Project Learning Blog | Information Strategy

While working in the IT industry as an Account Manager for large enterprise clients across APAC and the EU, I saw critical project delays caused by issues that sat below the surface, such as broken commitments and ignored tasks. They led to poor customer experience, which hurt the top line in the long term, and to unforeseen costs that squeezed the bottom line and profit margins. That experience became the starting point for Slipstream, an AI-powered project governance tool that acts as a silent observer for project managers. It connects with Outlook, Teams and Slack, reads project emails, messages and chats, and gives project managers insights and recommendations on these hidden issues before they turn into project delays. The idea sounds simple, but the problem behind it is anything but.

Because the idea came from my own experience, I took on the part of introducing it to my team. I explained how complex IT projects actually work and made them familiar with the pain points and bottlenecks project managers face.

Initially, our main question was about adoption and trust: would an organization be willing to hand over its confidential emails and chats for the product to work on? But we had missed an important point, which was the effectiveness of the product, measured against a baseline. This is where the constant support and feedback we received really helped align our thought process in the right direction. Feedback from our professor broadened how we evaluated the idea, and it opened up a whole new dimension: working with project managers to create a baseline and test the efficacy of our product.

To do this, I built a prototype as a custom AI model in Google Gemini Gems and met with two ex-colleagues who work as project managers to discuss how the product would fit into their workflow. I also shared anonymized project artifacts with them, asked them to find gaps in communication, and compared their responses (the baseline) against our model’s output.

In the initial runs, our AI model performed well at finding the “signals”, but it struggled to recognize and ignore false alarms. So we refined the instructions of our custom Gemini Gem, and in subsequent tests it identified and filtered out false alarms well.

Throughout the project, one key question kept coming up: are we building this product to cannibalize the role of project managers, or to make their job easier? After several discussions around ethical AI and AI governance, we all agreed that the tool has to help project managers and should not be portrayed as their replacement. The prototype results made this belief stronger. The tool performed better than our human baseline, but, being an AI tool, it was also prone to errors and would deliver its best results working alongside a project manager.

The same thinking shaped how we designed the output. We did not want the product to just add to the pile of tasks a project manager already has and might overlook. So we decided to deliver the output as a daily digest: an email or message that updates the project manager on what is happening across their projects, with the option to open a dashboard and act on the issues most likely to cause delays. The full product is designed to track four signals, but for the demo and prototype we focused on the top two, the ones our qualitative research with project managers showed to be most critical.

We truly worked as a team, splitting tasks and making sure we stayed on track. Coming from different backgrounds made it a little difficult at times, and sometimes we had diverging ideas, which we handled democratically by voting. Early on, we worked without timelines, phases or milestones and made little progress. Ironically, we were building something to reduce project delays, so we decided to split our own project into phases and milestones, just as a project manager would. Once we did, we all responded well to the defined structure. We trusted each other with our responsibilities, and everyone made sure we delivered a good end product. Personally, this taught me that a structured way of working is far superior to an unstructured one.

As a Business Development Manager in IT industry, I was already aware of how AI can transform a company’s top line and bottom line. Working on the AI Strategy Lab gave me the chance to think more deeply about how AI can also transform small parts of our everyday lives. Demo Day in particular, where I saw many live AI projects under one roof, showed me that AI’s impact goes far beyond financial results. I understood that AI alone can’t be the answer to all our problems, but it can be applied to even the small aspects of our lives, and make them easier and better.