The peak is late at night
Players play when the games are on. The question comes up at midnight on a Saturday, when the human team is at its smallest.
PushAgent · Conversational AI for iGaming
PushAgent: the AI that supports your player at 3 a.m. and knows when to call your team
An orchestrator decides on every incoming message: rules first, AI next, a human when it matters. Withdrawals, bonuses and rollover answered in seconds, with your rules and your tone. When there is no verifiable source to answer from, PushAgent does not make things up: it hands off to your team with the full conversation.
The agent joins your operation without replacing your CRM or migrating channels. You keep deciding what to communicate; it handles support and returns the context.
ask like a player
PushAgent
online
your brand's assistant
decision trail:
rule
ai
human
Illustrative conversation. In a real operation, every answer comes from your brand's knowledge base and rules.
3 layersrule, AI and human in the same decision
24/7late nights, weekends and event peaks
0 to 100explainable churn, propensity, engagement and risk
pt-BRtone calibrated for Brazilian players
the problem
iGaming support is not ordinary customer service.
The conversation is about money, happens outside business hours and clusters exactly where your operation earns the most. One wrong answer about a withdrawal or a bonus costs more than ten slow ones.
The question is about money
Withdrawals, bonuses and rollover account for most contacts. That is where the answer has to be exact, not just friendly.
Volume arrives in waves
Derbies, bonus resets and promotions create concentrated peaks that no on-call schedule absorbs comfortably.
There are several brands
Each brand has its own bonus rules, timelines and tone. What applies to one must not leak into another.
The market is regulated
A licensed operation is accountable for advertising, responsible gaming and the player's personal data in every message sent.
The same question keeps coming back
A large share of contacts are repeat questions with objective answers, taking up the team that should be handling exceptions.
the orchestrator
Rules first. AI next.
A human when it matters.
The difference between a chatbot and a reliable support layer lies in who decides to answer. In PushAgent, every incoming message goes through three layers in order, and the decision is logged. Pick a layer and see how it behaves.
Layer 1: the rule answers
Questions with a single, verifiable answer do not need a language model. They need the right data, fast.
how long does a Pix withdrawal take?
rule · brand payment policy
Pix withdrawals are processed within 1 hour after the review is approved. If it takes longer, I'll let you know right here.
Who decides: the rule your team configured. Deterministic, auditable and the same for everyone.
Layer 2: the AI answers from a source
The question is open, but there is verifiable content to support the answer. The AI interprets and explains, in your brand's tone.
if I withdraw now, do I lose the bonus rollover?
ai · brand knowledge base
Only bets placed with your bonus balance count toward the rollover, and withdrawals don't affect that count. Want me to show you how much is left?
Who decides: the AI, limited to what the knowledge base supports. Outside of it, there is no answer.
Layer 3: a human takes over, with context
An account-specific case, a commercial exception or anything without a source to support the answer. The agent recognizes the limit and passes the ball.
yesterday there was a bonus that never showed up in my account
no verifiable source for this account · handoff with history
This is the bonus team. I already have the whole conversation open, no need to repeat anything.
Who decides: your team, with the full conversation on screen. No starting over from scratch.
scroll to go through the three layers
Want to see this running with your brand's rules? The conversation starts technical: which questions come in most and what already has a written answer.
Talk to a specialistknowledge base
It does not make things up about money.
The rule is strict, and it is what makes AI usable in a regulated operation: without a knowledge base to support the answer, the agent does not answer. It hands off. A hallucination about withdrawals, bonuses or rollover is not a friendliness slip, it is a liability.
- The content is yours. The knowledge base is built from your text, your rules and your policy, not from generic internet knowledge.
- The tone is yours. Answers sound like your brand: direct, warm and not robotic, calibrated to the language of Brazilian players.
- The limit is explicit. When the knowledge base does not cover the case, the agent says what it will do and passes it to the team, without stringing the player along.
- The decision is logged. Every conversation shows which layer answered, which gives you an audit trail instead of a black box.
What changes in practice
- You write the policy onceWithdrawal timelines, bonus rules, rollover, limits and exceptions go into the knowledge base as versioned content.
- The agent answers by itEvery answer relies on that content, so changing the policy changes support the same day.
- What's missing becomes the team's queueQuestions without a source become the list of what needs to be written, instead of turning into improvisation.
customer intelligence
Explainable indicators, with safeguards.
Every conversation feeds 0 to 100 indicators about the player: churn risk, engagement, propensity and risk. Explainable, so your team understands why the number is what it is, and always as a recommendation for whoever decides, not as imposed automation.
- It serves your CRM's decisions. The indicator reaches the team that already runs the cadence, where the campaign is decided.
- It is read in the context of the conversation. Someone who asks about a withdrawal three times in one week is not the same case as someone who disappeared thirty days ago.
- It does not become a revenue promise. No model replaces testing on your own base: the number exists to prioritize, not to promise.
Responsible gaming safeguard
Players showing risk signals do not receive deposit incentives. This rule overrides any campaign, because in a licensed operation that is the line you do not cross.
72
85
64
12
The values above are illustrative. Your indicators are calculated on your own base.
how it fits into your operation
Your team decides. The agent executes and returns the context.
The boundary is explicit and does not change: your team chooses the audience, offer, channel and timing, inside the CRM it already uses. PushAgent handles the conversation, triggers the right channel for the follow-up and returns the data so the next decision is better.
Works with your CRM
Native integration with Smartico and FastTrack, and a documented public API for in-house CRMs. Pushfy does not build CRMs and does not intend to.
The follow-up goes out on the right channel
When the conversation generates a notice, it follows the operation's cadence: SMS, RCS, Voice, Push or WhatsApp, with fallback between channels.
Integrates via API and webhook
The same token-authenticated API that already sends on every channel, with status and reply webhooks so your system can keep track.
Isolated multi-brand
Each brand with its own knowledge base, permissions and reports. No rule from one brand answering for another.
Access governance
Role-based permissions, two-step verification and complete audit trails of who changed what.
Support from the platform team
You talk directly with the Brazil-based team that runs the platform, including on the Saturday night when the volume shows up.
regulated market
Built for a licensed operator, not just any website.
Fixed-odds betting operations in Brazil are governed by Law 14.790/2023 (Brazil's regulated betting law), and all of a player's personal data is subject to the LGPD (Law No. 13.709/2018, Brazil's General Data Protection Law). This is not legal fine print: it changes what an AI can and cannot say and do.
- Responsible gaming ahead of the campaign. A risk signal suspends deposit incentives, regardless of what the cadence would call for.
- Opt-out honored on every channel. Whoever asks to stop receiving messages stops receiving them, with a record of the request and suppression applied across the entire operation.
- No made-up answers about amounts. Without a verifiable source, the conversation goes to a human. That is the difference between a support mistake and a compliance problem.
- Decision trail. Every conversation records which layer answered, so you can audit instead of having to trust.
When it comes to money, the agent would rather not answer than answer wrong.
This is the one rule that makes AI usable in a licensed operation. A hallucination about withdrawals, bonuses or rollover is not a friendliness slip: it is a liability with the player and with the regulator.
the difference
A generic chatbot and a support layer are not the same thing.
The comparison below is not against any specific vendor. It is against the most common market practice: installing an FAQ bot on top of an iGaming operation.
With a generic FAQ bot
- Answers everything, including what it has no way of knowing.
- Built on generic content, not on your brand's policy.
- Hands off to a human and the player tells the story again, from the start.
- Treats iGaming like retail: knows nothing about withdrawals, rollover or bonus resets.
- Does not separate brands within the same group.
- Returns nothing to whoever runs the CRM cadence.
With PushAgent
- No verifiable source, no answer. It recognizes the limit and hands off.
- The knowledge base is yours. Your text, your bonus rules, your withdrawal policy.
- Handoff with the complete history, in real time, without the player repeating anything.
- iGaming vocabulary, with the tone calibrated for Brazilian players.
- Isolated multi-brand: knowledge base, permissions and reports separated by brand.
- Returns explainable indicators to the team that decides communication.
getting started
It starts with a narrow scope, on one brand.
Nobody puts an AI in front of the entire player base on day one. The path that works is narrow and verifiable: a small set of questions, one brand, handoff switched on and success criteria written down before starting.
-
Map what comes in most
We identify the questions that drive most of the contact volume and which ones already have a written answer in your operation.
-
Build the knowledge base and tone
Your policy becomes a versioned knowledge base, and the brand's tone is calibrated with your team.
-
Go live with handoff on
The agent goes live on one brand, answering only what the knowledge base supports. Everything else goes to your team.
-
Measure and expand
You track what was resolved by each layer and what became a handoff, and expand the scope with numbers on the table.
The first step is a technical conversation, not a proposal. Bring the five questions your support receives most and we will show how each one would be resolved.
Talk to a specialistfrequently asked questions
What people usually ask before switching on the agent.
What is PushAgent?
PushAgent is Pushfy's conversational AI, built for iGaming operators' player support. An orchestrator decides on every incoming message across three layers: the rule answers what has an objective answer, the AI answers what the brand's knowledge base supports, and human agents take over the rest, receiving the full conversation.
Can PushAgent make up an answer about a withdrawal or bonus?
This is the rule the product rests on: without a knowledge base to support the answer, the agent does not answer. It recognizes the limit, tells the player and hands off to your team with the complete history. In a regulated operation, a made-up answer about money is not a support mistake, it is a liability.
Do I need to replace my CRM to use PushAgent?
No. Pushfy is the execution layer and does not build CRMs: your team decides the audience, offer, channel and timing, inside the tool it already uses. There is native integration with Smartico and FastTrack, and a documented public API for in-house CRMs or any other platform.
How does PushAgent handle responsible gaming?
With an explicit safeguard: players showing risk signals do not receive deposit incentives, regardless of what the campaign cadence would call for. On top of that, opt-out is honored on every channel, with a record of the request, along with the obligations of an operation licensed under Law 14.790/2023 and the LGPD (Law No. 13.709/2018).
What happens when the conversation goes to a human agent?
The conversation moves to the agent with the complete history, in real time. The player does not repeat the question or start over, and the team receives the context of what was already answered and by which layer.
Does it work for a group with several brands?
Yes. Each brand has its own knowledge base, tone, permissions and reports. In multi-account operations, role-based permissions and isolated reports keep one brand's rules from answering for another.
How long does it take to get PushAgent live?
It depends on how much of your policy is already written, so there is no universal timeline and it would not be right to promise one. The path that works is to start narrow: the questions that come in most, one brand, handoff switched on from day one and success criteria agreed before starting.
Bring one brand and five questions.
We will show which layer would resolve each one, what would need to be written in the knowledge base and where the handoff comes in. No need to replace your CRM or migrate channels to have the conversation.
Free account and pay-as-you-go pricing, no monthly fee. The PushAgent conversation does not require migrating anything or switching vendors.