White-label conversational BI for software vendors and consultancies

If you sell a vertical ERP, a payroll system, a public-sector platform or data consulting, you hear the same request in most account reviews. Customers want to ask a question in plain language and get the answer, without opening a ticket or waiting for a new report. A distribution manager wants to know which salesperson slipped this month. A finance secretary wants committed spending against budget. Today that person depends on a fixed report or on someone who can write a query.

The partner's system already stores that data. What is missing is the layer that understands the question, writes the right query against the database and returns a number, a table or a chart the manager can check. Many vendors have tried to close the gap with more reports, more filters and more screens. The backlog never empties, because every new question becomes a development request.

There is a commercial reason too. A customer who finds new value in a system they already use tends to stay longer and accept an add-on plan. For a consultancy the gain looks similar: instead of delivering a dashboard and leaving the project, it keeps a recurring line of revenue on what it already implemented.

Building is what stops most of them. A natural-language-to-SQL translator that works on a real customer's real database takes far more effort than a demo suggests. The partner wants to offer the capability, not staff an AI team to maintain it. That is where the white-label model fits.

To understand the product behind the offer, read what conversational BI is and see how Entendo connects to an ERP database.

The partner operates Entendo under its own identity. The end customer opens an address that belongs to the partner, sees the partner's logo and uses the product as an extension of the system they already know. In practice, the partner gets:

• Its own brand: the partner's logo and colors across the interface.
• Its own domain: customer access lives at a partner address.
• Branded emails: invitations, notices and messages go out with the partner's identity.
• Branded exports: the CSV, XLSX, PNG and PDF files customers download carry the partner's identity.
• Customer management: the partner creates and administers its customers' workspaces in one place.
• A semantic layer per vertical: the glossary, targets and indicators for a segment (distribution, payroll or public administration, for example) can be built once and reused for the next customers.

The semantic layer deserves attention because it separates a demo from a usable product. It tells the AI that "revenue" means a specific column with specific filters, that "delinquent" follows the partner's rule, and that the fiscal year starts in a different month. Once the partner builds that layer for its vertical, each new customer starts from a ready base and only adjusts what is particular.

Beyond the question itself, the partner brings customers what Entendo already does: dashboards with filters and drill-down, shared links that refresh automatically, time-series forecasting and multi-section reports generated from a single question. The interface comes in Portuguese and English. The generated SQL and the reasoning stay visible, which helps the partner stand behind a number in front of the customer.

Software vendors handle sensitive data that belongs to other companies, so a partner's first worry is usually security. Access works like this.

Entendo Gateway is an agent installed on the end customer's network, close to the database. It opens an encrypted outbound connection to Entendo, so the customer does not open an inbound firewall port or whitelist an IP address. The database credential stays inside the customer's network. What travels to the cloud is the query result, not the whole database and not the password.

The product is read-only. Before running a query, Entendo blocks any write command. We also recommend that the customer create a read-only database credential, so the restriction exists on both ends. Every question and every SQL statement is logged, so you can audit who asked what.

Between customers, isolation happens in the platform's own database through PostgreSQL row-level security. On the Business and Enterprise plans, customers can use SSO through OIDC, configured per customer. On the Pro, Business and Enterprise plans, a customer who requires it can bring its own key for one of the supported AI models (OpenAI, Anthropic and Google Gemini).

Supported databases are PostgreSQL, MySQL, SQL Server, Oracle, Firebird and ClickHouse, plus Google Sheets, CSV and Excel. Firebird and Oracle matter a great deal for anyone serving the Brazilian market: legacy management systems and established ERPs still run on them, and many BI tools skip those databases.

Every software vendor with a development team has considered building this in-house. An honest estimate starts with what a plain-language question system needs before anyone can trust it on a customer's database.

1. A semantic layer. Table and column names in a system rarely look like business language. Someone has to map them, keep the glossary current and update it with every release.
2. A read-only guard. An AI that writes SQL can write a destructive command. Every query needs validation before it runs, and nothing can alter the customer's database.
3. Auditing. Who asked, what was generated and what came back must be recorded, because the customer will want to check.
4. Multiple databases and dialects. One customer runs Oracle, another Firebird, another SQL Server. Each dialect has its own date functions, pagination and types.
5. Secure access to the customer's network. Connecting to a third party's database without asking for an open port is a project of its own.
6. Continuous evaluation. AI models change, the partner's system changes, and answer quality has to be measured after each change against a set of reference questions.
7. The interface, dashboards, exports, permissions and user management that customers expect to find next to the question box.

Each of these is product work that runs for years, not a sprint ticket. When a partner licenses, it trades that investment for an operation under its own brand. Building tends to weigh more heavily on vendors with few customers in the vertical, because the AI team has to be paid before revenue arrives. For a company that wants full control of the code, building remains a legitimate decision. What matters is deciding from the list above, not from a demo alone.

The path has four steps, and there is no rush to sign anything before you see it working.

1. Conversation. We learn about your system, your customer base, the database it uses and your vertical. This is also where we both find out whether the model fits.
2. Pilot with one customer. We pick one of your customers, connect the gateway to its database, build the initial semantic layer and validate the answers with someone who knows the business numbers.
3. Assisted rollout. Once the pilot is approved, the Sozo Data team works alongside the partner on the first customers until the partner can repeat the process alone.
4. Operation. The partner sells and manages its customers under its own brand. Sozo Data maintains the product, the AI and the infrastructure.

Commercial terms depend on the partner's size, vertical and resale model, and are handled on request. Assisted implementation is billed separately. We do not publish prices on this page.

The white-label model does not fit everyone. Talk it through carefully if any of these describes you.

• You need an API to embed the question box inside your own application's screens. That integration does not exist today. Partners run Entendo under their own brand and domain next to the system, not inside it.
• Your system runs in the cloud and offers no database access or data export. Without database access, connecting takes an integration we build on demand, through whatever API or equivalent format the system offers, scoped in the project.
• You already have an AI team and want to own every layer of the product. Building may make more sense for you.
• You want a tool for your own use, with no resale. In that case, look at Entendo as a customer company.

If your case fits, the next step is a short conversation. If it does not, we will say so in the first meeting.