Three reasons you need Agent-Based Modelling for your business
The future of business is prescriptive analytics, and ABM specifically is what you are looking for.
Today markets are incredibly difficult to navigate. Sometimes taking a single bad decision is all you need to sink your ship.
There are many techniques out there that can help you keep control, even when the situation seems desperate. But none is focused on understanding a problem: it is always a matter of describing vs. predicting, never understanding.
Agent-Based Modelling (ABMs) is a powerful, consolidated, and inter-disciplinary simulation technique, rooted in the study of complex systems, that has seen a number of applications in many different fields, from cosmology to policy-making, including real-world business problems — i.e. from simulating the perfect marketing mix before starting an adv campaign, to simulating pedestrian fluxes to improve security measures, from traffic simulation in an urban context to prevent traffic jams, to simulation of monetary exchanges between banks to assess financial risk.
An ABM is a perfect approach to solve a problem with many parameters and complex, often non-linear relations among its elements. In our case, agents will be startups and companies or customers, depending on the chosen context.
Heterogeneity of agents (like for instance different personas in the consumers’ market) and different sources of noise will be the foundation on which we will build our models and explore scenarios.
Reason #1: human problems are complex
Of course they are: every time people are involved in some kind of collaborative effort to create/build/do something, interactions and relations and feedback are all elements that get in the middle of
Making sense of complex systems is, well, complex. Advanced methods of analysis are necessary to disentangle these complex mechanisms. So why model? Well, you definitely need to read this piece by Joshua Epstein.
Entering a market, especially a B2C market, needs a solid strategy because it is increasingly difficult to understand (not to mention predict) how social phenomena unfold and thus how people will react to your product or service.
Collective behaviours are not only hard to predict: they are often undesired at the level of the social actor. When the actions of many actors pile up, we can observe strong patterns at the global level that emerge as undesired consequences (have a look at the Schelling’s segregation model here).
Reason #2: you need robust scenarios to make better decisions
Mental experiments are a useful asset in the toolbox of these advanced methods — in a business context, this equates to push the analysis from the level of statistical correlations to causation, answering the question: why did it happen? However, instead of accumulating statistical evidence to back up some hypotheses, mental experiments allow proving that our hypotheses are sufficient, in principle, to generate (i.e. reproduce) the consequences we observed.
However, in order to explore such a level of complexity, we need computer simulation to help us understand the emergence of these complex and often unintended patterns: we just wait for the computer to finish its job!
Of course an agent can represent anything, from a node in a network to a customer or a company or a product. The idea is to disentangle the properties and behavioural rules of these agents in their specific setting, and then run multiple simulations to understand the central tendency of the system: if what we observe is similar to what we observe empirically, then our model can be a candidate to explain (not just describe) the world.
At this point, we have a solid grasp of the mechanisms involved in a particular phenomenon. What we can do now, is use the simulator to play with “what-if” scenarios, and explore different course of actions — before attempting those actions in the real-world, without risking money and reputation in business decisions that might have been better analysed thorough!
Reason #3: it’s fun!
Building a model is like building a map: maybe a scale of 1:1 will be too detailed (and probably useless), but a scale too small won’t help us understand the best path from place A to place B. Each task requires the right map but, of course, it takes a bit of experience to understand the right level of simplification and abstraction for each specific modelling task.
There are many course and summer schools out there open also to practitioners and aimed at teaching how to do ABM… One in particular, where I am involved, is the BEHAVE Summer School, which aims to train students on ABM in NetLogo by using modelling examples from the social science research.
An ABM is the perfect approach to solve a problem with many parameters and complex, often non-linear relations among its elements. In our case, agents will be startups and companies or customers, depending on the chosen context. If you are interested in more on the subject, drop us a message!
