Incentive is all you need

During my studies, specialising in “Entrepreneurship & Innovation”, one question fascinated me above all: why does the same service work brilliantly with one business model and get completely displaced by another? What is this “business model innovation” at its core?

The textbook example is Uber versus the taxi. A ride from A to B, the service is identical. Or Airbnb versus traditional holiday rental agencies. That digitalisation plays a role is clear. But on its own it does not explain the disruptive force, and this has been discussed at length in the literature. Network effects, for example, are another building block: every additional participant makes the system more valuable for everyone else. The business model decides whether this dynamic can emerge at all.

At Thinkubator we work intensively on the economic viability of circular business models. How do we get business model innovation right so that we close material loops? Business model innovation alone is hard enough. But when it is additionally restricted by given constraints (think 10R principles) and has to be modelled towards a concrete target picture (think climate-neutral circular economy), it gets really hard.

Despite all the advantages of the circular economy described in the literature and presented at conferences, the linear model still wins far too often. Our globally declining circularity rate proves it.

Circularity is a tough nut to crack.

And more than ever I am convinced that one thing above all is needed: business model innovation. At the end of the day, a business model is a construct that moves different parties to cooperate through incentives.

How strongly incentives work is shown by the linear model. We all know the example: lamp manufacturers that build a light bulb that lasts forever threaten their own business, because the market eventually becomes saturated. So they do not invest in durability. You can think that is good or bad, and of course there are nuances and exceptions. But broadly speaking, that is the incentive situation.

That does not mean technology and regulation play no role. But if you want to implement something concrete within an existing framework, we stick to the credo: Incentives is all you need.

In my spare time I have been working with machine learning, neural networks and reinforcement learning for several years. What fascinates me is the history of the field. Decades of theoretical basic research were followed by a phase of intensive practical application. The decisive breakthrough to modern AI as we know it today finally came in 2017 with the paper “Attention Is All You Need” and the introduction of the Transformer architecture.

Put simply, it was about this. Take the sentence “The dog followed the fox into its den.” Older models processed the word sequence step by step, from left to right, and each step only referred to the one directly before it. That is difficult here, because only the context shows whom “its” refers to, the dog or the fox. By the time the models reached “its” or “den”, they had often already “forgotten” the beginning of the sentence.

This is also the problem in many linear business models. We go through the value chain from start to finish and lose sight of which point is responsible for which outcome: what effect do product design decisions have on recovery? Where does the choice of retailers determine whether a product can be taken back at all?

“Attention is all you need” turns this principle around. Every element in the sentence asks at the same time: where do I stand, and how am I related to all the others? These relationships are weighted and computed in parallel, depending on the whole context. The model looks at “its” and simultaneously draws connecting lines to “dog” and “fox” in order to compute whose den is most likely meant (context weighting).

This is exactly what we need to do in circular business models: apply context weighting when modelling. Which point relates to the others in which way? What does it need from them? And how can we incentivise it? That is the step from “Attention is all you need” to “Incentives is all you need”.

That was the thought that has stayed with me since. The step from the chain to the relationship matrix in artificial intelligence is analogous to the step from the linear economy to the circular economy.

Whether the analogy holds and how far it can be pushed, I do not know. Whether tools or algorithms can be developed from it is equally open. Testing it will probably be at the core of our next projects at Thinkubator.

Classic value chain analysis asks: who supplies me, whom do I supply? One step up, one step down. For linear chains that is enough. In a loop, however, the value of a bottle, a battery or a component depends on many actors at the same time, and to different degrees depending on the situation. That is why I propose the Incentive-is-all-you-need concept, based on three ideas.

Idea 1: relationships with everyone

The framework describes a network as a matrix. The actors are the points (in network analysis, “nodes”), and the connecting lines between them are called “edges”. In the text I mostly speak of relationships. Each cell in the matrix asks: how strongly does actor A influence the economics of actor B? What matters is that this weight is not fixed. A used bottle means something different to the washing plant than to winemakers, and for both its value changes with the condition, location and timing of the bottle.

Idea 2: layers and states

A Transformer looks at the same data with several “heads” at once, and it receives several pieces of positional information. For us the analogy is: we need not just one matrix but several layers, one per type of relationship, because relationships in business models have so many dimensions: material, money, information, liability, regulation, reputation and many more.

Every relationship is also a function of the state, that is, of location and transport costs, the quality condition of the bottle and the current situation of supply and demand. Without this information the matrix would act as if a bottle were worth the same whether it is 20 or 400 kilometres from the washing plant or whether it is damaged.

Idea 3: incentives as currency

The core question per relationship is: does actor B, under realistic conditions, have a reason to do what the business model assumes? That is, return bottles, collect them sorted by type, share data, guarantee a residual value. Where the answer is no, an incentive gap arises. That is exactly where the cash flow depends on someone who has no reason to play along.

Not all of this can be expressed in euros. Lobbying, image and veto power can hardly be valued in money in a meaningful way. That is why the model should work with three categories.

Limits

If you build the model, the weights are expert estimates at first, not data. And once you start collecting and deriving the weights from reality, the learned weights are observed behaviour and not necessarily causality. Whoever learns from the status quo learns the linear status quo.

In an early edition of the Thinkubator Fellow Programme, Sebastian Seisel, Michaela Zuckerhut and Kilian Mazagg worked on the circularity of wine bottles. Almost every wine bottle in Austria becomes waste glass after a single use. This means new glass is melted, raw materials and energy are consumed, and the bottle is replaced even though it is usually undamaged.

A life cycle assessment from an Italian wine consortium concludes that reuse is superior to glass recycling in all impact categories considered.³ The avoided new production outweighs the effort for washing, sterilising and drying.

A 2005 study by the State of Styria (using mineral water as an example) reports an average of around 32 cycles for reusable glass bottles. It also notes that the ecological advantage shrinks with transport distance and that individualised bottles can hardly be refilled regionally.²

Step 1: Constraints

The 10R principles rank strategies from “Refuse” to “Recover” according to how much material and energy they save. Three apply to the wine bottle: Rethink (changing the business model), Reduce (producing fewer new bottles) and Reuse (using the same stock several times, according to the Styrian study around 32 times on average).

Today the bottle ends up in recycling, which is much lower on the ladder. Added to this is a second constraint, transport distance: the further the bottle travels to the washing plant, the smaller the advantage.

Step 2: Target picture

The target picture is the climate-neutral circular economy. The circular economy follows from the principles of step 1. For climate neutrality, the life cycle assessment from Italy helps: reuse performs better than recycling in all impact categories considered.

This is not an absolute assessment. The negative climate value is a credit for avoided new production, not a real removal, and the study applies to a regionally compact case. But the direction is right.

Step 3: Value network

Now all actors come to the table, including those that seem to be on the margins at first glance. The bottle runs from manufacturers via winemakers to retail and hospitality, to consumers and via take-back and the washing plant back to winemakers. Above them are actors who move nothing and still decide: who finances, who is liable, who sets the rules.

Step 4: Relationship layers

Between two actors there is never just one relationship. Winemakers and retail exchange bottles, money, information and liability. In our concrete case, the following layers seem sensible.

Each plate is a layer, i.e. a type of relationship. The same actors stand on each plate, with the same number, and the arrows show who depends on whom on that layer. Winemakers (1) and retail (2), for example, are strongly connected on the material layer because bottles and empties move between them, but hardly at all on liability. The arrows have a direction, because how strongly B depends on A is not the same as the other way round: winemakers deliver the filled bottle to retail (strong), the empties come back more weakly. Only the stack makes these differences visible. The table below lists for each layer what concretely flows for the wine bottle and which guiding question it raises.

Step 5: Weighting

Not every layer is equally relevant for every actor. Investors want predictable cash flows and therefore look first at money and information, because without data on the number of cycles there is no reliable residual value. A public authority would probably give regulation more weight. That is the context weighting from the attention idea.

Step 6: Assessing the incentive gaps

Now we assess: where does an actor have too little incentive for what the model needs from them? To do this we fill in the matrix. Each cell rates the incentive gap of an actor on a relationship layer, from 0 (none) to 3 (large). Multiplied by the weights from step 5, this gives a weighted gap sum per actor. The values are my assumptions for our example case.

Our assessment of the actors at a glance:

This is offset by effects that do not accrue to the investors: value creation in the country through washing plants and regional transport, and less dependence on glass imports. Subsidies and extended producer responsibility are possible levers. How reuse is treated there today is something I am still checking.

Step 7: Measures

With a better understanding of the incentive gaps, we can finally consider which mechanisms we can use to create new incentives.