Five days in India’s “Silicon Valley”: How Sii experts built AI tools at the ABB Accelerator 2026

More than 2,000 participants from 37 countries, over 600 submitted use cases, and two Stevie® Awards plus an International Innovation Award to its name – that’s ABB Accelerator in numbers, the global innovation program Sii experts joined in June 2026 in Bengaluru. In a few days we designed working AI solutions for real, everyday problems at ABB – and came home with one conviction.

A key takeaway from the week is simple: when you bring together the right people, a clear goal, and a few days of focused, shared work, AI stops being a promise and becomes a working solution. That’s how work on AI projects should look – fast, with effective communication, and across organizational boundaries.

ABB Accelerator is a global innovation platform, complemented this year by the AI Boost conference, product sessions, and countless opportunities to talk with people from industry and from India’s technical universities. Those conversations, between presentations and over coffee and kadak chai, carried the most energy.

Our speakers from Sii Poland – Bogusław Kosęda, Paweł Mazur and Patryk Wyżgowski – brought their own topics to that exchange, close to ABB’s business: predicting safety risks in industrial environments, AI-augmented field service built on Copilot Studio and Azure AI Foundry, and parameter-correction recommendations built from machine-data streams in waste management. Please find the key thoughts of the presentation summarized by one of the speakers.

“Predicting a safety hazard in industrial environment before it reaches a worker is no longer the hard part — that layer is already production-grade. The hard part is the line: AI predicts and advises, a human commits the decision, and a certified, deterministic function is the one that acts. The winning architecture isn’t the most autonomous one – it’s the one an operator trusts and an auditor can defend. The technology was never the obstacle. Trust is” – says Bogusław Kosęda, PhD, Senior AI Architect, Sii Poland.

Apart from taking part in the AI Boost talks, Sii experts built four hackathon solutions, of which brief summarization can be found below.

Too many approvals, too little time

When someone at ABB wanted to launch a new AI project, three teams: IT, Legal, and Data Privacy, each had to sign off, each running its own separate process. The person with the idea explained the same thing several times and waited a long time for an answer. We built a single AI-assisted process, AI4AI – Accelerate AI use case review, in which all three reviews happen together.

Result: faster, clearer, and without the frustration for whoever submitted the idea.

Invoices that sort themselves out

Invoice handling in a large, multinational organization easily becomes blurry – it’s hard to track what stage a document is at and who is approving it. With Common Report Source, every invoice lands in a shared inbox, catalogs and describes itself, and the right person approves or rejects it directly from Microsoft Teams, with the status updating automatically.

Result: finance always knows where things stand, and invoices are easier to pay on time.

“Invoice approvals rarely stall because people are slow – they stall in inboxes, manual handoffs, and when nobody knows where a document actually is. We put the whole flow on Power Platform: every invoice catalogs and describes itself, and the approval happens right in Teams, where people already work. That was our biggest learning – simplicity, security, and a Teams-first experience are what drive tool’s adoption. Automation only counts when people actually use it” – says Paweł Mazur, Senior Microsoft 365 Consultant, Sii Poland.

Ask the assistant for cloud resources

Getting cloud compute resources usually means forms, permissions, and waiting. With Conversational Azure Resource Provisioning, you simply ask ABB’s internal chatbot – behind the scenes, the system checks permissions, fills in the technical details, and routes the request for a brief human check.

Result: easy, secure, and with a full record of who asked for what.

“An assistant that provisions cloud on its own is a security incident waiting to happen. The point is that the AI never holds the keys – every request carries the user’s own identity, gets checked against their real permissions, and lands as a pull request someone reviews before anything is created. The conversation is flexible; the execution is deterministic and recorded. You get the ease of just asking, without giving up who did what” – comments Patryk Wyżgowski, Senior AI Engineer, Sii Poland

Laser-sharp focus on the Risk Assessments Review

One of the spectacular PoCs was related to improving the experience of the Activity Based Risk Assessment (ABRA) review. A custom interface – minimal and optimized for the task, a chat bidirectionally linked to guidelines and key sections of the documents was sure to boost the mundane process part. It also enabled self-service for the submitters – a refinement opportunity to reduce review time even further.

Result: more documents reviewed in the same amount of time, with less effort on the part of the person responsible for reviewing them.

“A generic chatbot will happily tell you a risk assessment looks fine from memory, with no source you can check. That’s exactly what you can’t have for safety. We built the standards in, so every finding cites the rule it comes from and you can click ‘Why?’ to see the reasoning. The assistant doesn’t replace the reviewer’s judgment; it hands every reviewer the senior expert’s checklist, with the evidence attached. Consistent, traceable, and fast enough to actually use” – says Jacek Skoczylas, Senior AI Engineer, Sii Poland

• 
  
  
• 
  
  

• 
  
  
• 
  
  

These weren’t demo-only prototypes

It’s easy to assume that something built in a few days of a hackathon is just a demo that will never reach daily work. Bengaluru showed otherwise: AI4AI, Common Report Source, ABRA review support, and the Azure resource assistant respond immediately to recurring, everyday problems at ABB – approving AI projects, processing invoices, accessing cloud infrastructure, time-consuming reviews. These initiatives got warm reviews and some of them immediately got traction for bringing them into production. We observed that short build time does not mean a short shelf life.

Lessons from Bengaluru

Here are some insights brought by Sii Poland Team from ABB Accelerator 2026:

• Put a team with different skills and from different organizations at one table, instead of splitting the work across distant departments.
• Give it one specific, narrow problem – not a vague “let’s do something with AI.”
• Shorten the feedback loop: a few days of focused, shared work beats months of email back-and-forth.
• Treat organizational boundaries – IT, Legal, Finance, different countries – as something to cross, not accept.
• Thank you to ABB for inviting us on this adventure – we can’t wait for the next one.