Can a travelling salesman fix our cities?

Decisions. An average adult makes 35,000 a day. That’s a lot of choices. Many of them are trivial: what we wear, what we eat for lunch, what we say. Some are less so. Where do we build a new hospital? When should we schedule an important delivery? What’s the fastest way to get from A to B? The answers to these questions determine how we interact with cities. But do we make these decisions well?

I don’t think we do. I also think this “decision drag” has a direct impact on us. Mispositioned hospitals, inadequate warehouses, and inaccurate scheduling affect us all. Resources stretch, delays compound, traffic builds, and we suffer. So how do we improve our decision-making?

Luckily there’s a dedicated field of study for this: operational research (OR). OR, coined by a British military scientist in 1940, referred to applying scientific methods to improve the military. As we have gotten better at maths and computer science, OR has since expanded to other domains. Today, it’s used in manufacturing, transportation, logistics and other sectors to make real improvements. But what does OR actually do? How can we use it? And what’s the benefit for us?

More than a feeling

This all sounds quite abstract. What problems can OR solve? The classic example, and the inspiration for the name of this editorial, is the travelling salesman problem. The travelling salesman (or salesperson) problem goes: “Given a list of cities and the distances between each pair of cities, what is the shortest possible route that visits each city exactly once and returns to the origin city?” While it seems simple at first, to date there is still no official solution to this problem. Instead, we have different algorithms which provide rough solutions to the problem. Once an algorithm is found, we can apply it to any OR problem. Armed with this, computers can find a good route, but finding the “perfect route” is still hard. This has real-world implications. Imagine you’re trying to deliver parcels to 500 homes. The travelling salesman is trying to find the fastest route that covers each delivery destination with minimal backtracking and time-wasting. A bad route leads to extra delivery time, and extra petrol. Extra delivery time means one driver might not be enough, increasing costs further.

This is just one problem in this broader OR area of graph theory. Graph theory is a way to represent and simulate the flow of objects between points (called nodes). We can see this by looking at the other OR problems in graph theory. They all have names like the Seven Bridges of Königsberg, or the Chinese postman problem, or the London Underground Tube Challenge. These problems are generalised ideas of common questions. The questions include how we plan routes, or where’s the best position for a common resource? By solving these generalised problems, we have answers to any specific question, no matter the details. Algorithms designed to answer the travelling salesman problem are now used in chip design and DNA sequencing. Semiconductors and genetics are wildly different, but OR provides a set of tools which we can apply to any discipline. In fact, OR is used to optimise the flow of electricity through the grid. This is a vital part of balancing our energy loads, and OR has kept the grid running today.

A more common area of OR is optimisation. Optimisation problems are relevant when there is a value we want to maximise, minimise, or keep constant. Grocery stores and retailers use OR to optimise their inventory volumes. They want to minimise empty shelf space, without having excess inventory. OR can find the perfect balance. Another example is queueing. There are people waiting in line. What’s the best way to minimise customer wait times without spending too much on empty counters? This has applications beyond restaurants: telephone exchanges used queueing theory to process calls.

Optimising for optimisation

Clearly OR is responsible for many aspects of modern society. But, I don’t think it’s running at its full potential. We have our arsenal of OR techniques, but we’re not using them enough. I spoke to Rango, the CEO and Founder of a startup called heySimulate. heySimulate works with dispatch and logistic startups to optimise their operations. Before founding heySimulate, Rango worked and consulted for logistics incumbents and startups. According to him, incumbents sometimes use data in decision-making, but exclusively for low-frequency and high-impact decisions. An example would be where to build a new depot. The other decisions (high-frequency and low-impact) were made entirely by gut feeling. These decisions include the number of vehicles needed in the fleet. The combined impact of these lower-impact decisions outweighs the rarer high-impact decisions. These smaller decisions are also responsible for the majority of the costs. This is an obvious area for improvement.

In theory, startups are better placed to adopt OR. Sadly, it’s harder in practice. Startups are under phenomenal pressure to maintain KPIs while scaling revenue. Maintaining a 99% delivery-on-time rate while trying to triple revenue year-on-year is a huge challenge. This occupies everyone’s time. OR is powerful, but it takes time to set up and implement. Sometimes it needs experts too: for example, Rango has a PhD in engineering from the University of Oxford). It doesn’t make sense for a resource-stretched startup to implement OR. They have other fires to fight.

It seems incumbents and startups both could do with OR support. Could new startups fill this gap? A few months after launching, heySimulate already has traction. It seems like the market is there. But is this market interesting to us at PT1? The challenge facing the built environment and real assets is huge. Anything to help make these sectors more productive and more efficient is a good thing. But for a market to be venture-backable, it needs to satisfy the following:

1. Is it big enough to incubate potential unicorns?
2. Is there a “hair-on-fire” problem in this market?

The global markets for logistics, infrastructure and construction are in the trillions. If better decisions can increase profits by a few percentage points, then the market is large enough. A few percentage points might be an understatement. In fact, Germany’s most valuable startup, Celonis, does this for all sectors. This Technical University of Munich spinout uses process mining to improve operations. Based in Munich and last valued at $13b, Celonis has helped clients such as Deutsche Telekom save over €66m. We also have a portfolio company in this space (albeit earlier than Celonis!) Proptly uses process mining and other OR techniques to accelerate ClimateTech installs. Their tech enables climate hardware OEMs and distributors to better coordinate external contractors. This sounds a bit abstract, but the results are clear. Proptly helped one client double their profit margins on climate tech installs.

Incredibly, this increase in margin was two-sided. They increased revenues by automating the “self-serve” parts of the customer journey. This increased conversion rates and also opened up new sales channels. They decreased costs in two areas. Firstly, their automated and better processes led to a c80% decrease in customer success costs. Secondly, better installer coordination and routing decreased installation costs. Proptly’s revenue model also means the payback time for using Proptly is immediate. It’s hard to say no to that proposition - and we’ve seen this from their traction: they’re already working with the world’s first and third largest EV chargepoint manufacturers. Clearly, the market is large enough to be interesting.

We just need a hair-on-fire problem to incentivise companies to adopt this technology. This is harder to determine - is suboptimal decisioning a real problem? Why should companies care? One reason is the magnitude of work needed for the net zero transition. Over two billion bits of hardware must be installed for electrification, including smart meters, EV chargers, and heatpumps. And OR is relevant for this. A heatpump distributor whose employees conduct site visits is an example of the travelling salesman. Inventory is just an optimisation problem. We know OR can fix this.

But why does it matter now? It never mattered before, so what does the net zero transition do? It adds urgency. We are all faced with a global challenge, larger than what’s come before. Throwing money and resources at the challenge isn’t enough. The past few years have pushed and stretched our supply chains to the limit. Firstly, Covid-19 affected global production and demonstrated the limits of just-in-time supply chains. 2021’s Suez Canal blockage was another example of supply chain fragility. Today, geopolitical tensions and climate change are continuing to add pressure. Celonis’ success stories show us the potential of OR. Implementing these techniques will take some pressure off, and lead to some quick and recurrent wins.

The PT1 verdict

The market is big and the problem is painful. This is a good space for new startups. And off the top of my head, I know some areas where the built environment could do with help:

• Construction procurement and delivery: construction delivery is hard because “packages” are differently shaped. Loading delivery vehicles could be optimised. This would minimise vehicle mileage, cutting emissions while saving time and money.
• Portfolio retrofit planning: real estate owners need to retrofit their portfolios. They can do different things: insulation, a heatpump, solar etc. However what they decide to do can vary based on their goals. Maximising the amount of renewable generation is a different goal to reaching EPC C across the portfolio. This is another OR problem: given the goal of the owner, what actions should they do and in what order?

There are plenty of examples I haven’t listed. Ultimately, OR can be used anywhere, as long as the underlying problem can be modelled mathematically. OR can’t make you a better poker player, because poker has a human and therefore non-mathematical component.

So OR can help in many different areas. But is it limited to only helping large corporates? No. HeySimulate and Proptly are working with startups and scaleups. The benefits can help anybody. The market for better decisioning is here. But what are some concerns and things to think about?

1. Trust. OR is a simple idea, but often involves complex mathematical models and a lot of buy-in from the client. Building trust is a challenge. There’s no third-party verification here. Founders must be able to convince potential clients that the solution works.
2. Complexity. Think about the problem space you’re operating in. What are the sources of uncertainty in the input data? What’s the goal for the output data? This affects how founders should build the tech. There’s no point investing engineering resources to build an accurate model if the inputs are bad. Similarly, for some goals, a basic solution might be enough, instead of a perfect solution. This matters when finding the perfect solution would require more processing time and more computing costs.
3. Scalability. Pick a problem space that’s big enough, and painful enough. A super niche problem won’t give a big enough market. Or, you’ll end up with a feature, not a product.
4. Pricing. A SaaS style flat license fee isn’t always the best approach. Experiment with other pricing models. Proptly learnt this from experience: switching to a transactional fee, where they get paid on every successful install, decreased their sales cycles and increased their lifetime customer value.
5. Defensibility. You need a lot of input data to make these solutions work. Who has the most data? The ERPs. It’s better to get them onside than to compete against them. Celonis quickly built a relationship with SAP, and this helped them scale.

I’m starting to see more and more startups in this space coming through, but it’s still overlooked. If you know someone building in this space - ask them to contact me. To paraphrase Rango, engineers simulate the design of a car, a building, or a bridge thousands of times before they finalise it, all to make sure the decision is perfect. We should do the same everywhere.