# From weeks to less than a day per dashboard at an investment advisory firm | Tog Lab

> Tog Lab replaced the BI tool with a custom web application and installed AI-Accelerated, Tog Lab’s model for AI-accelerated development, in the client’s environment. The business team now requests dashboards in plain language, approves the plan before execution and ships improvements reviewed by a T

URL: https://www.toglab.co/en/projeto-bi-investimentos.html

[Português do Brasil](https://www.toglab.co/pt-br/projeto-bi-investimentos.html.md) · [Português de Portugal](https://www.toglab.co/pt-pt/projeto-bi-investimentos.html.md) · [Español](https://www.toglab.co/es/projeto-bi-investimentos.html.md) · [Italiano](https://www.toglab.co/it/projeto-bi-investimentos.html.md)

[Projects](https://www.toglab.co/en/projetos.html) BI with AI agents

Financial servicesAI-accelerated developmentService as a Software

# From weeks to less than a day per dashboard at an investment advisory firm

Tog Lab replaced the BI tool with a custom web application and installed AI-Accelerated, Tog Lab’s model for AI-accelerated development, in the client’s environment. The business team now requests dashboards in plain language, approves the plan before execution and ships improvements reviewed by a Tog Lab professional.

Illustrative screen with fictitious data.

Preliminary metrics from July 2026, measured against the client’s own previous process, with a small team and the migration still underway. Well-defined domain and data already organized: on legacy systems the gain tends to be smaller.

## The challenge

The client published metrics for advisors and managers in an off-the-shelf BI tool. The databases and business rules already worked well. The problem was the path: every new dashboard depended on a technical person, several rounds of manual validation and one more license for each user.

Requests arrived as spreadsheets, sketches and prototypes that showed the intent but rarely the calculation rules, the exceptions or the acceptance criteria.

## How a request becomes a dashboard

Six steps for every request. Agents handle triage, planning and execution. Two people approve: the requester approves the plan and a Tog Lab professional approves the delivery.

Who does it

Approval

Agent workHuman work and approval

The workflows became simple commands for requesters: /nova-feature (new feature), /melhoria (improvement), /analisa-bug (analyze bug), /corrigir-bug (fix bug) and /deploy.

## The AI works only with the structure of the data

Agents read schemas, API contracts and business rules to plan and build. Queries go through a server-side layer that decides what each role can see, and credentials never reach the model.

Everything runs in the client’s environment, with the client’s own AI model key.

## Twelve days from kickoff to production

Timeline from July 2026.

### Before and after

AreaBeforeWith AI

A person from the business team, with no technical background, received an introductory training session, ran the project on their own machine, validated the changes and opened Pull Requests without help. The first ones needed no fixes, and new visualizations emerged that did not exist in the old tool.

## Does your operation depend on a BI tool?

A backlog of dashboard requests, license costs that grow with every user and business rules living in spreadsheets are signs that this model fits your company.

[Talk to us](https://www.toglab.co/en/contato.html) [Next case study: IDS Social](https://www.toglab.co/en/projeto-ids-social.html)

## More on this page

- 3 business days — for AI-accelerated development to go live in the client’s environment
- Less than 1 day — per complex dashboard. Before, weeks
- 5 of 13 — dashboards migrated in the first week
- Zero — licenses per new user

- Constant rework — Implicit requirements and differing interpretations led to repeated rounds of adjustments to the dashboard or the database.
- Validation at the end — The client only saw the dashboard once it was finished, when fixes cost more.
- Licenses limiting access — Sharing a dashboard with more people meant buying and managing new licenses.
- Concentrated knowledge — The tool and the databases depended on a few technical people.

- Request — The requester, on the business team — No approval at this step — The person describes what they need in plain language, the way they would explain it to a colleague: the metric, the breakdown, the target and who will use it.
- Triage — Agent orchestrator — No approval at this step — The orchestrator classifies the request (new dashboard, improvement or fix), checks whether the data already exists and asks any missing questions before moving on.
- Plan — Planning agent — Goes to the requester for approval — The agents put together the execution plan: data source, calculation rules, role-based permissions, files that will change and the tests that will prove it works.
- Plan approval — The requester — Approval recorded before any code — The requester reads the plan and approves, adjusts or rejects it. Nothing is built before this sign-off, and the record stays in the request history.
- Execution — Code, test and review agents — Tests and automated checks — Specialized agents build, test and review in parallel, each within its own permissions. Build, security and regressions are checked before delivery.
- Validation and delivery — The requester and a Tog Lab professional — Technical review before production — The requester validates the dashboard on their own machine and the /deploy command opens the Pull Request. A Tog Lab professional reviews the code, and only then does the change reach production.

- Web experience — Interactive dashboards, with a role-based interface and access without per-user licenses. — Chakra UI
- Server-side layer — Authentication and role-based access decide what each user sees. — Role-based access — Internal BFF — Server-only credentials
- Integrations and data — Existing databases and rules became the raw material, with no data migration. — Metrics APIs — Swagger contracts via MCP — Consolidated databases
- AI harness — What surrounds the agents: context, limits, checks and approval gates. — Context documentation — Agents and skills — Rules and commands — Tests and gates

- Jul 2 — Kickoff
- Jul 6 — Production ready
- Jul 8 — First autonomy
- Jul 13 — First deliveries
- Jul 14 — Current phase

- Weeks for complex dashboards — Less than a day in the measured cases
- Manual and at the end, after building — Plan approved up front, local validation and Pull Request
- Fully dependent on developers — The business team requests and improves dashboards in plain language
- Tied to per-user licenses — Web application with role-based access
- Scattered across people, spreadsheets and formulas — Documented and used by the harness