Focus on Future Growth
The following article was originally published on the AEO blog. It summarizes and expands on a session that I facilitated at the January 2022 AEO CEO Forum.
Summary
In the ‘Focus on Future Growth’ session at the AEO CEO Forum in January 2022, Mark Parsons of Events Intelligence shared his perspectives on how organisers should think about data (and especially big data) to understand their communities. This article summarises his main points and responds to questions raised in the room, concluding with a set of actionable next steps for organisers.
Classification vs. Clustering
Mark started by exploring the human need to classify things. In his own words: “the mind likes order and wants to put things in boxes and categories, as these heuristics help us make sense of the world. People are either attendees or exhibitors, delegates or sponsors. You’re either in this category or that category. But the real world isn’t like that, people are just people, companies just do things. When you seek to classify things, the edge cases are always a problem, no one company is the same as another and no two individuals are identical.”
“When you start to classify data there are always cases which just don’t fit – given we all want to create order, we go back and think about whether our initial categories were wrong and add another. Show directors do this all the time, and many are very proud of their taxonomies, which help them understand their markets and often took many hours or days to devise”.
Then, he shared his experience in his own business, years ago. He sought to categorise tradeshows into sectors. “I looked at a show and worked out what fit into a taxonomy of shows, and then moved on… but every so often there was a show which didn’t quite fit. There was a category called “automobiles and motor vehicle shows” and it was one where I couldn’t work out what should fit – does a manufacturing show sit in this if it focuses on car manufacturers, is it right that an aftermarket accessories show is treated the same as a car show? What about bicycle shows, they’re sort of like motorbikes but not automotive…” The edge cases create challenges.
While it can be a lot of work, and hard to categorise things, it’s easy to cluster things. Mark argued: “This is not a new concept, it’s something we’ve been doing for years as show organisers. Rather than using software we’ve been using humans and calling it a ‘hunch’. We talk to people, we understand their needs, and then we spot themes or patterns which help us launch a show or expand into a new sectors.” Given finding people who are good at this is not easy – it is a scarce skill, after all – there is appeal in finding a way in which organisers are not reliant on talent but can use the data they have to identify initial ‘hunches’ for further exploration.
The Power of Graph-based Data
Mark set the scene: “as organisers we have lots of data, but it is highly siloed. Some is attributable to individuals and some not. Some of our data is of high quality (as we use it for sales) but a lot of it is of low quality. In many cases it’s just downright wrong. Universally, it’s not very joined up. Where investments are made to join data up, it’s often in expensive complex business transformations to build data warehouses or customer data platforms.
Forcing structure on messy unstructured data means essentially classifying things into categories.”
In his presentation he proposed an alternative approach, namely finding something which is good enough, rather than investing disproportionate time and effort on structuring and classifying data.
While the ideal solution would be to use perfect ‘golden source’ data for analysis, is there a quick and dirty approach to understand trends and clusters within the data – can we shortcut the spend and get most of the benefit?
In the discussion Mark shared the concept of network graphs. “Network graphs are dense meshes of interlink nodes and edges. A node is a thing – it could be a person, a piece of content, an answer in a questionnaire, a company… An edge is the connection between two nodes – such as a person who engaged with content, is part of a company, attends a show, expresses an interest in meeting with a company… By linking things together in a freeform way, you end up with a dense interlinked matrix of information. Depending on the types of nodes and edges, and usage of the graph, these spiderwebs of nodes and edges are called network graphs, social graphs or knowledge graphs. For an organiser it’s best to think of these as ‘interaction graphs’, given organisers mainly know people and learn what these people interact with through their engagement with them.”
He went on to explain that network graphs are powerful because they have some unique properties. This is true especially when people are nodes on a graph, because they build connections in non-random ways. According to Mark, it’s possible to:
Solve an interaction graph mathematically to work out the most important and least important nodes. This allows to identify the most influential nodes on the graph, and to work out what matters most to nodes.
Use an interaction graph to understand the linkages between different nodes – finding similar things – this is one of the reasons that network graphs (or autoencoded versions of these) are commonly used for recommendation systems, like LinkedIn’s algorithms.
Cluster an interaction graph by removing the least relevant links iteratively. Given sufficient data, you can let data tell you what the clusters are within your communities, rather than have to manually classify data.
The reason Mark likes graphs and believes they may be of significant value to organisers is because they are very resilient to missing data. He explained that each node is evaluated in relation to its linkages to other nodes – if a connection is missing, or wrong, it only has a small effect on the overall evaluation of the graph.
“In simple terms, graphs are good at dealing with messy imperfect data. While outcomes will always be dependent on data, in a master thesis a few years ago I looked at the information loss attributable to using graph derived metrics. One of our findings was that the benefit of being able to deal with incomplete data was greater than the information loss from not working with perfect data. Organisers are now drowning in unstructured data, which – with the best will in the world – will never be transformed into something structured and usable. However, this messy incomplete interlinked big data is exactly the sort of data which graphs work well with.”
I’ve got some clusters, now what?
It’s an old ad agency adage that if you’re not paying for something, you are the product. While one can argue that people pay in different ways (e.g., through their attention or time), there is no viable business model without a paying customer somewhere in the equation. Mark stated: “as organisers, we either make money from selling something within a cluster or selling something between clusters. Essentially, we might call our clusters exhibitors and attendees, or delegates and sponsors, but we all know that events are complex melting pots of interlinked clusters of companies and people, with slightly different goals and objectives.”
Then he expanded on this concept: “The framework of understanding whether someone is paying for a product which helps them in their cluster or helps them access another cluster is helpful in defining the business models which organisers use:
An exhibition is primarily a between cluster linkage – one cluster of companies (exhibitors or sellers) pays to solve a problem they have in accessing another cluster of companies (buyers).
A conference is primarily an in-cluster linkage – a member of a cluster pays to attend a conference because it solves a knowledge problem they have.
A research or data business is primarily an in-cluster linkage because a member of a cluster pays for information which helps them solve a problem in their own business.
One-to-one models are a between cluster linkage – one cluster is valuable enough to another cluster that, as well as paying to access the cluster, they are willing to compensate the other cluster for their time (e.g., Hosted Buyer)
Content businesses are typically between cluster linkages – they provide relevant information that readers of a cluster are interested in, paid for by advertisers who include their message either in or around the content.”
The post pandemic landscape has led to organisers exploring new business models, creating new digital offers, and accelerating the roll out of one-to-one models. Mark added: “Despite all the recent change, the reality is that business model innovation comes back to two types: we either do something for a cluster and they pay us directly for a product or service, or we deliver this cluster to another cluster and they pay for this access.”
What to build
Every organiser starts with a different map of their landscape. They know certain people, companies, have different offers and propositions and different resources and capabilities to be able to execute a strategy. Mark explained: “It’s critical to understand the current landscape in detail, specifically moving from simple groups such as data on known attendees and exhibitors to a cluster-based approach. By understanding which clusters are served through your current business model you can identify where you might be able to build next.”
He then shared an example where an exhibition organiser might seek to introduce a meetings model alongside their existing exhibition. This is a good example of taking two groups of customers who attend the show and offering them a further between cluster experience. Mark explained: “typically we think of building one-to-one models between our exhibitors (the sellers) and a subset of our attendees (the buyers), however a cluster approach to understanding our customers opens up more possibilities. If you only offer a meetings model to exhibitors, you instantly reduce the addressable market size, whereas if you offer this to companies within the same cluster you expand the opportunity set of potential buyers. A cluster approach to thinking about our customers may also create new opportunities; for example, it may be that two clusters within the audience of a show have a specific need to meet, and they use the show to do this informally. Creating a meetings model addressing this need directly is a way to monetise the linkage.”
He then returned to the concept that the data an organiser holds about their customers is their map of where they might seek to grow. “It may be that this map isn’t extremely high resolution at the present time; we just don’t have much data on our customers. We don’t know who they are beyond an e-mail, we don’t know what they interact with or what their interests are, we can’t predict who they might want to connect with. In this case it may be useful to explore how to strengthen your knowledge of specific clusters, to improve the quality of the map around these areas. An example of this would be investing in content which helps understand intent, or enriching data using third party sources.
investing in content which helps understand intent, or enriching data using third party sources. When doing so it is important to remember that the goal is not to build gold standard data through extensive surveillance and tracking. Instead, we should aim to build messy, incomplete but interconnected data to deepen our understanding of the interaction graph.”
The interaction graph concept is helpful in thinking about this map expansion. Mark added: “Every time we attract a new customer, we add a node to our interaction graph. The value comes from understanding what other nodes are connected to it. This helps work out where the new customer is on the wider map, which clusters they belong to, and how much value they may have given your existing (and future) business model.”
How to build
While Mark briefly touched on the options for growth in the session using a buy vs. build vs. partner framework, the focus was on how to build this capability in house. These trade-offs are explored in more depth in an article he wrote with Trevor Foley for Exhibition News
It’s easy to think “All this is very nice, but I don’t have the data scientists to do it”. Mark firmly disagreed with this perspective. He said “I’m fortunate to know a few data scientists with brains the size of planets – the last thing you want to do is ask them to work in an events business. They’ll be difficult to recruit, get bored quickly and be hard to retain and develop. There is no real career path, and building a data-led organiser must be something at the core of the business model… This kind of talent doesn’t want to be hired into a support function but seeks something where there is a guiding vision of the type of company you want to be in five years’ time.”
However, he sees solutions as well: “I think organisers do have some employees with significant potential in terms of solving the ‘we need to hire data scientists’ point. It’s important to identify those employees within the organisation who are highly analytical – they’re typically marketers or finance analysts. They’re the sort of people who like building dashboards, visualising data or are doing complex analysis in excels. If you pair these people with a developer with a background in ETL and a curiosity to explore what can be done on cloud platforms, you have a potent combination – with a bit of training and guidance.”
In terms of building a business, many may think that they need a system to do this on a continuous basis, with many integrations. Mark again disagreed: “That would be gold standard and probably gold-plated. I don’t think you need anything nearly as sophisticated. Events are held infrequently, you don’t need a real time system for something you hold each year, you’ve got more than enough time to work out what to do and months to execute, tweak and calibrate.”
Thanks again to all participants for engaging in the discussion and to Mark for sharing his perspectives.