First published as a white paper “How the Hidden Costs of Inefficient Workflows Drain SME Profits — and How to Fix Them” on the 2nd of November 2025 “Case Studies” is the fourth chapter.
A recurring observation made of Octane’s client projects is the perception that they seem commonplace, the sort of thing that would be best served with something off-the-shelf, but appearances aren’t always what they seem.
Accommodation booking
Octane was introduced to a team fielding upwards of 2,000 calls during the summer months from a specific group of people who required accommodation near to academic institutions. Why? They were examiners, and their needs were much more specific than booking a room at a hotel.
What was the problem?
All of the calls were captured on written notes and then (when time allowed) put into a spreadsheet. The most significant problems were erroneous and lost bookings, and cancellations, each having financial ramifications. The challenge was to solve the problem of data error, loss, and duplication.
What was accomplished?
Within 12 months we had created a semi-automated workflow that allowed the team to manage thousands more bookings, each handled in a fraction of the time, not to mention a substantial reduction in cancellation costs. Because of this sudden widening of their productivity bandwidth, the service was expanded to more of their own clients.
Working with the team, the service evolved almost in real time, adapting to subtle and sometimes not-so-subtle changes in requirements. Such was the precision of each booking, we could attribute a cost to each action, allowing the team — for the first time — to know if a booking was within its assigned budget, and then make an informed decision.
Now the team could manage venues, had an audit trail of each booking, while management could perform complex searches to create valuable reports, and to populate invoices to be sent to hotels. In the end, we created a comprehensive workflow capable of accommodating tens of thousands.
Stock management
What had been built as a short-term fix became a long-term pillar to a workflow that managed the new build or refurbishment of hydraulic pumps for industrial use. The veteran piece of software was almost beyond serviceable use, and had to be replaced.
What was the problem?
Replacing such an aging system would have fixed one part of what were systemic problems within the workflow as a whole. A previous attempt was made, and failed, because those responsible created something how they imagined it should work as opposed to how the team needed it to work.
Adding parts to a job that was in progress had to reflect the state of stock across multiple jobs, or risk running out of stock, resulting in stoppages. The management of stock had nuance (managed by written notes, emails, and conversations) that had to be automated.
Imagine a printed sheet of A4 for a job listing the initial required parts. Now imagine this same job sheet passing into a workshop where men whose hands were thick with grease and dirt, had to amend it with a pen (adding, and sometimes remove parts from it with a scribble), before passing the sheet to the office for approval.
What was accomplished?
Within the space of two weeks from the initial sales-technical meeting, Octane built a working prototype of the core parts of the workflow: a multi-user dashboard; stock management; and quasi-invoicing (that encompassed the job sheet, among other things).
The team had a fondness for the tactile job sheet! So, we made it more compact but still legible (allowing room for more items to be added), added a watermark, and a version number, mitigating against the most common problem where the job sheet would end up on the wrong desk and acted upon when it wasn’t finished (parts would be ordered that sometimes weren’t required, and from the wrong suppliers).
Octane empowered a team and gave them access to data and information that hadn’t been accessible to them before, removing entire processes and improving those that remained, to create a precision workflow fit for a team of engineers.
An experiment in AI
As a business that creates software for a living, the recent surge in the abilities of AI agents has been an intriguing thing to witness and explore — but, computer scientists at the Model Evaluation & Threat Research (METR), a non-profit research group have claimed:
“… we find that allowing AI actually increases completion time by 19 percent — AI tooling slowed developers down.”
The study involved 16 experienced developers who work on large, open source projects, and this limited sample size did at least warrant some scepticism.
Having used AI as an assistant for the last 5-6 months, I decided to build my own version of Google Maps, as a technical exercise, to explore — to a limited extent — the potential of AI.
Viaje
Viaje is best thought of as side project to build an understanding of new technologies and techniques that were passed into commercial projects.
Viaje is a simplified Google Maps with route planning, but using open data, as is sources of data from the government, and those made free to use.
The most consequential part of the project was that I built it using AI from OpenAI (ChatGPT) and Anthropic (the range of Claude models), at a time when it wasn’t as commonplace a thing to do as it is now.
Using the AI, we built a plan of action consisting of two parts: the frontend (the client), which is the part of the application we interact with; and the backend (the host), which is where the requests we make are translated into responses, containing data.
Viaje was a success, but what did I learn?
- AI is excellent at building plans, and at executing specific tasks. On the whole, AI is terrible if allowed to execute an entire plan without human supervision.
- There was a lot of overlap between the two parts of the plan, and the agent(s) — in spite of understanding the connection — couldn’t implement something nuanced and structured alone, hence the constant guidance.
- Each agent is different: I found that ChatGPT could do the bulk of the work until it encountered what were to it insurmountable problems, where I’d then have to switch to Claude to get things moving again.
- A human-like absent mindedness would creep in from time to time where it would forget some technical specific, or something that it had already been done, and I would have to remind it! This loss is attributable to what’s known as the context window, analogous to short term memory in humans, but not as large.
- As the project grew the more precise its suggestions and recommendations became, demonstrating an understanding of the project.
- If allowed, the AI would keep adding and adding code, through a series of guesses, drawn from their formidable data models.
How long did it take to build Viaje?
I chose several cutting edge technologies which would have incurred a steep learning curve, so if we combine the learning with the actual building, 2-3 weeks of contiguous time would be a good estimate, in stark contrast to the actual 4-5 days it took the AI and me.
It’s worth pointing out that I was also learning how best to use AI as part of the software development workflow, so that also contributed to the amount of time it took to build Viaje.
Using AI in everyday tasks
I’ve enjoyed the most success with the AI when I followed these 3 simple rules:
- precision prompts;
- specificity of task(s);
- constant supervision.
In addition to software development, I’ve also used AI to do research, to brainstorm ideas, and then attempt to validate their fitness.
Here, point 2 is critical, in that it’s best to keep the tasks simple but additive, such that they’re chained: Task A contributes to Task B; Task B contributes to Task C and so on. Asking the AI to implement Tasks A through F is often when the problems begin.
AI sometimes gets things wrong, the same as we do, but the perception is that it shouldn’t. AI is not magic, and while that must seem obvious, a lot of the confusion I’ve seen has been in how people have attempted to use it, expecting magic things to happen from a prompt lacking specificity in instruction.
AI is nascent, evolving, flawed, but also compelling and promising. Remember that we are the training data of AI.