Before founding Perthshire, I worked as Co-founder and Chief Operating Officer at Propellus, an AI-first Singapore based travel technology company. Along with my fellow founders, I helped build Propellus from an idea in 2022 to a working platform, signed enterprise partners, and a valuation exceeding $10 million. This is the story of how we used AI to do it, and what it taught us about where AI implementation succeeds and where it fails.
At a glance
- Sector: Singapore based AI-first travel technology
- Role: Co-founder and Chief Operating Officer
- Timeframe: 2022 to 2025
- Outcome: Idea to a valuation exceeding $10 million
- The lesson: AI created extraordinary leverage in some parts of the business and almost none in others
The Challenge
Propellus set out to make travel more equitable for people who need a visa.
The idea began in 2022. The ambition was large and the resources were small. The founders put in the early capital, and the first objective was to build a prototype, take it to market, and prove there was real demand. This is the position most ventures start from. Limited funding, a wide problem, and a need to create value quickly.
My remit was broad. It covered HR, operations, admin, legal, and growth, including sales and marketing. I had run several of these functions before inside large enterprises, but always with full teams behind me. At Propellus I was carrying the same portfolio with one full-time junior operations manager, who was still completing her final year of study.
The operating context made the gap larger. Propellus was global from the first day.
- Investors based in South Africa, and angels supporting us from Singapore under Singapore regulations.
- Customers in the United Kingdom and Bangladesh.
- Suppliers across Vietnam, Pakistan, and India.
We had depth in legal, operations, admin, HR, and B2B sales. We did not have experience in travel technology, and we had not run a venture across this many jurisdictions at once. The work required capability and capacity the team did not have. That was the real challenge. Not access to ambition, but the ability to execute a wide, complex, global mandate with very little.
What We Did
We treated AI as the way to close the gap between what the business needed and what the team could deliver.
We started before ChatGPT existed. When it launched, we recognised early that this was a significant shift and a real opportunity. The constraint was practical. We had an enterprise-sized portfolio and a very small team. So we used AI to build the capability and capacity we were missing, and we did it hands-on, designing, implementing, and operating the systems ourselves.
When custom GPTs arrived, they became our delivery mechanism. We trained each one as an expert team member in a specific domain, and built agents to carry the recurring work across the portfolio. The point was always the work being performed, not the tools themselves.
✓ Expert advisors on demand. We trained custom GPTs as specialists in areas such as Singapore corporate law, Pakistan labour law, and SaaS growth and sales. They gave us qualified input across jurisdictions and functions we could not afford to hire for.
✓ Growth Agent. Partner and prospect research, outreach preparation, and go-to-market materials, so a one-person growth function could operate across multiple markets.
✓ Legal and commercial work. Contract drafting and review, clause analysis, and negotiation preparation for enterprise partner agreements.
✓ Finance and pricing work. Pricing analysis and unit economics, including modelling serverless architecture costs that were not available in any ready report.
✓ Admin and Writing Agents. Process documentation, internal communication, and the day-to-day operational output that keeps a lean company moving.
The insight was simple and it held. A well-trained AI agent could stand in for a role we had no headcount to fill. That allowed a very small team to perform like a much larger one, and it let us create real value on very little money.
This worked because the work was owned directly. We were building the systems, running them daily, and seeing the value as it landed. The benefit was immediate and visible in the work itself, from signed outreach to negotiated enterprise agreements.
The Outcome
Propellus moved from an idea to a real business, and AI was central to how a lean team got there.
By 2025 the company had grown from an idea in 2022 to a valuation exceeding $10 million. We had a working web app platform and signed enterprise partners. The progress was a collective effort across the founders and the team. AI was the leverage that let a small operation create value well beyond its size.
Results at a glance
- Valuation: $10M+
- Enterprise partners: 2 (Pakistan and Bangladesh)
- UK SME partners: 12+
- Team size: ~20 at peak
- Platform: Built and operational
- Early test conversion: 40%
What surprised us came next. AI created extraordinary leverage in some parts of the business and almost none in others. That contrast became the most important thing we took from Propellus, and it is worth examining closely.
Where It Worked, and Where It Did Not
AI created immense value on one side of the business and stalled on the other.
The business, communications, and customer acquisition side adopted AI quickly and gained from it. The founders embraced it early, and the commercial work compounded. The delivery and product side was different. Despite good people and an open mandate to move fast, AI adoption did not take hold there, and the result did not change in any material way.
We first saw this with an external development vendor. We expected the developers to adopt AI fastest. We even showed them a case where AI debugged their own code faster than they could, without writing a single line manually. The response was resistance and argument, and we were unable to implement AI with that vendor. At the time we put it down to the vendor, and assumed it would be different with our own people.
After funding, we built our own team. Good people, open to new ideas, with a mandate to move fast. We expected this to be the moment AI implementation finally succeeded. We ran AI hackathon weeks and pushed hard. The outcome did not move. Month after month this burned precious runway without the speed or the gains we had expected.
The pattern was consistent and worth naming plainly.
✓ The non-technical functions adopted AI and created value.
✓ The technical and delivery functions resisted it, and adoption did not scale.
That left the company operating in two worlds. One side was transformed by AI. The other was not.
Key Lessons from Propellus
Looking honestly at what happened, the cause was not the team. It was the approach. We had handed tools to people and expected transformation, without doing the work that makes implementation succeed.
- AI creates the most value when tied to a business outcome. Tools introduced without a clear goal fragment quickly and rarely scale.
- AI adoption is an organisational challenge, not a technology challenge. Capability and mindset matter more than the model, and adoption has to be led.
- AI requires workflows, context, governance, and leadership to scale. Without structured data and sound process underneath it, AI exposes the gaps rather than closing them.
We have since seen the same pattern across many other organisations. Capable teams, real ambition, and AI adoption that stalls because the foundations were never put in place. Propellus gave us the opportunity, and we created real value with it. It also showed us clearly where AI breaks inside an organisation, and that there had to be a better way to implement it across a wider team.
Why This Matters for Perthshire
The question that stayed with me after Propellus was simple.
Why did AI create extraordinary leverage for some teams and almost none for others?
That question is the foundation of Perthshire. We have lived both sides of it. We used AI to help take a company from an idea to a valuation exceeding $10 million, and we also watched AI fail to scale inside the same business. The success and the failure are equally valuable, because together they show where implementation actually succeeds.
Access to AI is easy. Scaling AI is hard. The challenge is implementation.
The organisations that win with AI follow a method. They choose goals over tools. They pick the right use cases. They redesign their workflows for AI. Perthshire has systematised that method in its proprietary AIOS Blueprint, so leaders can capture the value of AI through implementation that succeeds rather than experimentation that stalls.
We are open about the parts of the Propellus journey that did not work, because the lesson is the most useful asset in it. Perthshire's approach comes from real implementation experience, not theory. That is the work. Helping organisations move from AI ambition to operational reality, and create lasting business value in the process.