“twigbit moved us from AI experiments to AI in production. The support-triage agent paid for itself within months, and the assistant we built into Projektron BCS has quickly become one of our customers' favourite features.”
Francisco Josué Gutendorf, CEO at ProjektronClient
Projektron develops Projektron BCS, the web-based project-management software used by hundreds of organisations to plan, track and bill their projects. As an established software vendor, Projektron wanted to move quickly — and responsibly — on AI, both inside their own product and across their own teams.
Objective
Projektron came to us with three connected goals:
- Cut the cost and response time of customer support without growing the team.
- Build AI into Projektron BCS as a real product feature their customers could rely on.
- Adopt AI internally for the everyday work of their product, support and services teams.
We partnered with them across all three — from first prototype to production, with the evaluation and guardrails an enterprise product needs.
What we built
1. A production support-triage agent
We built an autonomous agent that reads every inbound support request, classifies it, drafts a grounded first response from Projektron's knowledge base, and routes it to the right queue — escalating cleanly to a human when its confidence is low. It runs in production against live traffic, monitored with automated evals so quality never drifts silently.
2. An in-product AI assistant (RAG)
We helped Projektron ship an AI assistant inside Projektron BCS: a retrieval-augmented search and Q&A layer grounded in the product documentation and the customer's own project data. Users ask questions in natural language and get answers with citations back to the source — turning years of BCS know-how into an on-demand expert. We owned the retrieval architecture, the evaluation harness, and the integration into their product.
3. An internal AI rollout
Beyond the product, we rolled out AI across Projektron's own teams — assistants and automations for drafting, search and repetitive operational work — with the governance and training to make adoption stick.
Results
- 3.9× — return on the support-triage agent in its first year
- −58% — median first-response time on support tickets
- 71% — of inbound tickets auto-classified and routed (≈40% resolved with no human touch)
- 92% — answer accuracy of the in-product assistant on our evaluation set, grounded in 14,000+ pages of BCS documentation
- ~6 hrs/week — saved per employee across the teams using the internal tools
From the support desk to a shipped product feature to internal tooling, Projektron now runs AI where it moves the numbers — built to enterprise standards of evaluation, guardrails and monitoring.