Market research’s evolving AI needs and the teams needed to support them

Click here to access the slides.

Transcript of recording with Kyle Findlary – generated automatically by HappyScribe which means it will be about 80% accurate – if you spot confusing errors, please email ray@new-mr.com. The timestamps are included to help you jump directly to a point of interest.

[00:00:08.170] – Kyle Findlary

Right. So this talk is about how our industry has evolved over recent years, I’ve been in the industry for 15 years, which is longer than some and not as long as others. But in that time, we’ve seen the way in which we deliver insight has evolved and changed quite dramatically. And what I’m going to be speaking about today is the implications for those that evolution, how it’s affected, the way we constitute our business, the teams, the skill sets that we we have our teams and things like that.

[00:00:41.280] – Kyle Findlary

First of all, to say that I’m actually presenting many people’s work here, this is the reality of market research more than ever before, is that when you build things, you need an ever increasing number of skills. And there’s nothing can be built in isolation if it was. I’m also going to try and avoid using the term A.I., because it’s obviously a buzz word that is overused and and can become quite meaningless. So let’s see if I can avoid using A.I..

[00:01:12.220] – Kyle Findlary

First thing I want to impress upon you, though, is that all companies really are tech companies or 99 per cent of companies. There are very few companies these days that can do what they need to do without leveraging technology. And often case, it behooves those companies to take control of their technology and bend it to their world. And to the extent that companies do that, they are tech companies. Nothing can be done without technology these days. And that applies to market research as much as anything else.

[00:01:41.740] – Kyle Findlary

So why now, what has brought about this shift where we’re leveraging technology, machine learning, all these kinds of software development techniques, that kind of stuff, why now? What’s changed? The first thing, a change that clients are asking different questions. They have access to different data sources. They have the enhancing of capabilities, and they have different client internal capabilities that are changing the conversations that we’re having. And all that’s being underpinned by new technology, cloud infrastructure or machine learning, I use the term they forgot about that.

[00:02:22.440] – Kyle Findlary

And all these pieces that we’ve been maneuvering into place slowly over the years have suddenly been forced to we’ve been forced to accelerate this metamorphosis due to covid-19, where traditional data collection techniques were no longer available to us, people who had to be reached remotely on Web applications, mobile, or the kind of places where technology really can be brought to bear. What we’re really seeing is a shift in the way that we deliver insights from reactive to proactive insights from looking backwards and what has gone before in the last few months to trying to anticipate what’s coming ahead as we are buffeted by an ever accelerating cycle of marketplace’s economics and society.

[00:03:12.320] – Kyle Findlary

We’re trying to more than ever try and be proactive in anticipating what comes ahead. And this has been facilitated through technology. For example, we’re trying to surface emerging trends and detect anomalies. We’re trying to alert researchers to those changes that there can be more tactical within the scope of their strategies as they are buffeted by these changes on a on an increasing basis. And essentially, we’re just trying to predict ahead for planning rather than look backwards. What we’re trying to do, as well as we’re trying to empower researchers, we’re trying to make access to data more democratized so that you have the right data at your fingertips when you need it in a more of a data agnostic form, whether it’s unstructured data, unstructured data, having the right data streams at your fingertips.

[00:04:03.760] – Kyle Findlary

And the reason why you’re doing that is we’re trying to increase situational awareness like a soldier on a battlefield who where wars are won, when your soldiers have more information, more contextual information about the environment than the competitors in the same way. What we’re trying to do through technology is trying to increase situational awareness of the people making decisions. And from an inside provider perspective is we’re trying to be more consultative. We’re trying to be more domain specific and we’re trying to have the right insights at the right time for the context of our clients.

[00:04:33.850] – Kyle Findlary

And all this comes through technology, empowerment. So let’s have a look at the way in which things have changed over the years. Let’s first look at how we used to build products in the old days. What we did was we saw what the need was. We had some ideas. We consulted the theory, what was the frameworks, what was the literature review saying? We worked with marketing, science and TV and we validate models and they’d often be serving based models and then we’d socialize those within the market.

[00:05:00.770] – Kyle Findlary

The new world has far more steps in and those steps come to us from the world of software development. But they’re proving to be steps that each one is developing. A complex data product needs to follow or can can benefit from following the story, eliciting ideas, getting ideation, doing R&D, but then using agile processes, you know, building entities, AlphaBeta, getting our products into the test environments, the environment, production environments, measuring, evolving and iterating and all that kind of thing.

[00:05:27.980] – Kyle Findlary

And all that stuff comes to us from all the software development. And then the reason why we’re all tech companies these days, given the complexity of what we’re building, there seems to be one of the best approaches to doing that kind of thing. The data sources available to us have also has also the list of data sources has expanded a lot. We used to deal with surveys, diaries, online communities, panel data and media data and advanced analytics, and we still relying on all those things, but now we’re further augmenting them with a greater variety of data sources, social media data, search engine data, customer reviews, chat board conversations, mobile diaries, mobility data and all of this.

[00:06:05.140] – Kyle Findlary

All is new data. There is new data streams coming online and we need to be able to deal with it. And while the new data streams coming online, what we’re really dealing with is a few different kinds of data modalities. Whereas before we dealt with survey panel, data, media data, candidate, etc. again, we’re dealing with all those kinds of data, but we’re also dealing with unstructured text from a variety of sources like social media, more time series, telemetry data, time in form of Time series, different kinds of things as time series, image data, video data, chatbot logs, transcripts that have been auto transcribed by machine learning approaches.

[00:06:40.990] – Kyle Findlary

All those kinds of things are necessitating that necessitate us to process them in different ways and we need to have those abilities on hand. And bring it all together, the way we build products, the different kinds of data sources and data modalities that we deal with means that we have to deal with different skill sets. While traditionally within market research and insights, we had researchers with data processing, marketing, science, et cetera, et cetera. Now we’ve got all of those again.

[00:07:08.630] – Kyle Findlary

We’ve got business analysts, project managers, grandmasters, Davis database administrators, data scientist, data engineers, Mellops business intelligence dashboard professionals, front end developers, experts, back backend developers, cloud architects, etc., etc.. It’s an entirely different list of skills, and it’s a far more extensive list of skills because we’re dealing with far more complex data products. OK, so that’s just a bit of a juxtaposition showing the way in which the world has changed.

[00:07:43.240] – Kyle Findlary

Let’s talk a bit about the kinds of things that need to be taken into account when you’re actually building data products and there’s two large sides to building a data product that need to be taken into account to ensure that your data product has a defensible USPI. A holistic data product consists of two halves, embedding what we know already into the product, the domain expertise of our people, of our clients, the context of their categories and markets, and also the market research paradigm, our knowledge about how brands work and how that differs from the way in which consumers speak.

[00:08:20.250] – Kyle Findlary

We speak of brands in very unique ways that are specific to our industry, and embodying that paradigm within our machine models and our data products is very important. And then there’s the other side of the equation, which is how do we find the things that we don’t know, we don’t know. And that’s things like emerging trends, evolving, consumer language, evolving, consumer needs, etc., etc. And this is about the different techniques that we have at our disposal.

[00:08:43.920] – Kyle Findlary

Let’s look at how do we embed what we know really, what are the things that we need to be external from our domain into our machines in order to create meaningful contextual insights? For example, we need to take account of our training data sets of fine tuning data sets, things like document level coded data sets, and we need to put all those things into one place will make the Middle East easily accessible, because the more machine, the more data machines have, the better they do at what they do.

[00:09:17.720] – Kyle Findlary

And that’s the kind of things we’re talking about here, or attribute lists, topics, tag rules, grouping rules, the kinds of implicit domain knowledge that we have as experts that we need to externalize into our machines. For example, taxonomies like these kinds where you have hierarchical grouping rules or ways in which make sense of otherwise messy data. The other side of the equation is how do we get the stuff that we don’t know, how do we surface and make things make sense of the things that we don’t know, that we don’t know?

[00:09:49.380] – Kyle Findlary

And this is really where machine learning comes in and data science and analytics. And this is where we have to have different capabilities for dealing with different data modalities. For example, when it comes to natural language processing topic mortlock sentiment analysis, chatbot conversations, how do you know the intent of someone in that conversation? How do you generate a response automatically to their responses, etc., etc.? All these kinds of things are capabilities that we’re not really leveraged a few years ago, but are absolutely vital to this modern era.

[00:10:22.740] – Kyle Findlary

Identities like image and video, how do you ensure that you’ve got the capabilities for detecting the right objects, tracking client logos or brand logos through a video or piece of creative, how do you deal with identifying high order concepts like emotions and videos or things like that while those capabilities need to be brought to bear? Voice the audio making sense of transcriptions so that you can use those transcripts, an LP, natural language processing Hadean for the mood of a particular piece of creative, perhaps things like that.

[00:10:54.130] – Kyle Findlary

And then very importantly, even you’ve got these capabilities. You also need to have your architecture and your plumbing in place. How do you ensure that you are bringing your different data sets? Data assets do better on these capabilities, making sure that they are continuously learning and putting out the right contextual information. And how do you ensure that the kinds of things that your your your your human assets, the people that are working in your data sets, whenever they do something, they code new data or encode new information, how do you ensure that that’s always being fed back into your systems and so that your systems are continually, continually learning?

[00:11:37.510] – Kyle Findlary

How do you use your different data sources to get through data fusion? What that kind of stuff is absolutely fundamental to getting? Absolutely. It’s absolutely fundamental that you get that stuff right and you get your plumbing rat. And if you are able to capitalize, able to get both these things together, you’ve got a holistic brand by brand, a brand, a data product, but doing so means that you need to build those capabilities. And I like to think of the analogy of Lego bricks, getting those capabilities in place requires you to have the right technological resources, the right cloud architecture and processing capabilities you need to bring to bear the right data assets that contextualize your data that fine, tune it for your context, that help your machines invade the market research paradigm.

[00:12:28.240] – Kyle Findlary

And you need those methods are going to help you find the patterns, the right kinds of patterns contextualized in the way that you get assets have helped you contextualize them, things like MLP, those different data modalities. For example, we’ve got a product, and if I were to break up the way in which this product fits into this kind of analogy, our technology resources, we have cross organization brand connectors where we’re pulling in data from different parts of the organization that previously didn’t speak to each other.

[00:12:58.720] – Kyle Findlary

And we’re using data fusion to to harmonize that. We’re putting all those kinds of things, all that data and all the kind of downstream data that comes from that into a centralized place for logging purposes, for meta analysis, for understanding user experiences, for cross-selling and all those kinds of things. And then in terms of actual data assets, we’re listening to get the media spin data sales data and in terms of methodologies where we’ve got ways of identifying anomaly detection to trigger smart alerts, Times series decomposition and prediction and simulation and all that together becomes a a holistic data product that is kind of improbabilities.

[00:13:44.550] – Kyle Findlary

Continuing the kind of leg analogy may be kind of like restorationist adults’ once you have all these capabilities, you need to combine them into data products and into solutions. A capability might be leveraging a bunch of methods, some data storage, some data assets and technology resources. Maybe you’ve even wrapped it into a service API and you’re finding it out to different parts of the of your data ecosystem. And when you get these different capabilities together, you might define a product out of that which does a specific thing, answers a specific need or a specific question.

[00:14:16.890] – Kyle Findlary

And when you get products together, you get a solution, a suite of products that that all play in a similar kind of area. And it’s this kind of laddering approach using Lego bricks that is being underscored and is being empowered through all these new skill sets that all these new ways of working. And as you can see, it’s it’s a far more complex reality than what we used to deal with when it came to more traditional market research even five years ago.

[00:14:48.540] – Kyle Findlary

Once we put all this together, what we’re really trying to do is to try to create a closed loop, deadly ecosystem. What I mean by that, well, in many cases, in many of the things that we’re building in the data products, what we do is we and just to say that even even within our own organization, to use the famous, quote, the future. Yeah, it’s just not evenly spread. This is something that we’ve got right in some places, not right in other places.

[00:15:15.120] – Kyle Findlary

But the general idea is that when you have raw data, whatever the data modality is, you are first trying to enrich that data. And the way that you enrich that data is you bring your methods to bear, whether it’s in OP or object detection or whatever the case may be. And you’re further enriching your data with your data assets. You’re imposing structure and often unstructured data through taxonomies. You’re creating better classifications of your data through training data sets that improve your machine, learning models, for example.

[00:15:45.180] – Kyle Findlary

And then you’re creating a feedback loop with your human assets, your human in the loop coders who further bootstrap the quality of that data to a level that is acceptable to research grade level, where we can actually start to derive insights from it with some level of confidence. And once you reach that level, we start farming it out, whether it’s to online platforms and dashboards or reporting whatever the case may be, but importantly, what we always need to do is we need to close that loop.

[00:16:16.680] – Kyle Findlary

So all the work that the human assets have done to get the data up to a certain level needs to be fed back into our data assets so that our machines can continue to learn from them and continue to improve over time. And that’s the general kind of framework that we try to put in place when we deal with these data products. And again, you can see all those new skill skill sets scattered throughout this process with its cloud architects, data scientists, engineers, dashboard, whatever the case may be, project managers that are shepherding this all the way through the process, they’re all scattered throughout this process.

[00:16:53.640] – Kyle Findlary

So. What if I just what’s the kind of ground that I’ve discovered now? What have I what I hope to impart upon you as we end this presentation? Well, I hope you agree that all companies are tech companies. Obviously, some caveats that, you know, companies of a certain size. But generally speaking, we’re all dealing in the world of technology. We’re all enabled through technology. What we do is done through technology. And as I said in the beginning, it therefore behooves us to take some control and have some understanding of what’s going on under the surface.

[00:17:29.730] – Kyle Findlary

The weather we’ve both data products and the skills required have changed dramatically, the list of skills needed has increased a lot. The complexity of what we’re building has increased a lot. The ways in which we build it has become more complicated and more complex. And so we’re dealing with a very different world today. What we’re trying to do in the days we’re trying to create these kind of closed loop data ecosystems where we leverage what we already know, we external as the domain knowledge, we embed the market research paradigm into our machines.

[00:18:03.150] – Kyle Findlary

And we also need to have ways of surfacing the things that we don’t already know the best in class methodologies, the kind of machine learning tools that we need to make sense of the increasingly messy, unstructured data that we have at our fingertips these days. And the analogy that I like to use for this is we build capabilities that can be maneuvered into place like shared Lego bricks and build up block by block into increasingly complex data products. So I hope you find that interesting and hopefully find it useful.

[00:18:39.680] – Kyle Findlary

A look at my take on the way in which the industry has changed in the last few years. I’d be interested to hear how that compares to your own experiences of this. And myself and Ray will be ready to take some questions now. Thank you for your time.