



Setting up ADA is, above all, a series of decisions about how your brand should sound and when it may speak.
Communities comment around the clock and on every network. Whoever asks expects an answer, whoever praises expects a nod, whoever criticizes expects to be heard. But the team has working hours, priorities, and a limit to how many comments it can handle by hand.
ADA, Bunker's AI assistant, replies to comments on behalf of your brand, by your rules and only when it's sure. This guide walks through the step-by-step setup inside Bunker Analytics: what you define, in what order, and what each feature unlocks.
Where it's set up
ADA lives in the Sentiment module: the ADA button opens its hub, and that is where the two pieces you configure sit, the brand tone and the automated responses. Order matters. The tone first, so ADA knows how it should sound; then the rules for when it may speak.
Before touching anything, it helps to see what it rests on. Three pieces of Bunker Analytics work in sequence. Inbox Management brings the comments and messages from Facebook, Instagram, LinkedIn, TikTok, X, and YouTube into a single inbox, Ads posts included. Sentiment Analysis classifies each interaction as positive, neutral, or negative. And Automated Actions tags by topic and intent and triggers the actions. On that foundation, ADA decides whether to reply.
The first benefit shows up right here, before you configure anything: a single omnichannel inbox. The team stops jumping between apps and losing conversations across tabs, and ADA's entire scope is governed from one screen.
Step A · Brand tone
A correct answer in the wrong tone is still a mistake. That is why the tone comes first: everything you define here applies to every automated response. Tuned well, ADA does not sound like a bot. It replies the way the brand would, with a naturalness that feels organic, as if someone from the team were writing on the other side. That is the potential of this step: one consistent, distinctive voice, instead of a tone that shifts depending on who answers.
The hub asks for six definitions:
To grasp what this step achieves, it is enough to see the same ADA on two opposite brands. With a "formal, reserved, no humor" setup, faced with "it's great but the learning curve is steep," it acknowledges the praise and validates the friction in a professional register. With a "casual, playful, emoji-friendly" setup, the same engine answers loosely, with a wink and an emoji. Same brain; different voice.
None of this goes live blindly. That tester works like a playground: it lets you simulate conversations, type the comments a real user would send, and see on the spot how ADA would respond in each situation that might come up. It is the place to anticipate the odd cases and adjust the tone until it convinces you, before anything goes public.
Step B · Automated responses
With the voice ready, an assistant guides the setup of the automated responses from start to finish. You can save a draft at any point, and nothing goes live until the last step. There are four steps, and each one unlocks a different capability.
Heads-up · X
On X, ADA cannot verify whether a comment was already answered manually, so it might reply to one the team has already handled. Keep it in mind when you enable it.
What it answers and what it routes
Once the rules are in place, ADA stops being a mechanical filter and works as a system that understands what each interaction is about. It reads the topic and the sentiment of every comment and every direct message, and from that it makes one decision: reply automatically or route the conversation to the right Community Manager.
An example makes it concrete. A customer comments, annoyed, that the product quality was not what they expected. ADA recognizes the topic (Quality and Product), the sentiment (negative), and the intent (a complaint). Because it is a sensitive complaint, it does not improvise an answer: it tags it, prioritizes it, and routes it to a person, with the context already sorted so they can reply in seconds instead of hours.
The value is not only speed. Every conversation reaches the right person already classified, and the ones ADA can resolve on its own stop queuing behind the complex ones.
The intelligence
Behind that decision to reply or route there is a concrete mechanism, and it is what separates a reliable automation from a bot that slips up in public. Before replying, ADA reads the comment the way a person on the team would, running it through six comprehension filters:
That confidence is what governs the final decision, with a simple rule: if it clears 80%, ADA publishes on its own; if it is 80% or below, it holds back and routes the comment to the team. The number reflects whether the answer can go out alone or needs a person. Real cases from the tests make it clear:
| Comment | Confidence | Decision |
|---|---|---|
| "great post" | 95% | Replies: clear praise, an answer that covers it fully. |
| "it's great, but the learning curve is steep" | 84% | Replies: mixed, but the answer addresses the criticism. |
| "the dashboards take forever to load" | 63% | Routes: a technical complaint that deserves a person. |
| "loved it, do you have a newsletter?" | 33% | Routes: it would claim a fact it does not have. |
On top of that come the by-design pass-overs, even when the comment meets the rules: missing commercial data (prices, stock, dates), user-to-user interactions, technical complaints that call for real support, and sensitive or off-topic matters (politics, religion, or questions about the support itself). This filter is what makes automation worthwhile: ADA does not just reply fast, it knows when to stay quiet.
And there is a benefit that goes beyond the reply: every comment classified by sentiment and topic stops being loose text and turns into business data that feeds your Analytics dashboards.
We would rather ADA not reply than reply badly: every automated response is the brand speaking in public.
Not just automation
Automating does not always mean replacing. In Suggest-only mode, ADA works alongside the team: it drafts a ready reply for each comment, which the Community Manager reviews, adjusts, or publishes as is. That suggestion works like a template, a starting point that turns writing from scratch into a quick review.
On top of that comes auto-tagging. ADA classifies every interaction by topic, intent, and sentiment with no one doing it by hand, so the team finds, filters, and prioritizes without wasting time. And the sentiment radar gets ahead of problems: a spike of negative comments on a post is an early warning, the signal to act before an annoyance becomes a crisis.
The effect is a team that is faster across the whole cycle: replying, analyzing, and above all resolving emergencies when every minute counts.
Before you activate
The setup lets you switch everything on day one, but that is not what we recommend. The rollout that works is gradual, and it leans on exactly the controls you have seen so far:
The logic is the same one that holds up the whole system: control is never handed over, it is dialed in. First you observe, then you enable little by little, and full automation arrives only once the evidence has given you the confidence to do it.
Setting up ADA is not handing the conversation over to a machine: it is writing down, once, the decisions about how the brand sounds and when it may speak, so replies land on time even when the team is off. All of it lives inside Bunker Analytics, together with the data every comment generates.
To see it on your own channels, let's book a walkthrough and set it up together with real comments.
Lucas Suarez
Marketing Analyst
1/9