Testing the brand character on the target audience
Why testing the character is a separate task, not part of the design
Most articles about mascots end at the stage of confirming the image: drew it, coordinated it, launched it. Then the zone of uncertainty begins — does the character work for business metrics or just pleases the art director. An academic study of animated mascots showed a specific chain: the growth of the character's social presence correlates with audience engagement, which in turn is linked to the growth of the intention to revisit and the intention to purchase, according to a study published in MDPI. That is, between 'the character is beautiful' and 'the character sells,' there are several measurable steps, and each of them can and should be tested separately.
At what stage to test and when it's already too late
Testing the character is not a one-time procedure after launch, but a series of checks at different stages of development.
- Concept stage (sketches, 2-3 character options). It's cheaper to make a mistake and cheaper to fix it here. Suitable methods are quality ones — focus groups, in-depth interviews.
- Prototype (ready visual, draft usage scenarios). First quantitative reactions can already be measured: recognizability, associations, emotional response.
- Pilot launch in one channel. A/B-test on real traffic — comparison of creatives with a character and without.
- After scaling. It is too late to change the character or visual code without reputational losses — one can only fine-tune the context of use.
The rule is simple: the earlier a problem is found, the cheaper it is to solve. Changing the 3D-model after a character has been fixed in all contact points — as happened with SberKot when transitioning from 2D to 3D — is possible, but it is an expensive and risky step that is only justified by significant business indicators.
Methods and tools: what to choose for your task
There is no single 'correct' method — the set of tools depends on the budget, timeframe, and which specific question you are addressing.
| Method | What it provides | Example of application |
|---|---|---|
| Focus groups | Quality response, verbalization of associations and character | Character testing for a cartoon in three age segments (4-6, 7-9, 10-12 years) with a request to describe the character and provide associations |
| Online survey + in-depth interviews | Massive quantitative evaluation plus depth of understanding motivations | The comprehensive character test included quantitative surveys and in-depth interviews, with the data analyzed and systematized |
| Eye tracking | Checking whether the audience notices the character physically and within what time frame | Eye tracking on specialized equipment allows measuring the speed of character fixation and assessing whether users notice it during the critical first seconds of viewing |
| Neurotesting | Emotional reaction at a level not consciously recognized by the respondent | SberKot was developed using research in the field of neuroscience and social psychology |
| A/B test in a channel | Direct impact on business metrics: clicks, adding to cart, purchases | Monarch 3v1: personalized AI creatives delivered a 5 times higher cart conversion than control |
| Multi-criteria creative evaluation | Comparison of options based on a set of parameters before launch | Cross-test evaluated creatives by 13 criteria on the Yandex Glance platform, ratings 4,39 and 4,24 matched the final VTR |
Good practice is to combine methods rather than choose just one. Aitracking will show that a character was noticed, but won't explain whether it inspires trust. A survey will provide numbers on recognizability, but won't show what happens at the emotional level. Neurotesting and in-depth interviews close this gap
Metrics and KPIs: from awareness to conversion
The mistake of many teams is measuring only 'like/dislike'. A character must work throughout the funnel, so metrics should also be structured by levels
| Funnel level | Metric | How to measure |
|---|---|---|
| Awareness | Character and brand recognizability | Survey, brainstorming of associations; as a result of the campaign, SberKot's recognizability among youth increased by 15 points |
| Attention | Speed and probability of gaze fixation on the character | Aitracking, thermal maps |
| Ratio | Loyalty, trust, emotional response | Neurotest, deep interviews; Sber loyalty increased 1,5 times |
| Engagement | Viewing time, engagement with content, engagement | Channel analytics; SberKot's ad engagement increased by 3 times |
| Intention | Intention to buy, brand association, consideration for purchase | A/B testing; with Monarch 3v1, brand association increased by 91%, consideration for purchase — by 36% |
| Conversion | Adding to cart, purchasing, repeat visit | Direct CRM/analytics data; increase in intent to revisit by 29%, purchases — by 33% |
Looking only at top-line metrics — a typical mistake. A character can be easily remembered and yet not move sales at all, and vice versa: a quiet, not the brightest image sometimes yields better conversion precisely because it doesn't distract from the product.
How to design an A/B test for a character without false conclusions
The main technical complexity of the mascot A/B test is to separate the character's effect from the effect of the text, design, publication time, or ad placement segment. Without this separation, the test results will prove nothing.
- Formulate a hypothesis in measurable terms: 'a creative with a mascot will yield a CTR higher by X%', not 'the mascot should be liked'.
- Create two scenario variants — with and without a mascot — keeping the text, design, offer, time, and channel of display identical. Only one variable changes.
- Fix the minimum sample size in advance, before launching, and do not adjust it to fit a convenient result afterward.
- Choose the test duration — in practice, two weeks is sufficient for a quick check of a mascot, provided traffic is stable and not skewed by seasonal peaks.
- Randomly split the traffic, not by convenient segments — otherwise, the difference in results might be explained by audience differences, not the mascot.
- Count not only clicks, but also lower funnel metrics — conversion and repeat actions — because top-line metrics can sometimes be misleading.
The Perekrestok case clearly illustrates why this is needed: a preliminary evaluation of creatives against 13 criteria almost exactly matched the final 13.6% VTR gap between the versions. This means a well-designed preliminary test can actually predict audience behaviour in a real campaign — saving budget that would otherwise have been spent on a weak version.
Audience segmentation: why one character does not work equally for everyone
Different audience groups react differently to the same image, and ignoring this means testing an averaged response that does not match the reaction of any real segment. In the test of animated characters, three age groups — 4-6, 7-9 and 10-12 years — gave noticeably different feedback in terms of character and associations within the same focus group, which allowed us to adjust the image for each audience layer separately, rather than seeking a compromise.
Practical conclusion: testing should not be done for the character 'as a whole', but for the character in specific segments — by age, communication channel, stage of brand familiarity. The example of Monarch 3v1 shows the economic effect of this approach: AI creatives personalized for segments gave 57% higher memorability than the control group and conversion to adding to the cart 5 times higher. One universal character variant almost always loses to several variants adapted for segments, even if the basic image remains unchanged.
How to make decisions based on test results
Data without interpretation rules turns into a reason for disputes, not for decisions. To avoid this, it is worth establishing a simple protocol.
- The alignment of signal directions is more important than absolute figures: if surveys, heatmaps, and A/B tests all show the same direction — the results can be trusted with high confidence.
- A difference in evaluations that is less than the statistical error is not considered a win for one variant — either a larger sample size is needed, or a repeat test.
- Upper funnel metrics (awareness, likability) do not replace conversion metrics — final decisions should be based on lower indicators if the business goal is sales.
- Any divergence in results between segments is not a reason to average them, but a signal to develop variability in how the character is presented to each segment.
Character testing pays off not when the team gets confirmation of their sympathy for the character, but when the data allows unambiguous choice between variants and predicts the effect before a large-scale launch. This is what distinguishes a systematic approach from intuitive mascot design.
Why testing the character is a separate task, not part of the design
Most articles about mascots end at the stage of confirming the image: drew it, coordinated it, launched it. Then the zone of uncertainty begins — does the character work for business metrics or just pleases the art director. An academic study of animated mascots showed a specific chain: the growth of the character's social presence correlates with audience engagement, which in turn is linked to the growth of the intention to revisit and the intention to purchase, according to a study published in MDPI. That is, between 'the character is beautiful' and 'the character sells,' there are several measurable steps, and each of them can and should be tested separately.
At what stage to test and when it's already too late
Testing the character is not a one-time procedure after launch, but a series of checks at different stages of development.
- Concept stage (sketches, 2-3 character options). It's cheaper to make a mistake and cheaper to fix it here. Suitable methods are quality ones — focus groups, in-depth interviews.
- Prototype (ready visual, draft usage scenarios). First quantitative reactions can already be measured: recognizability, associations, emotional response.
- Pilot launch in one channel. A/B-test on real traffic — comparison of creatives with a character and without.
- After scaling. It is too late to change the character or visual code without reputational losses — one can only fine-tune the context of use.
The rule is simple: the earlier a problem is found, the cheaper it is to solve. Changing the 3D-model after a character has been fixed in all contact points — as happened with SberKot when transitioning from 2D to 3D — is possible, but it is an expensive and risky step that is only justified by significant business indicators.
Methods and tools: what to choose for your task
There is no single 'correct' method — the set of tools depends on the budget, timeframe, and which specific question you are addressing.
| Method | What it provides | Example of application |
|---|---|---|
| Focus groups | Quality response, verbalization of associations and character | Character testing for a cartoon in three age segments (4-6, 7-9, 10-12 years) with a request to describe the character and provide associations |
| Online survey + in-depth interviews | Massive quantitative evaluation plus depth of understanding motivations | The comprehensive character test included quantitative surveys and in-depth interviews, with the data analyzed and systematized |
| Eye tracking | Checking whether the audience notices the character physically and within what time frame | Eye tracking on specialized equipment allows measuring the speed of character fixation and assessing whether users notice it during the critical first seconds of viewing |
| Neurotesting | Emotional reaction at a level not consciously recognized by the respondent | SberKot was developed using research in the field of neuroscience and social psychology |
| A/B test in a channel | Direct impact on business metrics: clicks, adding to cart, purchases | Monarch 3v1: personalized AI creatives delivered a 5 times higher cart conversion than control |
| Multi-criteria creative evaluation | Comparison of options based on a set of parameters before launch | Cross-test evaluated creatives by 13 criteria on the Yandex Glance platform, ratings 4,39 and 4,24 matched the final VTR |
Good practice is to combine methods rather than choose just one. Aitracking will show that a character was noticed, but won't explain whether it inspires trust. A survey will provide numbers on recognizability, but won't show what happens at the emotional level. Neurotesting and in-depth interviews close this gap
Metrics and KPIs: from awareness to conversion
The mistake of many teams is measuring only 'like/dislike'. A character must work throughout the funnel, so metrics should also be structured by levels
| Funnel level | Metric | How to measure |
|---|---|---|
| Awareness | Character and brand recognizability | Survey, brainstorming of associations; as a result of the campaign, SberKot's recognizability among youth increased by 15 points |
| Attention | Speed and probability of gaze fixation on the character | Aitracking, thermal maps |
| Ratio | Loyalty, trust, emotional response | Neurotest, deep interviews; Sber loyalty increased 1,5 times |
| Engagement | Viewing time, engagement with content, engagement | Channel analytics; SberKot's ad engagement increased by 3 times |
| Intention | Intention to buy, brand association, consideration for purchase | A/B testing; with Monarch 3v1, brand association increased by 91%, consideration for purchase — by 36% |
| Conversion | Adding to cart, purchasing, repeat visit | Direct CRM/analytics data; increase in intent to revisit by 29%, purchases — by 33% |
Looking only at top-line metrics — a typical mistake. A character can be easily remembered and yet not move sales at all, and vice versa: a quiet, not the brightest image sometimes yields better conversion precisely because it doesn't distract from the product.
How to design an A/B test for a character without false conclusions
The main technical complexity of the mascot A/B test is to separate the character's effect from the effect of the text, design, publication time, or ad placement segment. Without this separation, the test results will prove nothing.
- Formulate a hypothesis in measurable terms: 'a creative with a mascot will yield a CTR higher by X%', not 'the mascot should be liked'.
- Create two scenario variants — with and without a mascot — keeping the text, design, offer, time, and channel of display identical. Only one variable changes.
- Fix the minimum sample size in advance, before launching, and do not adjust it to fit a convenient result afterward.
- Choose the test duration — in practice, two weeks is sufficient for a quick check of a mascot, provided traffic is stable and not skewed by seasonal peaks.
- Randomly split the traffic, not by convenient segments — otherwise, the difference in results might be explained by audience differences, not the mascot.
- Count not only clicks, but also lower funnel metrics — conversion and repeat actions — because top-line metrics can sometimes be misleading.
The Perekrestok case clearly illustrates why this is needed: a preliminary evaluation of creatives against 13 criteria almost exactly matched the final 13.6% VTR gap between the versions. This means a well-designed preliminary test can actually predict audience behaviour in a real campaign — saving budget that would otherwise have been spent on a weak version.
Audience segmentation: why one character does not work equally for everyone
Different audience groups react differently to the same image, and ignoring this means testing an averaged response that does not match the reaction of any real segment. In the test of animated characters, three age groups — 4-6, 7-9 and 10-12 years — gave noticeably different feedback in terms of character and associations within the same focus group, which allowed us to adjust the image for each audience layer separately, rather than seeking a compromise.
Practical conclusion: testing should not be done for the character 'as a whole', but for the character in specific segments — by age, communication channel, stage of brand familiarity. The example of Monarch 3v1 shows the economic effect of this approach: AI creatives personalized for segments gave 57% higher memorability than the control group and conversion to adding to the cart 5 times higher. One universal character variant almost always loses to several variants adapted for segments, even if the basic image remains unchanged.
How to make decisions based on test results
Data without interpretation rules turns into a reason for disputes, not for decisions. To avoid this, it is worth establishing a simple protocol.
- The alignment of signal directions is more important than absolute figures: if surveys, heatmaps, and A/B tests all show the same direction — the results can be trusted with high confidence.
- A difference in evaluations that is less than the statistical error is not considered a win for one variant — either a larger sample size is needed, or a repeat test.
- Upper funnel metrics (awareness, likability) do not replace conversion metrics — final decisions should be based on lower indicators if the business goal is sales.
- Any divergence in results between segments is not a reason to average them, but a signal to develop variability in how the character is presented to each segment.
Character testing pays off not when the team gets confirmation of their sympathy for the character, but when the data allows unambiguous choice between variants and predicts the effect before a large-scale launch. This is what distinguishes a systematic approach from intuitive mascot design.
Recent articles
MegaFon, BMW, VTB and more 240+ companies. See how it looks in the showcase reel:
Our branding agency is approached to order:
Branding agency services
