Productivity · News · 🇧🇷 Brazil
Vale uses Accenture AI to turn meetings into requirements and code in its ERP migration
In the Neo Vale project, Cognitive Digital Brain reviewed more than 270 requirements and brought the initial definition stage forward by up to three weeks in the move to SAP S/4HANA.
- Profession
- Engineer, Consultant, Developer
- Tools
- Cognitive Digital Brain, SAP S/4HANA
- Source
- exame.com ↗
- Country
- 🇧🇷 Brazil
- Report date
- November 5, 2026
- Verified on
- November 6, 2026
See more: AI for engineers · AI for consultants · AI for developers
What they did
Vale is migrating its ERP to SAP S/4HANA in the internal Neo Vale project, with Accenture as the main implementer. The current system connects to more than 350 others and has more than 30,000 users. In workshops with commercial, maintenance and finance teams, meetings were recorded and Accenture's Cognitive Digital Brain agent platform extracted requests, built the requirements traceability matrix, functional and technical specs, extension code when standard SAP was not enough, test plans and documentation. Consultants review the material and Vale approves it before any operational use.
How it went
According to Vale and Accenture in Exame's 5 November 2026 report, the AI reviewed more than 270 requirements and brought the initial definition stage forward by up to three weeks. For the miner, the gain was planning time, not shortening the two-year project. Vale says Neo Vale is on schedule and on budget. Accenture says it uses the tool with about 40 clients in Latin America, most of them in Brazil. Project cost and cash savings were not disclosed.
Good to know
Nothing goes straight into production: consultants review and the client approves. The figures and timeline come from Vale and Accenture, from a public source checked before publishing.
The prompt
I am taking part in the migration of [company]'s management system to [SAP S/4HANA or another ERP]. I will paste the transcript (or minutes) of a requirements workshop [paste the text with no personal or confidential data]. Extract the requirements as a numbered list, name the business area for each one, say whether standard system coverage looks enough or an extension is needed, and draft a half-page functional spec per requirement. Flag anything ambiguous and list questions for the next meeting. Do not invent a business rule that is not in the text.
Replace what is in [brackets]. Prompt written by us from the account, not a literal copy of the source. Do not paste other people’s personal data (privacy laws such as LGPD and GDPR apply).
Summary in our own words. Results as stated in the original account; we have not tested or audited them. Read the source ↗ Found a mistake?
Similar cases
01Allie Miller tests agents with an unusual request: buying a yak
Among several agents tested, Manus researched on its own how to buy a yak for the family.
02When your AI limit runs out in the middle of the day
A dev running four subagents across four repositories ran out of tokens and reflected on their dependence on AI.
03Running an offline AI assistant on your iPhone
A dev built Anvil, which runs models on the device for use on a plane, on the subway or without roaming.
04Estimating how long your AI plan will last before it runs out
A dev published an open calculation to estimate how many tokens fit into the Pro/Max plans.
05Building a "thinking system" with Obsidian and Claude
A developer started recording what he learns in Obsidian and using Claude to connect notes.
06Sending AI summaries and research straight to your Kindle
A developer built a Go MCP in minutes that lets Claude or Codex send texts to a Kindle.
07Setting up complete Discord servers with an AI bot
An open source bot uses Gemini to plan categories, channels, roles and rules from a description.
08Use a local AI to detect phishing
A security professional showed a Python script using a local model to analyze the intent of suspicious emails.