Artificial intelligence consulting for enterprises, applied to corporate data
We design, build and run AI solutions on the environment the company already has: agents, conversational assistants, document search and process automation. We start from Power BI, the ERP and the AI contracts already in place, with the source and the query shown in every answer.
Platform approved in a formal information security, legal and AI ethics review by one of the world's largest technology companies, where it runs in production. Qualification materials are shared under NDA.
What an AI consultancy does for enterprises
An artificial intelligence consultancy chooses where AI produces results, prepares the data, builds the agents and puts everything into production securely, measuring usage, cost and answer quality. Active BI does this work on the data the company already governs (Power BI, ERP, databases and documents), with the source visible in every answer and each user's permissions respected.
The path starts with a proof of concept on real data, moves to deployment in the company's environment and grows one agent at a time. Remote delivery, for companies across Brazil and with operations abroad.
From the analytics environment to an AI decision layer
The company has already invested in data. What is missing is a layer that can interpret that environment and deliver context, analysis and foresight to decision makers.
What you have already built
- Power BI, semantic models and reports
- Microsoft Fabric and Databricks, Lakehouse and pipelines
- Databases such as SQL Server, PostgreSQL and Oracle
- ERP, CRM and corporate systems
- Documents: policies, contracts and manuals
Where deciding still takes effort
- Questions answered by digging through dashboards, databases and spreadsheets
- Reliance on specialists: decision makers wait for whoever knows the tool
- Backward-looking analysis: explaining variances and anticipating risks is still manual
The decision layer
- Ask in plain language
- Understand why things changed
- Anticipate deviations and risks
- Decide with context, not just numbers
AI on the ecosystem the company already uses
We do not start by replacing tools. We assess the environment, the security requirements and the AI contracts in force, and choose the path that makes the most of the investment already made.
| Path | When we recommend it |
|---|---|
| Active IA, the enterprise AI hub | A single portal, with corporate sign-in, where people use the company's agents, in-house or from other platforms, in Chat, Journal and Signal. Observability shows usage, cost and answer quality by department and by agent. |
| Active Agents API | Active BI's integration layer that works with agents and models from any vendor: Microsoft Fabric, Copilot Studio, Databricks Genie, Azure OpenAI, OpenAI, Claude and Gemini. The company can combine or switch solutions without rebuilding what is already in production. |
| Agents in Microsoft Fabric | When the data environment is already on Fabric, or moving there, and agents can work on the lakehouse and the models governed there. |
| Databricks Genie | When the data is in Databricks and the team wants natural-language questions over Unity Catalog. Genie joins the AI hub through the Active Agents API, with the same governance as the other agents. |
| Copilot and Copilot Studio | When the company works in Microsoft 365 and wants agents inside everyday tools. The semantic model we prepare improves Copilot's answers as well. |
| Azure OpenAI, ChatGPT Enterprise, Claude, Gemini and open models | When there is already a contract with a provider, or a hosting and region requirement. The contract stays in place; Active BI adds the business context, the rules, the accuracy tests and the governance. |
| Predictive models | Demand forecasting, churn, propensity and anomaly detection, built on Fabric, Databricks or Python, or reused when they already exist. Generative AI then explains the forecast and alerts when the forecast deviation exceeds the threshold. |
Generic AI or AI on corporate data
The question that separates a pilot from a solution in production is simple: can you trust the answer without checking the report?
| Generic AI | AI on corporate data |
|---|---|
| Answers with public knowledge | Answers with company data, from the sources it already uses |
| Calculates its own way | Runs the validated measure in the semantic model. With no measure, it says it does not know |
| Shows everyone the same content | Inherits each user's permissions: everyone sees what they could already see |
| Answer with no origin | Source and query visible in every answer |
| Usage with no record | Question, query, model, time and cost logged |
Copilot, ChatGPT Enterprise or your own AI hub?
It is the most common question from companies that already have a contract with Microsoft or OpenAI. The options are not mutually exclusive: many companies use all three, each for a different type of question.
| Criterion | Copilot | ChatGPT Enterprise and similar | Your own hub (Active IA) |
|---|---|---|---|
| Best for | Productivity in Microsoft 365: email, documents, meetings | Writing, research and analysis of files by the user | Business questions on the company's official data |
| Data sources | Microsoft ecosystem and the Power BI semantic model | Whatever the user attaches or connects | Power BI, databases, ERP, documents and other agents, combined |
| Business rules | Depends on how the semantic model is prepared | Interpreted in each conversation | Validated measures and company vocabulary configured |
| Logging and cost | License per user | License per user | Every answer logged, with cost by agent and by department |
When the company already uses Copilot, we prepare the semantic model and the rules to improve its answers too. See Power BI and Artificial Intelligence.
Where AI fits in each department
The same layer serves different departments. A new department is configuration on sources that are already connected, not a project from scratch.
| Department | Questions AI can now answer |
|---|---|
| Sales | "Why did revenue in the South unit drop, and what does the price adjustment contract say about it?" · "Which of our twenty largest customers have gone 30 days without buying?" |
| Finance | "Where did margin fall below target this month, and which products explain the gap?" · "Which cost centers went over budget, and why?" |
| Operations | "Which branch had the worst SLA performance this week?" · "Where is production behind plan?" |
| Supply chain | "Which products are at risk of stockout in the next two weeks?" · "Which supplier's delivery time got worse, and what penalty does the contract set?" |
| HR | "Which departments have the highest turnover this year, and what changed?" · "Which positions have been open for more than 60 days, and why?" |
| Legal and contracts | "Which contracts expire in the next 90 days and have automatic price adjustment?" · "What does the contract with this supplier say about late-delivery penalties?" |
| Customer service | "Which customers opened the most tickets this month, and about what?" · "What is the standard response to this request under the current policy?" |
| HR | "Which departments have the highest turnover this year, and what changed?" · "Which positions have been open for more than 60 days, and why?" |
| Legal and contracts | "Which contracts expire in the next 90 days and have automatic price adjustment?" · "What does the contract with this supplier say about late-delivery penalties?" |
| Customer service | "Which customers opened the most tickets this month, and about what?" · "What is the standard response to this request under the current policy?" |
AI also works outside the chat: reading documents and executing process steps, in Intelligent Process Automation, and inside work screens, in custom software.
AI also works outside the chat: reading documents and executing process steps, in Intelligent Process Automation, and inside work screens, in custom software.
AI approved by security, legal and ethics, and in production
One of the world's largest technology companies wanted its business areas to query data in plain language without giving up the security rules that apply to any corporate system.
Bringing AI to sensitive data, with users of different profiles, and passing the formal review by the information security, legal and AI ethics committees.
Active IA, with corporate sign-in, permissions inherited per user, a record of every answer and a dashboard of usage, cost and quality by department.
The platform was approved in the formal review by all three committees and runs in production. Qualification materials are shared under NDA.
Will your security committee review the AI? We present the architecture, the controls and the qualification materials in the same conversation.
What our clients say
Our partnership with Active BI is providing invaluable support to Lenovo throughout our digital transformation journey. Professionalism, technical proficiency and proactive communication.
The Active BI team is extremely punctual and responsible, committed to its deliverables and solutions. Both the technical and management teams are excellent.
Active BI has highly skilled, up-to-date professionals who deliver fast, efficient solutions. We also highlight their clear communication, availability and ability to simplify complex topics. An excellent service that we recommend without hesitation.
Active BI artificial intelligence services
These services rely on Active IA, Active BI's enterprise AI hub, with the Chat, Journal and Signal modules and observability of usage, cost and quality, and on the Active Agents API, which connects agents from any platform. The product is described on Active IA, and the service for plain-language questions on data, on Conversational AI for Data.
How to start an AI project: from proof of concept to operation
No stage starts until the previous one has been evaluated. At each stage, the company decides whether to continue, adjust or stop.
One agent, on one source, in the chosen domain, with the company's real data.
The company defines the domainThe criteria agreed at kickoff are measured together with the people who used it.
The company decides whether to proceedThe platform runs in the company's environment, with corporate sign-in.
The company approves the environmentNew agents on other sources and departments, and alerts configured by department.
The company prioritizes the backlogMonitoring of usage and answer quality, support and evolution.
The company sets the paceBefore the proof of concept, when numbers still differ
If the company does not yet know whether its data is ready for AI, the starting point is the Information Assessment. In four to eight weeks, we assess data and AI maturity and define what needs to be prepared before investing in agents.
How much an AI project costs: an investment in stages
Each stage delivers something different and is contracted when the company decides to move forward. Only the proof of concept needs to be decided at the start; the others serve as a reference for budget planning.
| Stage | What it delivers | How it is contracted | When it is decided |
|---|---|---|---|
| Proof of concept | One agent in use with real data, a measured result and a decision on whether to proceed. | Fixed scope | Now |
| Deployment | Platform in the company's environment, with corporate sign-in and governance, and the proof-of-concept agent migrated. | Once | After the proof-of-concept evaluation |
| Production | Team using it, with a usage and cost dashboard from day one. | Included in deployment | After validation by the business areas |
| Evolution | New agents, sources, alerts and capabilities. | Per agent, as a fixed-scope deliverable | According to the prioritized backlog |
| Support | Ongoing oversight of what is live: quality, cost and adjustments. | Monthly | From production onward |
Deployment happens once
It brings the platform into the company's environment, with security, governance and the proof-of-concept agent. New agents run on the same platform without a new deployment: the company does not pay twice for the same work.
Each agent has a known investment
We have objective criteria to size each agent, based on the data source and the complexity of the use case. The price is presented before work starts, with no surprises along the way.
Support with a clear scope
After go-live, a monthly pool of hours keeps the portal and the agents aligned with real usage. What counts as support and what counts as evolution is defined in the contract, so there are no scope disputes.
In the conversation, we show how the investment works
We present the criteria for sizing agents, what support covers and the estimated technology costs for your volume. AI model consumption, cloud and licenses vary with the architecture and usage, and are handled separately from services.
Discuss the investmentThe simulator estimates AI model consumption based on the number of users, the volume of questions and the type of use.
Simulate AI costWhy so many AI projects stall at the pilot stage
The AI model is rarely the problem. Pilots that never reach production usually stumble on the same five points, and that is where we start.
Ungoverned data
If numbers differ across reports, AI repeats the discrepancy, only faster. We start from validated measures and, where they are missing, prepare the data first.
A question with no owner
Without an accountable department, no one validates the answers or drives usage. The proof of concept starts with a defined domain and owner.
Success not agreed upfront
Without criteria agreed in advance, the pilot becomes a matter of opinion. We agree on the test questions and the expected accuracy at kickoff.
Uncontrolled cost
AI consumption grows without anyone noticing. The dashboard shows cost by agent and by department from day one.
No one uses it after launch
AI that does not become part of the routine dies quietly. We track adoption by department and adjust the agent to real usage.
How we measure results
The same indicators a director needs to justify continued investment, drawn from Active IA observability:
AI models and data sources
Each task goes to the most suitable AI model. If a better model appears tomorrow, switching does not require rebuilding the solution.
Models and assistants
Data sources
Further reading
Frequently asked questions about AI for enterprises
What is artificial intelligence for enterprises?
What is the difference between using ChatGPT and having AI on the company's data?
How much does an enterprise AI project cost?
How much does AI consumption cost per month?
Where should an enterprise AI project start?
Do I need data engineering before deploying AI?
How is the investment in a new agent defined?
How does support work after go-live?
Can AI make up numbers? How do you prevent that?
Is company data used to train AI models?
Which AI models and platforms does Active BI work with?
How is LGPD (Brazil's data protection law) handled when AI queries personal data?
What is RAG and when does it make sense?
How long does an AI proof of concept take?
We already use Copilot or ChatGPT Enterprise. Does an AI project with Active BI still make sense?
Do you also build predictive models?
Does Active BI offer AI training for teams?
How do you measure the return on an AI project?
Does Active BI work remotely?
What question does your leadership team ask every week?
We start with it. Bring the question and the report that answers it today: in one conversation, we choose the domain, the source and the success criteria for the proof of concept.
- You tell us the question and where the data lives
- A specialist replies within 30 minutes during business hours (Brasília time)
- A conversation to define the domain, source and success criteria
No commitment and no follow-up call sequences.
Propose a proof of concept
2 steps, under 40 seconds
Step 1 of 2