Atacadão Macre · Retail
+42% active retention
behavior-based push notifications in the loyalty app.
View case study →AI automation for your operation
AI agents, chatbots, and automation connected to your ERP, CRM, and messaging channels. Built by a software company founded in 2012 that uses AI in production. A 30-day pilot with measured results before scaling.
- Since 2012
- AI-first in production
- 30-day pilot
- Fixed pricing, no hourly billing
Where AI can help
Where AI can deliver early value
Customer support that cannot keep up
An agent on your website or messaging channel that answers, qualifies, schedules, and hands over to a person when needed, with CRM context.
Manual back-office work
Document processing, data entry, checks, and email replies organized into workflows with human review.
Sales follow-up gaps
Lead qualification, follow-ups, draft proposals, and pipeline updates, with salespeople involved at the appropriate stage.
Decisions without clear data
Reports, exception alerts, and daily summaries based on the information your business already records.
The engagement
An AI pilot in 30 days
We choose one process, establish a baseline, build the solution, and measure the result.
What you receive: a working pilot, before-and-after measures, documentation, and a fixed proposal to expand if the results support it.
Not sure which process to start with? Start with the AI assessment (two weeks, fixed price) to choose the pilot together.
Explore the assessment →| Week | What happens |
|---|---|
| 1 | Map the selected process and establish a baseline for time, cost, volume, or errors |
| 2–3 | Build the integrated agent or automation with human review |
| 4 | Assisted operation, measurement, and a decision to expand, adjust, or stop |
Connected to the tools you use
Models and platforms: Claude (Anthropic), OpenAI, and open models where infrastructure requirements make them appropriate
Case studies
View all case studies → AI & automation in practice
Comparison
Chatbot, AI agent, or automation: which do you need?
These terms overlap. Here is a practical distinction:
| Rule-based chatbot | AI agent | AI automation | |
|---|---|---|---|
| What it does | Follows a predefined sequence of buttons and menus | Interprets requests, consults connected systems, and takes permitted actions | Processes documents, data, or emails without a conversational interface |
| Suitable for | Simple FAQs and basic routing | Support, qualification, and scheduling with relevant context | Back-office work, finance, checks, and reports |
| Limitations | Limited when a request falls outside the defined flow | Needs clear permissions and human handoff rules | Needs measures and review of exceptions |
| Relative implementation effort | Low | Medium | Varies with integration and review requirements |
Your business may benefit from a customer-facing agent and one or two back-office automations. The pilot identifies the most suitable starting point.
Pricing
What determines the cost of an AI project?
- Connected systems. An agent that answers questions and one that checks an ERP, creates CRM records, and schedules appointments have different integration requirements.
- Range of requests. A limited FAQ and a full support process with exceptions, complaints, and changing context require different scopes.
- Accuracy requirements. Drafting a response for human approval and responding independently require different controls and testing. The permitted level of autonomy affects the work.
- Data policies. If information must remain within your infrastructure, that requirement affects model selection, architecture, and cost.
- Volume. Higher conversation volumes can require additional monitoring, limits, and model-cost optimization.
Model usage is a separate operating cost. We explain the expected cost and assumptions before the engagement begins.
Choosing a partner
How to choose an AI implementation partner
| Ask | Why it matters |
|---|---|
| Have you deployed AI in production? | Ask about exceptions, failures, monitoring, and model costs. We use AI in production every week. |
| How will results be measured? | Agree on the baseline and success measures at the start so the pilot can support a decision. |
| Where does a person stay involved? | Human review and escalation are part of the design, especially when an error has significant consequences. |
| What happens to my data? | Discuss the model, hosting, retention, and access arrangements before implementation. |
| Do you also develop software? | The solution needs to work with your existing systems. Ask how the provider will handle integrations and operational exceptions. |
| What if the pilot does not support scaling? | You should still receive the measurements, documentation, and a clear recommendation to adjust or stop. |
Common mistakes
Common mistakes when starting with AI
- Choosing a tool before defining the process. Define a measurable problem, such as reducing response time, before choosing the technology.
- Automating a poorly understood process. Map the workflow first so automation does not reproduce its existing problems.
- Expecting full autonomy on day one. Start with AI suggestions and human approval. Expand autonomy only when observed results support it.
- Ignoring exceptions. Include complaints, unusual requests, missing information, and changes of topic in the project scope and testing.
- Leaving operating costs out of the plan. Account for model usage, monitoring, and adjustments before expanding the pilot.
How we evaluate the solution
- Human review where errors carry meaningful consequences: AI suggests and a person approves until measured results support a change.
- Before-and-after measures. A pilot needs defined measures to support a decision.
- Defined data handling. We choose models according to the use case and your data requirements, including infrastructure constraints.
- The agent identifies itself as an assistant and hands over to a person when needed.
FAQ
AI and automation questions
Do we need organized data before starting?
Not necessarily. The pilot includes assessing what exists and what can be used. Some workflows can begin with email, messaging, and spreadsheets.
Will AI replace my team?
Our goal is to reduce repetitive work and improve what your team can accomplish. Human responsibilities are agreed with you.
How much does a pilot cost?
The pilot uses a fixed scope and price agreed before implementation. A prior AI assessment can help define that scope. Model usage is listed separately.
Which models do you use?
It depends on the use case, cost, and data requirements. We work with Claude, OpenAI, and open models where appropriate.
Can it use the official WhatsApp platform?
Yes, through the WhatsApp Business API, subject to the account and platform requirements. We also connect websites, email, and internal systems.
What if the pilot does not deliver the expected result?
You receive measurements, documentation, and a recommendation on whether to adjust, expand, or stop.
Can you add AI to an existing app or system?
Yes. We assess the existing code, data, and available interfaces to plan the integration.
How soon can we evaluate results?
For a 30-day pilot, before-and-after measures are reviewed in week four. Expansion is a separate project with its own agreed scope and price.
How do you handle customer data?
We define access, data handling, model selection, and any required anonymization during discovery. Requirements depend on the information involved and the agreed architecture.
Do you work with companies in Brazil?
Yes. Megaleios serves companies in Brazil and the United States in English and Portuguese, with a U.S. company based in Orlando, Florida.
Which process would you automate first?
Describe it in a sentence. We’ll discuss whether a pilot makes sense and how it would be scoped.
Chat on WhatsApp