HUBSPOT
Summary
When Boca Rosa experienced a peak of post-live crisis (high volume, integration and quality issues), service became a bottleneck and the reputation suffered: NPS 32, Reclame Aqui 3.3, and TMA of 25 hours. With Cloud Humans' AI (the Rosa, brand persona), the team regained performance and customer trust: in 6 months achieved NPS 78, earned RA 1000, reduced the TMA to 3.3 hours, raised CSAT of the AI to 84.3, and automated 54% of the tickets — even with 80% of the contacts being logistics.
The challenge: when the buzz is huge and the experience can’t keep up, the queue explodes
Boca Rosa Company began to operate 100% "in-house" from October of the previous year (and B2B in January). In June 2024, a live event generated a lot of revenue and expectations — and then came the combo that no one wants:
Integration issues and order updates
Cases of defective product/packaging
Digitally native consumers demanding answers "yesterday"
Demand spikes on a lean team
Impact before AI (at the height of the crisis):
NPS 32
Reclame Aqui 3.3 (not recommended / red)
TMA 25 hours
FCR 7/10 (3 out of 10 came back)
At one point, the company made a rare decision:
pause the operation to fix what needed to be fixed and reconnect with the community
“Post-sale crisis equals high volume of contacts + misaligned experience.”
— Xavier
Why AI has become inevitable
in native digital branding, "square chatbot" destroys identity.
The team needed to scale without becoming a generic service.
What was at stake:
Maintain the "Boca Rosa way" of speaking
Ensure agility and empathy during seasonality (launch = peak)
Get the team out of the conveyor mode (copy and paste)
Free up space for CX to influence the business (social, e-commerce, supply)
What made the difference in building Rosa
1) Start small and map frictions (real Pareto)
They already had a very clear picture of where it hurt the most: peaks post-live, logistics becoming 80% of the demand, anxiety for status, and a native digital customer without patience for long SLAs. So AI first entered where it truly made a difference:
tracking
order status
repetitive questions that congested the queue
sorting and pre-collecting data
2) Minimum viable content, with curation
Boca Rosa didn't need a "perfect center" to start. They needed a useful V1, aligned with the brand's voice and the themes that appeared the most in crises.
What was included at the start:
logistics and deadline responses
support policies and flows (exchange, return, damage)
common exceptions at launch
messages of empathy and transparency (without becoming a robot)
3) Continuous auditing with a clear owner (this speeds things up immensely)
Another point that almost everyone ignores: having someone from CX as the owner of the audit changes the game. It’s not an “IT project.” It’s customer experience.
And when the owner is close to the operation:
feedback becomes a short cycle
the corrections are quick
AI starts “ramping up” like a real person
4) Personality and tone of voice (without destroying the branding)
Here there was an obvious requirement: it wasn't possible to scale with a clunky chatbot.
AI needed to speak as Boca Rosa speaks.
So the persona was designed to:
maintain dynamics and connection
use emojis in moderation (and with context)
vary responses (not become a script)
mirror the client's tone when it makes sense
Xavier said something that sums it up well: there are clients who end the conversation without realizing it was AI — and this only happens when the language is human and consistent with the brand.
5) Integration to move from "FAQ" to becoming an agent
For me, this is the dividing line: chatbot answers. agent resolves.
And in Boca Rosa's case, since 80% of the tickets are logistics, simply having content isn't enough.
The leap comes when AI can:
consult updated status
pull tracking automatically
pre-collect data (photo, evidence, purchase info)
deliver to the human a "chewed-up" case when it needs to scale
And this is exactly where AI becomes even stronger: connecting backoffice and taking the repetitive “copy and paste of tracking” work off the human's hands.
Example of a real conversation
If you prefer, I can also transfer you to the human team to finalize any adjustments now.
What is happening behind the scenes:
status inquiry via integration (logistics)
tone of voice mirroring the customer (without overdoing emojis)
clear overflow option
reduction of public friction (fewer complaints on Reclame Aqui)
Key metrics
NPS: 32 → 78 (in ~6 months)
Reclame Aqui: 3.3 → RA 1000
TMA: 25h → 3.3h
FCR: 7/10 → 9/10 (only 1 return)
AI CSAT (Rosa): 84.3
Retention/automation: 54% (previous slide 44%)
Demand mix: 80% logistics
How it was to implement (step by step, concise)
1️⃣ Setup and connection with previous channel (without internal dev team)
Boca Rosa did not have developers
Cloud Humans entered strongly as an implementation partner
Onboarding with close monitoring
2️⃣ Persona and brand voice (Rosa)
native digital language
consistency with social/content
emoji in moderation
variable responses (not “robot script”)
3️⃣ Continuous auditing (with owner in CX)
dedicated person to train and adjust
analysis of what improves CSAT vs retention
easy feedback (like/dislike + correction)
4️⃣ Sensitive case rules
product quality
returns
adverse reaction
AI recognizes and transfers transparently
5️⃣ ED in the backoffice (logistics and editing)
integration with logistics operator and purchase data
tracking automation
next step: registration update (email/address) without human
WHAT CHANGED THE GAME: From AI in Zendesk to the complete Cloud Humans solution (AI + Cloud Chat)
An important point in Boca Rosa's journey: at the beginning, Rosa operated integrated with Zendesk.
It worked.
But it wasn't ideal.
Over time, it became clear that it made more sense to use the complete solution from Cloud Humans — AI + Cloud Chat in the same ecosystem.
What changed in practice
1️⃣ Real WhatsApp (without brokers)
In Zendesk: it needed to work with brokers because the tool was not native for WhatsApp.
In Cloud Chat: official and directly integrated WhatsApp, without workaround.
2️⃣ Simpler tool
Zendesk was complex and expensive for what they used.
Cloud Chat is more intuitive, easier for the team, and faster to operate.
3️⃣ More aligned pricing model
Zendesk charged for modules and add-ons, with long contracts.
Cloud Humans has a more predictable model without "tricks".
4️⃣ Much closer support
In Zendesk, support was distant and often pushed towards expensive third-party consulting.
In Cloud Humans, implementation and follow-up were done in real partnership.
5️⃣ Easier AI to train
The AI within Zendesk was more complex to maintain.
Rosa is simple to audit, adjust, and evolve — directly through the interface.
The result was not just technical.
It was operational.
The side effect I like the most: human team becomes strategic
When the repetitive disappears, the team stops surviving and starts building.
What began to fit into the agenda:
group delight actions
customer voice committee
rituals with social, supply, and e-commerce
root cause analysis (instead of "complaining team")
