Brand character in chat bots and automated customer support
Why does a business need a personality in a bot: numbers, not feelings
Talks about «a bot's empathy» sound abstract until you apply them to performance. The AI agent of the brand Feastables (MrBeast project) with a defined personality generates 6,5% of the company's total revenue — that is 6,24 million dollars, while the average order value with its participation is higher by 20% and NPS stays at 52, which is 10 points above the average e-commerce indicator. The chat bot «Chernogolovka» under the character Cola Cosmic Taro achieved 78% repeat usage — against the norm for FMCG 15–25%. The difference is not in text recognition technology, but in the fact that the customer returns to someone they were not opposed to talking to.
SberKot is a rare example of long-distance engagement: over 8 years, the character has gathered 13 million active subscribers in a chat-bot and 4 million in a separate community, known to 80 million Russians, with every second bank customer having opted for a card featuring his image. The brand's ER communication has grown 2,1 times, and loyalty — 1,5. This is no longer a metric of engagement, but a direct contribution to customer acquisition and retention.
The character versus the faceless bot: where does the irritation come from
53% of Russians are irritated by chat-bot communication, and 25% of customers outright refuse to purchase after receiving a 'scripted' response. The issue is not with the automation itself — the issue is with fake care. The phrase 'we're always happy to help' after the third incomprehensible request reads as mockery, not politeness. The customer's brain reacts not to words, but to incongruity: the system promises dialogue, but delivers a menu tree.
The character works differently: he honestly indicates his boundaries and rules of engagement. The client understands who he's speaking with and what to expect — and this reduces anxiety better than any apology. That's where the fundamental difference lies:
| Parameter | Impersonal bot-script | Bot with a character |
|---|---|---|
| Tone of communication | neutral-office, identical for all scenarios | stable personality, recognizable speech manner |
| Response to an error | standard apology, repeating the same answer | acknowledging boundaries, explicitly offering to call a person |
| Client expectations | underestimated, so failures are less noticeable | overestimated — if the character is 'human-like', the error hits harder |
| Effect on failure | moderate irritation | sharp disappointment, if the mask slips |
Precisely because of this paradox, 33% of users of financial services negatively evaluate AI assistants, although formally bots have become 'smarter' and 'more human' over the past few years. Only 33% of Russians in general had a positive experience communicating with smart chatbots — and this is not a problem of speech recognition technology, but a problem of mismatch between promise and result.
One character — four different channels: why copy-paste doesn't work
The mistake that almost everyone makes who has 'already created a character' is to transfer it unchanged from Telegram to the website widget and the IVR call center. Each channel requires its own response speed, attention format, and acceptable degree of 'character'.
| Channel | What the character determines | Key metrics | Format feature |
|---|---|---|---|
| Telegram bot | retention, regular contact | DAU, retention, share of repeat sessions | allows humor, memes, long conversation — example: Beauty Bomb, 5,5 million views, 2x exceeding the plan |
| Website widget | removing the barrier before purchase | conversion to cart, CTR, time to first reaction | the character must be concise: at Monarchy, using AI creative, conversion increased by 5 times, CPM dropped by 22% |
| Social networks / content avatar | brand reach and memorability | reach, subscribers, association with a slogan | the format of 'the character lives and uses the product', not advertising — an avatar for an insecticide gave 2,5 million reach and 15 000 subscribers in one month |
| Call center / IVR | reduction of anxiety around complex topics | Containment Rate, the share of escalations, CSAT | a more restrained tone, a minimalistic character — in banks, Containment ~30% is maintained, but this is not a guarantee of trust |
A telling example of format change is the AI character for a book service that significantly exceeded planned metrics for visits and conversion of dialogues into purchases. The character on the Rafinad platform increased engagement by 92,5% and doubled average reach — but this is a content channel metric, not a support indicator, and comparing it to the Containment Rate of the call center makes no sense.
The task defines the degree of 'humanity', not the other way around
A character for sales can flirt with the audience and joke — as in the case of Beauty Bomb with the 'Main Heroine' bot that collected 1,5 million messages. A support character must above all be predictable. And a character for financial consultations is a rather questionable territory: automation works well as a navigational layer for typical questions, but explaining loan terms requires a live person, because the client values the feeling of control rather than wit. Eight out of ten bank customers still contact an operator even after a bot's response — due to model hallucinations, when the answer sounds confident but is incorrect.
Expensive mistakes: when 'humanization' works against the brand
- A knowledge base with no content. A character with personality but without facts produces confident hallucinations — and trust is lost faster than with a neutral bot, because people expect competence from 'personality'.
- Too similar to a human. 57% of Russians cannot distinguish a bot from a human in chat — and disappointment occurs precisely at the moment of truth being revealed, not because of the fact of communicating with AI.
- Template phrases of politeness. «glad you to help» after an unresolved issue reads as ignoring rather than caring — this literally amplifies the irritation that the character must release.
- The character in a theme where control is needed, not emotion. Financial consultations, legal issues, complaints about fund deductions — here the bot's excessive 'friendliness' is perceived as evading responsibility.
- A different character at different stages of the funnel. If a Telegram bot makes jokes, but the website suddenly becomes dry and formal — the brand loses integrity, and this is more noticeable than the absence of a character altogether.
General principle: a character is harmful where the stakes are higher than the cost of one dialog error. If the topic involves client money, health, or legal risk, the right step is not 'make the bot more friendly,' but honestly acknowledge the limits of automation.
Handoff: the moment when the character must silently yield to a human.
Quiet dialog handoff is not a 'call an operator' button after three failed attempts, but a pre-defined trigger. Practice shows three reliable moments for handoff:
- The client rephrased the same question twice — this means the character did not understand the core, and a third attempt will only increase irritation.
- The topic involves money, health, or legal risk — here the client needs control, not wit, and an operator reduces anxiety better than any script.
- The client directly asks for a person — this request cannot be handled by retrying automation, even if the bot 'almost guessed' the answer.
The main metric trap here is Containment Rate. In banks, it stays around 30%, and on paper, this looks like automation success. But if a client finds the live chat button themselves after five failed bot attempts, the metric won't see it — it will count the dialog as 'closed by bot,' although in fact the client spent extra time and lost trust. The correct handoff metric is not the fact of transfer, but the number of attempts by the character before it: the fewer, the cleaner the transition and the lower the chance that the client will remember the brand as 'the one where the bot was dumb.'
Why does a business need a personality in a bot: numbers, not feelings
Talks about «a bot's empathy» sound abstract until you apply them to performance. The AI agent of the brand Feastables (MrBeast project) with a defined personality generates 6,5% of the company's total revenue — that is 6,24 million dollars, while the average order value with its participation is higher by 20% and NPS stays at 52, which is 10 points above the average e-commerce indicator. The chat bot «Chernogolovka» under the character Cola Cosmic Taro achieved 78% repeat usage — against the norm for FMCG 15–25%. The difference is not in text recognition technology, but in the fact that the customer returns to someone they were not opposed to talking to.
SberKot is a rare example of long-distance engagement: over 8 years, the character has gathered 13 million active subscribers in a chat-bot and 4 million in a separate community, known to 80 million Russians, with every second bank customer having opted for a card featuring his image. The brand's ER communication has grown 2,1 times, and loyalty — 1,5. This is no longer a metric of engagement, but a direct contribution to customer acquisition and retention.
The character versus the faceless bot: where does the irritation come from
53% of Russians are irritated by chat-bot communication, and 25% of customers outright refuse to purchase after receiving a 'scripted' response. The issue is not with the automation itself — the issue is with fake care. The phrase 'we're always happy to help' after the third incomprehensible request reads as mockery, not politeness. The customer's brain reacts not to words, but to incongruity: the system promises dialogue, but delivers a menu tree.
The character works differently: he honestly indicates his boundaries and rules of engagement. The client understands who he's speaking with and what to expect — and this reduces anxiety better than any apology. That's where the fundamental difference lies:
| Parameter | Impersonal bot-script | Bot with a character |
|---|---|---|
| Tone of communication | neutral-office, identical for all scenarios | stable personality, recognizable speech manner |
| Response to an error | standard apology, repeating the same answer | acknowledging boundaries, explicitly offering to call a person |
| Client expectations | underestimated, so failures are less noticeable | overestimated — if the character is 'human-like', the error hits harder |
| Effect on failure | moderate irritation | sharp disappointment, if the mask slips |
Precisely because of this paradox, 33% of users of financial services negatively evaluate AI assistants, although formally bots have become 'smarter' and 'more human' over the past few years. Only 33% of Russians in general had a positive experience communicating with smart chatbots — and this is not a problem of speech recognition technology, but a problem of mismatch between promise and result.
One character — four different channels: why copy-paste doesn't work
The mistake that almost everyone makes who has 'already created a character' is to transfer it unchanged from Telegram to the website widget and the IVR call center. Each channel requires its own response speed, attention format, and acceptable degree of 'character'.
| Channel | What the character determines | Key metrics | Format feature |
|---|---|---|---|
| Telegram bot | retention, regular contact | DAU, retention, share of repeat sessions | allows humor, memes, long conversation — example: Beauty Bomb, 5,5 million views, 2x exceeding the plan |
| Website widget | removing the barrier before purchase | conversion to cart, CTR, time to first reaction | the character must be concise: at Monarchy, using AI creative, conversion increased by 5 times, CPM dropped by 22% |
| Social networks / content avatar | brand reach and memorability | reach, subscribers, association with a slogan | the format of 'the character lives and uses the product', not advertising — an avatar for an insecticide gave 2,5 million reach and 15 000 subscribers in one month |
| Call center / IVR | reduction of anxiety around complex topics | Containment Rate, the share of escalations, CSAT | a more restrained tone, a minimalistic character — in banks, Containment ~30% is maintained, but this is not a guarantee of trust |
A telling example of format change is the AI character for a book service that significantly exceeded planned metrics for visits and conversion of dialogues into purchases. The character on the Rafinad platform increased engagement by 92,5% and doubled average reach — but this is a content channel metric, not a support indicator, and comparing it to the Containment Rate of the call center makes no sense.
The task defines the degree of 'humanity', not the other way around
A character for sales can flirt with the audience and joke — as in the case of Beauty Bomb with the 'Main Heroine' bot that collected 1,5 million messages. A support character must above all be predictable. And a character for financial consultations is a rather questionable territory: automation works well as a navigational layer for typical questions, but explaining loan terms requires a live person, because the client values the feeling of control rather than wit. Eight out of ten bank customers still contact an operator even after a bot's response — due to model hallucinations, when the answer sounds confident but is incorrect.
Expensive mistakes: when 'humanization' works against the brand
- A knowledge base with no content. A character with personality but without facts produces confident hallucinations — and trust is lost faster than with a neutral bot, because people expect competence from 'personality'.
- Too similar to a human. 57% of Russians cannot distinguish a bot from a human in chat — and disappointment occurs precisely at the moment of truth being revealed, not because of the fact of communicating with AI.
- Template phrases of politeness. «glad you to help» after an unresolved issue reads as ignoring rather than caring — this literally amplifies the irritation that the character must release.
- The character in a theme where control is needed, not emotion. Financial consultations, legal issues, complaints about fund deductions — here the bot's excessive 'friendliness' is perceived as evading responsibility.
- A different character at different stages of the funnel. If a Telegram bot makes jokes, but the website suddenly becomes dry and formal — the brand loses integrity, and this is more noticeable than the absence of a character altogether.
General principle: a character is harmful where the stakes are higher than the cost of one dialog error. If the topic involves client money, health, or legal risk, the right step is not 'make the bot more friendly,' but honestly acknowledge the limits of automation.
Handoff: the moment when the character must silently yield to a human.
Quiet dialog handoff is not a 'call an operator' button after three failed attempts, but a pre-defined trigger. Practice shows three reliable moments for handoff:
- The client rephrased the same question twice — this means the character did not understand the core, and a third attempt will only increase irritation.
- The topic involves money, health, or legal risk — here the client needs control, not wit, and an operator reduces anxiety better than any script.
- The client directly asks for a person — this request cannot be handled by retrying automation, even if the bot 'almost guessed' the answer.
The main metric trap here is Containment Rate. In banks, it stays around 30%, and on paper, this looks like automation success. But if a client finds the live chat button themselves after five failed bot attempts, the metric won't see it — it will count the dialog as 'closed by bot,' although in fact the client spent extra time and lost trust. The correct handoff metric is not the fact of transfer, but the number of attempts by the character before it: the fewer, the cleaner the transition and the lower the chance that the client will remember the brand as 'the one where the bot was dumb.'
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
