Editor's note: Chrissie Wells is head of quantitative methods at Leapfrog Research, London. She can be reached at firstname.lastname@example.org. This article appeared in the January 13, 2014, edition of Quirk's e-newsletter.
For many multinationals, global segmentations now seem to be a must-have. The advantages are obvious: enabling the organization to identify common targets across markets, with consequent efficiencies of marketing and management effort. So why are global segmentations so often rejected by the local offices, who will often favor a localized segmentation of their own market?
Exist across cultures
Let's start by considering why we believe global segmentations are relevant: It's because we believe that the same meaningful human variations exist across cultures. For example, let's consider health care. In all cultures people want to remain healthy. However, for some people personal health will be a core need, whereas for others it may be less of a priority compared, for instance, with affluence or immediate gratification. Some may trust in modern science and medicine, while others place more trust in traditional or homespun remedies. Some may believe that by their own actions they can improve or maintain their health, while others may hold a more fatalistic view. It follows that groups of consumers with common outlooks and needs can be identified across markets and that brands can exploit these by addressing the same core needs.
On the other hand, there are considerable cultural differences in exactly what being healthy means. Health care product categories can be at very different stages of development in different markets. There are huge differences in distribution channels, particularly for medicines - to say nothing of differing brand presence. It is also possible that the same brand may be perceived very differently from one market to another. These influences do not prevent global segmentations being relevant but they do mean that local adaptation and flexing of a global model is vital.
Right and wrong ways
Within companies there will always be an internal political dynamic. It is perhaps natural for local management to resent the interference of a global owner. Beyond that however, there are right and wrong ways of developing a global segmentation which can be implemented at a local level. These are the focus of this article.
To start with, there are six methodological steps that are essential for success in developing an effective global segmentation. These might appear to be basic or hygiene factors but can easily be overlooked in adhering to time and cost parameters.
1. Research consumer differences very thoroughly at a local level.
This is essential for a comprehensive investigation. In an ideal world, bespoke qualitative research would be commissioned to generate hypotheses about the ways in which consumers differ. If this is not possible then hypotheses can be based on analysis of previous work. Without these working hypotheses at a market level, important features of consumers in that market may be overlooked.
2. Involve local stakeholders from Day 1.
Local stakeholders need to understand what the segmentation is intended to achieve and how it will be used both globally and in their market. They need to be given an opportunity to comment on the working hypotheses, as well as the proposed sample definition and methodology in their market. This might involve teleconference sessions, with information being sent in advance and follow-up confirmation of points agreed.
3. Choose interview methods carefully.
While online interviewing has become the norm in most developed countries, there are many markets where only certain sectors of the population are accessible via online panels and others where online panels are practically nonexistent. So it may be necessary to use a mixed methodology but care should be taken to avoid interview mode effects. For example, a self-completion approach can be used within a personal interview for attitude batteries to more closely mirror the online experience.
4. Take great care with translation.
At minimum, translations should be carried out by a native speaker of the language translated to and checked by a second independent native speaker. If prior qualitative research has been carried out, the local qualitative moderator is an ideal candidate for this second check, as they will have recent experience of the exact way local consumers describe the category. And of course, the local client stakeholders must sign off the final translations.
5. Control for differences in response style.
Cultural differences in response style (e.g., the way attitude scales are used) are well-known. There are a number of ways of addressing this. We favor analytical approaches, such as latent class, that are independent of response style. Within a latent class analysis we can set up variables that represent individual response patterns (e.g., tendency to rate high or low on each statement) or tendency to use the extremes/middle. Using these as latent variables in the model sweeps up any difference in response style, leaving only true differences in response patterns.
6. Try a number of different ways of analyzing the data.
The ideal segmentation model maximizes differences between groups in both their needs and their behavior. For this reason, a number of approaches with different driver variables should be attempted. Examples might include core psychographic attitudes, category perceptions, expressed category needs, behavior, etc. - each alone or in combination.
If all of these steps are taken, there is a better chance that a single segmentation model will emerge that meets certain criteria at a global and local level. Such criteria might be:
Does it pass some basic sense checks?
Number of segments. Not too many to be unmanageable, not too few that utility is limited.
Size of segments. None too small to be useful, none too big.
Replicability. Statistical measures such as r-squared are a good indication of this.
Basic meaning. Do the segments make sense? Are they easy to explain? Do possible names spring to mind?
Does it deliver market insight?
- Does it explain key behaviors such as volume purchase, brand choice, premium or value brand orientation, etc.?
- Does it give a clear view of each segment's needs, including unmet needs?
Can it inform a clear action plan? Consider the following:
- strategic opportunities and potential marketing initiatives;
- priorities for product and/or service development;
- brand alignment;
- choice of channel; and
- communication of key messages and/or tone of voice.
Cannot be found
However, it is possible that despite all best practices, a segmentation solution cannot be found that works both globally and locally. Then what options do we have?
Option No. 1
The first option is to accept that the global solution may have more segments than is ideal but that only a limited number will be relevant for each individual market. Opinions vary on the ideal number of segments but conventional wisdom would suggest around seven, with three or four being deemed appropriate target groups.
Figure 1 shows a segmentation on a general health topic across six markets. At a global level, there are nine segments of significant size. However, no individual market includes meaningful numbers of all nine. For instance, in the U.S., segments 8 and 5 are of negligible size so there are effectively only seven segments in this market. Likewise, Brazil has only seven segments to consider and Russia only six.
In this particular case, global segments do emerge. Segments 9 and 4 proved to be significant in all global markets, allowing global initiatives to be addressed to them. Segments 8, 7, 6 and 3 are not present everywhere but exist in more than one market, thus being relevant at a regional and local market level. (Segments 1 and 2 were less engaged with the topic so were not regarded as target groups.)
Option No. 2
In another segmentation on a household product category, we did not find a good solution this way. The five markets involved proved to be very different in the way attitudes and needs were expressed and in key behavior indicators, such as the frequency with which products are used in different areas of the house. Figure 2 is typical of the output we observed.
Even with nine segments, nearly half of the Colombian respondents fell into a single segment and four out of 10 in Korea fell into another segment. While it was thought acceptable that some segments may effectively not exist in all markets and even that some might exist in one market only (e.g., Segment 7 in Russia), each market should include several segments of reasonable size with no single segment dominating. This was clearly not the case for Colombia and Korea.
We tried 16 different model approaches each with four-to-nine segment solutions. All of them had the flaw described above. Or if they did not, the segments emerging were not fit for purpose. For example, one solution had good behavioral discrimination but little attitudinal difference from one group to another.
The route to success in this case was to build the global segmentation from the bottom up. This four-step process took place in a two-day workshop involving agency researchers, client researchers, client stakeholders and a consultant statistician.
Step 1 (pre-workshop): A number of solutions were run separately for each market and the agency team selected the most promising two or three against agreed criteria.
Step 2: Workshop syndicates considered each market separately, chose one winning solution and presented it to the whole group.
Step 3: The group then mapped the segments, grouping together those from each market that had similar characteristics.
Step 4: The mapped segments were combined to form a new global model that was then checked conceptually and statistically.
The final model is shown in Figure 3. There are three truly-global segments: Segments 1, 3 and 6. Of these segments, 1 and 3 were designated global targets for product development. In addition, local markets have the option to include relevant cross-market groups (such as Segment 2) or market-specific segments for targeting specific local initiatives.
While successful in this case, the bottom-up approach will not always be appropriate. When the number of markets is more than five or six, it will become unwieldy.
Option No. 3
The third option is to use a statistical approach to removing variation between markets. This is possible within a latent class approach and is one example of the flexibility that this approach has over other methods. Market variation can be removed by including market as predictor in the model alongside the segment variable which is unknown. This effectively sweeps up the differences on the items down purely to country, leaving the segmentation to focus on other differences that apply across all markets.
An example of this is a 10-country segmentation of a widely-purchased category. A latent class segmentation method was applied, using a battery of attitude statements as drivers (Figure 4).
The total sample and the eight segments are profiled by market. Significant differences from the total sample are shown in green if significantly higher, orange if significantly lower. The extreme market skew in this solution is clear and discriminates between markets, not within (Figure 5).
In this case, the segments are much more evenly distributed by market but still showed distinct and meaningful attitudinal differences. A development of this solution formed part of the eventual segment model, which was based on a combination of broad attitudinal outlook with specific needs from the category.
Put into action
So once we have a segmentation that is both meaningful and actionable at a global and local level, what steps can be taken to ensure that it is put into action locally?
It goes without saying that the findings should be reported on local data. We saw earlier that the presence and size of segments can vary by market. Beyond that, while the segments will be defined similarly by their core attitudes and needs, their profile of behavior will vary noticeably at a market level. For example, in one project we identified a segment in every market defined by a greater concern, even an obsession, with having a spotlessly-clean home. However, exactly what that means in terms of how often each area of the house is cleaned - and with what tools and materials - is very different in, say, Brazil compared with Australia. Local management need to gain a deep understanding of who each segment are and what they do and why, in terms of their own market.
Face-to-face interaction remains the most compelling way of communication and segment deployment events, involving a mix of presentations, workshopping and hands-on working with the data, can be particularly effective. Segments need to be communicated at a local level in two ways.
A human level
First of all the segments need to be understood at a human level (i.e., pen portraits, videos, ethnographic data, etc.), even to the extent of having actual consumer representatives of each segment attend a workshop and answer questions. Personal involvement is also important so workshop members can be invited to guess their own or their colleagues' segment - or an algorithm (i.e., short question set) can be used to identify them.
A business level
The segments also need to be understood at a business level. Their size needs to be measured not just by volume but also of value based on their spend within the relevant category. Their local alignment with brands is clearly important in terms of their current loyalty to our brands or to competitors. However, the key task for local management is to understand the segments in terms of future potential. Which groups are most aligned with relevant trends in our category? While a global perspective can help here, the local team must have the chance to debate the issues and come to their own conclusions on which their action plans can be founded.
Tools and processes
Beyond the initial event, what tools and processes will help a segmentation to be used effectively? Again, we should think both at the human and the business level. At the human level, key segment characteristics can be captured on posters, mouse pads, screensavers, etc., and short films are a great way of communicating segments to a wider team.
At the business level, segments need to be integrated into the organization's metrics wherever possible. This may mean applying them to ongoing brand tracking or customer satisfaction metrics. This requires an easily-usable but accurate way of reproducing the segments via a short question set and algorithm (sometimes referred to as golden questions or typing tool).
The usefulness of these can be maximized in a number of ways. For example, as well as outputting a segment allocation, the algorithm can calculate a percent likelihood of that allocation being accurate. When recruiting to qualitative research, this can be used as a criterion so that only the most typical segment members are included. Another variation is the development of algorithms at a local level, rather than using a one-size-fits-all global version, which may be less accurate at a market level.
Think global, act local
From planning through to execution and final implementation, segmentation is an area where the adage "Think global, act local" truly represents the path to success.