Dario came up with the idea to cut down empty container trips in Rotterdam and Yet, Ruben and I quickly saw the opportunity. We chose Track B because most of us had build prototypes and three of us had completed an MSc in Strategic Entrepreneurship. Our goal was to make street turns not just possible, but routine throughout the entire service area of the port of Rotterdam. A street turn happens when an import container is immediately reused as an export container after being emptied.
One of my responsibilities was the economical analysis of our idea. To do this, I made a value model per turn. Through this model, I concluded that a succesful turn in the base case would lead to about 80 kilometres and two hours of driving saved. The 80 kilometres not having to be driven will save €54 and those two hours will result in another €110 saved, but only when those hours are productively resold. This gave a gross value of €164 per succesfull turn. The total costs coverd about €52, leaving a net value of about €112 per turn.
Following this, I created a break-even table to see the feasibility when the basis under our most important assumptions change. I also checked what it would costs if a match was wrongly accepted, which caused for us to set an acceptance threshold of the platform much more conservative that originally planned. Mainly due to the fact that one mismatch costs both a wasted trip, but also fees for keeping a carriers container for longer then agreed, called a detention fee. These two combinded easily negate the positive value of the turn.
What mostly surprised me was the cost of the driver time. If the hours that get saved do not get resold, a turn is only worth about €2. This was also shown by Dario’s simulation, since our demo turn was worth €147, while the average is €49.
I found it most difficult to decide what to use in the analysis and what not to use, since every new stakeholder througt of, added multiple new angles on both the costs and income side. I took to my team and LLM brainstorming sessions to discuss where to draw the line.
My most important decision was the pricing strategy. At first, we wanted to charge per seat + a succes fee for each trip. However, I suggested a change since I believed it to be necessary for the model to be viable on a long term basis. The main reason for my believe was BCG’s report saying that seat-based pricing would dissappear since agents will reduce the amount of seats necessary for succesful completed work. We researched the options as a team and decided on changing the strategy to a share on savings pricing model. In the end, we had two more main reasons for this change:
- The AI costs was not high enough to have that be the main driver behind the pricing
- During one of the classes, we talked about incentive compatibility, making us realise the small trips could end up in a net loss in our earlier pricing strategy.
By switching to a share of the savings we wanted to align our objective and incentive to the clients’. This being: Save the client as much money as possible.
From this experience I have learnt that pricing is not something that should be the final stage in a business model, but it is something you can tweak througout the process to make sure your product benefits who you want it to benefit.
The most difficult situation was when we came across Avantida. Their whole product was a solution for the same issue and they were already managing about 1,600 turns each day. This made us see that just matching the supply and demand of containers was not enough to make a differnce and impact, since Avantida was already linking 2,500 companies and 12 shipping lines.
Finding this was really demotivating because you think your idea is in the pits now. However, it ended up being the most useful moment because we had to think of a pivot.
Rather than just supplying our clients with an app for them to search themselves, we wanted to proactively give them signals from analysing information that forwarders already have in their booking emails to find the best time and cost-saving options for our clients
My other responsibility was testing our streetturn idea against the future agent market. Since this product needs physical movements to be made in order to work, the scarsity is in the physical world. Agents will probably make coordination and analysis essentialy free, but they don’t effect who owns the container, the condition it’s in or who is liable. So I created four scenarios to test. A verification API did not score negative in any of them. So our strategy was to begin with a dispatcher tool, but to develop all future features in such a way that, through an API, any agent can use it.
We also listened to the feedback received on our proposal. There it said that value capture is generally weaker than value creation and that we should focus on things that could kill the model. Dario’s tests showed that AI handled all 45 emails correctly, compared to 28.9% for the rule-based reader, but dispatcher acceptance is still untested. This is why we recommended a bounded pilot and not a full scale-up.
We were working on the same document, each of us being responsible for our own section. My financial analysis was based on Dario’s simulation and Ruben’s governance section made use of my figure for the cost of wrong matches. The figures also differed between the document and the presentation deck, which we had to fix last minute. Next time, we should use a single source for all the shared figures from the beginning. The fact that we communicated mostly through WhatsApp and very brief real-life meetings made this extra difficult. This is something to be improved on as well by meeting more in person.
This project has altered my view of the role of AI in business. Initially, I thought that introducing AI into most processes would automatically lead to improvements. However, I have since realised that it is important exactly how, where for which functions you apply AI. The problem isn’t that the AI lacks intelligence, it’s that it can’t make the correct decision on its own 100 per cent of the time. There is still no definite answer as to who is to be held responsible if the AI makes a costly error. That is the reason why in our design the AI merely reads the documents, fixed rules determine what is possible and a human gives the final approval. Since AI efficiency is becoming more and more available to all, this means value is more easily created, but not neccesarily captured (Barney & Reeves, 2024). It is the physical assets, such as the carrier’s container, network density and accountability, that remain scarce. For future projects, including possibly my own startup, I will first ask: what stays scarce once everyone has the same access to AI?
Reference
Barney, J. B., & Reeves, M. (2024). AI won’t give you a new sustainable advantage. Harvard Business Review, 102(5), 72–79.
