AI & Matchmaking; A Match Made in Heaven?

I do like a good conference, or exhibition, or trade show, or whatever you want to call the whole affair. They're a maelstrom of barely ordered chaos; bustling and humming with conversation, collaboration and commerce.

Some of the biggest I've been to have exhibitor stands a-go-go, numbering in their thousands across square miles of show floor. It can easily become a daunting affair trying to have meaningful conversations in such an environment; how the hell do you pick the right people to talk to? The right companies?

One way to approach this problem is to wander the floor looking for names you know, yet you run the risk of missing out potentially transformative opportunities just because you'd not heard of them. It also defeats the object of your attendance; why go if you're only going to look at things you already know about?

Another tactic is to see which stands have the biggest crowds around them; herding. This is a natural instinct for the biologically bound, and we are drawn to crowds in unfamiliar settings. After all, if all these people are interested, must be worth a look, right?

Well, not necessarily. Just because there's a big crowd around the stand for the free "swag and smoothies" being doled out in dozens, this doesn't mean you're going find that killer new cloud based CRM platform / diamond wholesaler / couture fashion distributor for micro pigs; I hear mud baths are all the rage in Milan this summer.

Logically, you'll want to plan ahead; maybe research the exhibitors that are going and make some decisions about which you'd like to see. Great, excellent idea! How are you going to do that?

Most major events have search capabilities that let you quickly skim through what's available on their websites, so let's start there. The fundamental tech beneath this in my experience is a simple static keyword or tag search; the function looks for tag associated with the exhibitor or phrases/terms taken from their description. So, you better know the exact words used, tags assigned and have perfect spelling or the ambitions of your porcine fashion empire will get lost in the swill.

No truffles for you. Sad face.

It's at this stage the event organiser steps in with the silver bullet to solve all those problems; matchmaking.

Making Matches for Catches in Batches…

For those firmly embedded in the event's sector, matchmaking will be a concept you're intimately aware of; but for the benefit of those who aren't I want to give a quick primer. At its core though, the process is exactly what it sounds like; making matches between entities.

The basic premise is that you, as the attendee, want to have meaningful conversations with exhibitors at this event. However, the choice of exhibitors is dizzying for large expo's making the vetting process challenging. Event organisers act as the middlemen to address this challenge, typically by first understanding the goals you have (via a phone call, questionnaire, profile info etc.) and then match this against the exhibitor list for the event.

The feedback is a list of potential meetings that could aid you in your quest, vetted by a team of match makers using a mixture of manual and semi-automated means. This can range for keyword searches across profiles to simply reading your business cases and making experiential judgements as to which exhibitor’s best suite that need.

One key point to take away from this is that no two companies do this the same; I've worked with several of the biggest global events firms in recent years and can testify to the vast differences not only between companies, but between shows run by that company.

There is no standardised method for doing this, and the level of manual intervention / attendee effort required varies wildly. At its core though, the logic is simple; understand your situational truth via whatever medium used, and match this to exhibitors with compatible truths ranked by the match makers confidence in the success of the match.

This is where I think things get interesting; where Natural Language Processing & Artificial Intelligence can make a real difference.

Semantics, Shemantics…

Language is a tricky beast; you only need to consider that Q, Cue, Queue & Kew are all the same phonetically whilst having a plethora of different meanings that you really don't want to mess up. I've never tried playing pool using a royal botanical garden, but I imagine it has many shortcomings.

It's in this space that Natural Language Processing (NLP) engines really fly, the basic science of which I covered a few weeks back in my article about Deep Learning & Conversing.

An important facet of human conversation, and the way we think, is linearity. We write like we think, following a path with branching options that we select as we go, choosing our words to match the meanings we form inside our skulls. Often the process of writing evokes thinking as well, and ideas and/or concepts we hadn't thought to consider emerge from the process.

Ask someone to describe what they want from a watch in 5 key words, you may get something like this;

Gold, Swiss, Upmarket, Gift, Treasured

Ask to describe what they look for in a watch in a paragraph, you get this;

I want a gift for my wife, it has to be delicate but not fragile, in rose gold. Self-winding preferable but not essential, must stand the test of time.

Now in those examples, I spent more time writing the first example than I did the second; this is my mind at work, forming ideas in a flow rather than singular, descriptive terms. Conversational prose gives a much richer context, a meaning that is deeper than simple keyword tags.

By using an NLP engine (which in itself is a collection of models and methods, such as Recurrent Neural Networks, Multi Layered Perceptrons, Markov Chains etc.), we can pull the semantics out of that sentence and turn those into terms for comparison; tags.

On the other side of the fence we have equivalent information from Exhibitors about their services and product ranges; a comparative data set. We can easily enhance these generated comparison tags using data from the companies own website, articles in the news, whitepapers etc. By adding more data, you have a stronger association with certain derived tags, which can then be given a greater weighting in the match process.

All of this is 100% possible manually, but it is impractical. Manually reading through all this data and collecting a truly representative selection of tags, accounting for spelling errors, semantic similarities and the wider context of the user’s goals and ambitions for the event is an enormous task for even a larger team of people to achieve.

Even with the best processes in place consistency goes out the window, with opinion colouring what one-person construes is a match as opposed to another member of the team; consistency is key, without rigour there is no confidence.

This is where AI really, really helps. It can consistently analyse all delegates against all exhibitors with unerring rigour or bias, and draw a ranked match order within the context of the event being run.

Of course, if the model designer is bias then the model output will be, but that philosophical headache is for us to debate another day…

State of the Nation

There are some products already alive in the market making use of these technologies. From first-hand experience I can see that, whilst moving in the right direction, they have some way to go. The matches are clunky, more tuned towards keyword searches and common skills listed on social profiles, versus a true Language Understanding engine that pulls out the semantics and context from swathes of free text, giving a much stronger dataset to go off when making matches.

This restrictive approach, whilst giving tightly defined parameters to feed nicely into a model, does not necessarily tell your truth; your context. It's easier to build, but does not provide the customer, the delegate, with an engaging experience; customer centric design is a must in the modern marketplace.

There's no scope to enhance this match either with other artifices, such as articles, whitepapers, missives, maybe even requirement specs; all of which could give a much deeper context to your situation and that which you seek. More data is better, if it's relevant, and with each piece analysed comes more context that supports, enhances and interprets your needs with ever tighter granularity.

In that marketplace of existing tools, there also seems to be a bias towards delegate matchmaking; making contacts at an event based on common likes, roles and industry (C2C).

It's an important distinction to make, as this is a completely different challenge to the B2B / B2C matchmaking that forms the lifeblood of the event's industry. After all, the value proposition Organisers provide to Exhibitors at an event is a stream of well-matched Delegates that have potential to become Customers. Without that, there's no point even setting up store.

AI has the potential to utterly transform the matchmaking process, by making it quick and easy to understand what Delegates want, and what Exhibitors provide. Through Natural Language Processing comes the ability to understand vast quantities of text and content that could quite easily have been ignored, engaging Attendee's in the matchmaking process during the run up to the event and incentivising engagement through the promise of ever more accurate matches both now, and in the future.

In short, you get to know your each and every customer through their own words, and they in turn trust your judgement in providing reliable, meaningful matches; all on a scale hitherto impossible without an enormous team of match makers until recently.

Artificial Intelligence is intelligent; it learns, and in turn we can learn from it if given the chance.