficients and the market attributes are illustrative throughout · the two market-research figures quoted here are sourced and attributed on the page
Contour holds a panel of 12,000 illustrative respondents across nineteen markets. Each of them scores twenty experience factors from 1 to 10 for how much that factor matters to them. Averaged across everybody, those twenty scores are the global baseline, and every persona in the tool is measured as a distance from it.
The twenty baselines run from 4.70 to 7.20. The whole average consumer fits inside a quarter of the scale, which is what makes the baseline itself a weak brief. The useful signal is how far a real group sits away from it.
Scroll the chart sideways on a narrow screen. Mulberry marks the one factor followed through the rest of this page.
The persona controls express 30,240 combinations: seven age bands, nine employment types, four income groups, six household types, twenty market selections. Four of them sit within a tenth of a point of the global average on all twenty factors, and all four have no market, no age and no household set. Choose a real city and a real life stage and the average stops describing anybody.
What replaces it is one line of arithmetic, run once per factor:
The persona sets the three axis values. The factor carries its own three coefficients. The market contributes a tilt applied by topic. The rest of this page follows one factor, for one persona, one term at a time.
25 to 34 · higher income · living with children. Axes: life-stage +0.60, affluence +0.95, household +1.00.
Global baseline 5.70. Topic: Urbanisation. Coefficients +0.9 life-stage, +0.3 affluence, −0.6 household.
The rail is drawn at the same scale in every section below, so the marker can be compared by eye down the page.
The 25 to 34 band sets the life-stage axis to +0.60. Urban amenities and nightlife carries a coefficient of 0.9 there, the second highest of the twenty factors, so the term is 0.9 × 0.60. The marker moves right by just over half a point, which on this scale is a fifth of the entire baseline spread.
Affluence is assembled from income and employment together. Higher income contributes +1.00, the 25 to 34 band contributes −0.05, and employment is left open, so the axis reads +0.95. This factor’s affluence coefficient is 0.3, so the term is small and the marker barely moves.
Two terms in, and the group looks like it values nightlife more than the average consumer does.
Living with children sets the household axis to its maximum of +1.00. Urban amenities and nightlife carries −0.6 there, the strongest negative household coefficient anywhere in the model. The term is −0.60, and it is the first one that moves the marker back the way it came.
Hong Kong is described to the model by three attributes: development stage −0.20, technology optimism +0.50, sustainability focus +0.20. Each topic takes a different one at a different weight. Urbanisation factors take development stage at a fifth, so the tilt here is −0.04 and the market is almost silent.
On a technology factor the same market would have contributed +0.45, because Technology and AI takes the technology optimism attribute at nine tenths. The market matters a great deal or hardly at all depending on which factor is being scored, and the model is explicit about which.
Urban amenities and nightlife finishes at 5.89 against a baseline of 5.70. A gap of +0.19 on a ten-point scale, ranking eighteenth of twenty for this persona. Read that number on its own and the conclusion is that this group is indifferent to nightlife, and a strategy built on it would leave the subject alone.
The four terms say something more useful. Age and income together added +0.83. Household took −0.60 of it back and the market took −0.04 more. The score sits close to the average because two forces of real size cancelled, and the same near-average number would have been returned by a group that simply did not care either way. Those are different audiences and they call for different work.
This is the argument for the whole design. A clustered segmentation hands back a segment label. Contour hands back an arithmetic that can be read out loud in a meeting: when somebody asks why smart infrastructure scores 7.65 for this persona, the answer is 6.20 baseline, plus 0.42 for age, 0.48 for income, 0.10 for household and 0.45 for Hong Kong’s technology attribute.
Two things stop the arithmetic. The first is the clamp: scores are held between 1 and 10, so a persona at the extremes eventually stops moving. The largest divergence the model can produce is +2.83, on cost of living, for an 18 to 24 student on lower income living with children in India.
The second is the base size, and it is the one that bites. Contour multiplies the market’s respondent count by the population fraction of each persona choice. Every filter is a multiplication by a number smaller than one, so the base falls fast.
Logarithmic scale. The persona built through this page is the fifth row.
The worked persona resolves to 7 respondents. Contour withholds a score below five, so it still reports one. Market research has worked to a minimum of thirty for a long time, justified by the central limit theorem, and below that the convention is a footnote reading small base, interpret with caution. Measured against that convention the persona crossed the line two filters earlier, at 20.
The same persona is 69 people in the USA and 267 globally. The constraint is the size of the Hong Kong cell, which is a sampling decision rather than anything in the model. Reading Contour properly on Hong Kong means buying more Hong Kong, and the tool is built to say so on screen while the persona is being assembled.
Both personas below sit above Contour’s reporting floor, and both are ordinary descriptions of a Hong Kong resident.
| Rank | A · 25 to 34, higher income | Score | B · 55 to 64, lower income | Score |
|---|---|---|---|---|
| 1 | Smart infrastructure & connectivity | 7.54 | Cost of living & cheaper options | 7.64 |
| 2 | Digital city services & apps | 7.37 | Affordability of housing | 7.54 |
| 3 | Green & public spaces | 7.34 | Safety & local development | 6.71 |
| base | respondents in the cell | 20 | respondents in the cell | 9 |
Illustrative model output. Scroll the table sideways on a narrow screen.
They share nothing in their top three. On luxury and exclusivity they sit 2.73 points apart, which is wider than the entire 2.50-point spread of the twenty global baselines. One number for Hong Kong would have to sit between them and would describe neither.
Solid marks a score above the global average. Hatched marks one below it, the way a contour map hatches a hollow.
Harvard Business Review research, quoted by Greenbook in December 2025, puts the gap plainly: 81% of businesses call segmentation critical for growth and 25% believe they use it well. The usual failure is activation. The segmentation is delivered as a report, and the report never reaches a brief, a media plan or a database.
Contour is built as an instrument. Nothing is precomputed. The persona is assembled in front of whoever is asking, and the ranking, the divergence and the base size all move while they watch it happen. The decomposition is what makes that survive the meeting, because every number on screen can be traced back to the four terms that produced it.
The engine does not need rewriting to run on real data. Three inputs are synthetic, and those three are the whole of what a live version would replace: the twenty factor baselines, the persona coefficients, swapped for grouped benchmark scores by characteristic, and the per-market respondent counts. The interaction, the ranking, the divergence and the suppression logic stay as built.