Case study · Consumer goods
Thums Up Believer Bot
Turning World Cup doubters into believers, one reply at a time.
- Client
- Thums Up
- Service
- AI campaign platform, design and development
- Sector
- Consumer goods
- Moment
- ICC World Cup 2023
At a glance
What the platform did
Fans who posted with a campaign hashtag got a personalised response back in seconds, not hours.
Crew, editors and a final reviewer signed off. Nothing reached the timeline unread.
The whole campaign ran on the platform where World Cup conversation actually happened.
Built for the run up to the tournament, the loudest month in Indian sport.
The brief
A brand cannot reply to a nation by hand
A cricket World Cup is the busiest month on Indian social media, and Thums Up wanted to be in the conversation rather than beside it. The campaign invited fans to post with its hashtags. Every one of those posts deserved a reply that read as though a person had written it for that fan, about that tweet.
That is a volume problem and a trust problem at the same time. A human team fast enough to answer a World Cup timeline does not exist. A language model left to post on its own is quick enough, but a brand this size cannot risk a single bad reply going out under its name, and the incoming posts are unfiltered public text.
The build is the reconciliation of those two facts. Generation is automated. Publishing never is.
How it worked
Four stages, one of them human
Collect
Sprinklr watched the campaign hashtags and pulled in matching posts as they landed.
Filter
A profanity screen stripped anything the brand should not be engaging with in the first place.
Generate
An LLM drafted a personalised reply to that specific post, not a template with a name dropped in.
Approve
Crew, then editors, then a final reviewer. Three sign offs stood between a draft and the timeline.
The build
Why the slow step is the point
Most of this pipeline exists to make the last stage affordable. Collection and filtering mean reviewers never see posts the brand had no business answering. Generation means they are editing a draft rather than composing from scratch. By the time a reply reaches a human it is already close, so a person can clear it in the time it takes to read it.
That is what lets a three step approval chain run at campaign speed instead of becoming the bottleneck. The model does the drafting. The crew, the editors and the final reviewer keep the brand's voice and its judgement in the loop, which on a public timeline during a World Cup is not a formality.
Wow Labz built the platform end to end, from the collection and moderation layer through the generation service to the review tooling the approval chain ran on.
Stack
What it was built with
Questions
Believer Bot, answered
What was the Thums Up Believer Bot?
A Twitter/X campaign platform Wow Labz built for Thums Up in the run up to the ICC World Cup 2023. Fans posted using the campaign hashtags and received a personalised reply within seconds.
How were the replies kept brand safe?
Two ways. Incoming posts passed through a profanity filter first, so the brand never engaged with content it should not have. Outgoing replies were drafted by a language model but never published automatically, every one cleared a three step human approval chain across crew, editors and a final reviewer.
Were the replies automated or written by people?
Both, deliberately. An LLM generated each reply against the specific post it was answering, which is what made the speed possible. People approved every reply before it posted, which is what made the speed safe.
What did Wow Labz build?
The AI campaign platform end to end, plus the design and development around it. That covers tweet collection through Sprinklr, the content moderation layer, the LLM response generation service, and the review tooling the approval chain ran on.
Let's talk
Got a developer product nobody finishes integrating?
Tell us what you are building. We will tell you how we would ship it.
