Skip to main content
Search form
  • 00:00

    [MUSIC PLAYING][Using Conversational Methodology for MarketResearch--Streetbees][MUSIC PLAYING]

  • 00:26

    OLIVER MAY: I think the vision for Streetbeesis to be the largest database to help companies understandhuman behavior.[Oliver May, Co Founder, Streetbees]That's really the essence of what we do.And the way in which we do that is wecollect what we call real life moments from people's lives

  • 00:46

    OLIVER MAY [continued]: by asking them to capture these moments on mobileat very intimate moments in their lives.And we will ask them about, what is their emotional state, whoare they with at the time, what is the broader contextsurrounding their situation.And it's almost like you're capturing this metadata set

  • 01:07

    OLIVER MAY [continued]: around that particular moment.And we're then able to encode that information in clever waysand actually distill that down to insight for our clients.But the long term vision is reallyto create this repository or this huge databaseof human behavior and to be able to tap into that at will

  • 01:30

    OLIVER MAY [continued]: on behalf of a variety of different clientsto answer their inside questions.

  • 01:35

    NINA KANIN: At Streetbees, we think of ourselvesas kind of the nexus between qual and quant.[Nina Kanin, Client Strategy Director, Streetbees]So probably unsurprisingly, a very hybrid approachis really useful.So I think you have to be really comfortable with numbers,and you have to not necessarily be able to do--I mean, our clustering is machine learning-based.So you have to be familiar with the philosophiesor the theories of how those work,

  • 01:57

    NINA KANIN [continued]: but you have to be comfortable reporting with those numbers.You have to be really comfortableat statistical analysis.I think really importantly though youhave to be good at tying the depth of the qual that we getand those individual consumer submissions to those numbersto make a story.So that's, I think, where the real storytelling comes in.That's the fun part.

  • 02:18

    NINA KANIN [continued]: And I think sometimes that putting numbersinto a human form is not always intuitive or easy.So I think that's the magic part of the job.

  • 02:31

    OLIVER MAY: So the conversational researchapproach that we have here at Streetbeesis fundamentally different to any other market researchfirm's methodology.So if you think traditionally about the way in which marketresearch has worked, you've alwayshad to choose between the richness of small scalequalitative research, which would typically be something

  • 02:53

    OLIVER MAY [continued]: like a focus group, where you would bring together10 individuals and ask them a whole bunchof detailed questions.Or at the other end of the scale,you could get mass scale research--so quantitative research--but you would be stuck with thingslike multiple choice questions, radiobuttons, that sort of thing.

  • 03:14

    OLIVER MAY [continued]: That's great from a scale perspectivebut it has genuine challenges methodologicallybecause as a researcher, you're forced to actually guesswhat the answer to your question is when you'rewriting the question before you've releasedthat survey to the market.So let me give you an example.Imagine that you are an insights manager for ice cream,

  • 03:36

    OLIVER MAY [continued]: and you're looking at launching a new flavor in China.And I ask you, what is the flavorsthat you would expect Chinese consumers to want next?And you actually have to compose a multiple choice questionasking that question to your Chinese consumer.When you're doing that, you will be forced

  • 03:56

    OLIVER MAY [continued]: to say, OK, well, let me guess.Is it chocolate?Is it strawberry?Is it vanilla?You're probably sitting here in Londonwith a very Western mindset, and it will be very difficultfor you to guess what the likely response is going to befrom your Chinese consumer.But that's the approach that you'reforced to take if you are only giventhe option of multiple choice questions.

  • 04:18

    OLIVER MAY [continued]: What we do is we have an open text or conversationalmethodology where we say to you as a researcher,you can ask an open ended question.So you could say, what are the flavorsthat you would like to see, or what are some of the flavorsthat you're seeing that are trendingin China at the moment?And we can then apply machine learning

  • 04:40

    OLIVER MAY [continued]: to all of those responses that still come throughat massive scale and, in doing so,actually reveal some new flavors that you wouldn'thave discovered otherwise.So that's a trivial example, but it justgives you a quick demonstration of the powerof the conversational research methodology.The other interesting thing that we foundis that conversational research from a user perspective

  • 05:03

    OLIVER MAY [continued]: is something that also feels a lot more natural.So the interface, if you're familiar with the Streetbeesinterface, it looks very much like WhatsApp or FacebookMessenger.And just psychologically, users connectwith those kind of apps, and they'rea lot more willing to share when youhave that type of interface.

  • 05:23

    SAM LOWE: Technology for us is kind of three different worlds.[Sam Lowe, Chief Technology Officer, Streetbees]There's the technology that we useto work with the people who are giving us the information.They're telling us their stories as we call it.We call them the Bees, hence the name Streetbees.So we have a certain amount of technologywe use to gather the information with them.We have other technology that we use to runthe operations of our business.

  • 05:44

    SAM LOWE [continued]: So that's, what are the tasks that we're doing,how many people do we need in orderto have a representative sample, and running those projects,collecting the data, processing the data.And then we have the technology that weuse to then allow our clients to gain access to the insightsthat we have.So it's a different set of technology

  • 06:06

    SAM LOWE [continued]: that is based around things like dashboards and datavisualization so that marketeers working in big, global consumerbrands can really make sense of what we found.Streetbees is a business that's been founded specificallyto turn traditional approaches to research on their headin that of course we use all the conventional approaches

  • 06:27

    SAM LOWE [continued]: to gathering research information and processing it,but the ability that we have with the technology thatnow exists is to do things a different way around as well.And that is to be gathering a lot moreorganic, natural data--open text and images and videos, thingsthat are quite difficult to cope with in very large volumes

  • 06:48

    SAM LOWE [continued]: if you're processing them in spreadsheets or databasesusing humans.But machines can now understand some of this data.So for an image, that will be all of the pixels thatmake up the image.All the red and the green and the bluein each of those pixels is transformed into a numberto show basically its brightness-- the amount of redin that pixel.

  • 07:09

    SAM LOWE [continued]: And that goes into your machine learning model alongwith the models that process the natural language.And these models that process the natural language have allcome about since the huge amount of data on the internethas allowed us to be actually able to train machinesto understand words based on whichwords are used around them.

  • 07:30

    SAM LOWE [continued]: And if you have enough data-- and there's an awful lotof data on the internet--then you can suddenly get a machineto understand the patterns in the words and what that means.And so it will transform a sentenceinto also a list of numbers that capturesthe essence of the context and the information there.And both of those can go into the same model.So you can then have a much better accuracy

  • 07:53

    SAM LOWE [continued]: if your machine learning model istaking in both the mathematical representation of the imageand of the text because it may wellbe that if somebody is recording for us what they're eating,they may not mention everything in what they sayto us that the photo tells us.And also, the photo may have information in it that the text

  • 08:16

    SAM LOWE [continued]: doesn't.So the combination of the two allowsus to get a much better view as to what'sin what they're experiencing, what they're recording.The final one of the three areas of technologythat we build products with is in data visualization.So that's almost, as you've takenthis huge amount of information in from our conversations

  • 08:38

    SAM LOWE [continued]: with the Bees and you process it down,but you're processing it down to what is effectivelymathematical structures.Mathematical structures are not reallygoing to be suitable to support a marketeer in makingdecisions.I mean, they need something that they can manipulate and workwith, and they can cross-cut, and theycan analyze in order to work out what should they do next?

  • 08:58

    SAM LOWE [continued]: Where are the opportunities for product improvementsor new products?So we spend an awful lot of time, then,on data visualization, about how to makethat information accessible and howto make it able to be manipulatedby marketeers without them having to have PhDsand without having an understanding of the mathsinvolved.

  • 09:19

    SAM LOWE [continued]: So that is a major area of kind of product creativity for us.How can we design user interfaces for marketeersto use that gives them the power to manipulatethese things but without losing the depth that is behind it?You know, our goal is to make thingsthat are as simple to use as Google Analytics

  • 09:40

    SAM LOWE [continued]: but can manipulate all the power of this machine learning.In fact, maybe a better example to Google Analyticsis maybe Google Maps because everybody can use that.You don't require any training to use Google Maps.You're actually processing a huge amount of informationwhen you use it, but it's entirely natural.And that kind of way of working with this very abstract,complex data that's come from this huge quantity

  • 10:04

    SAM LOWE [continued]: is the way that you can really use it to support decisions.

  • 10:07

    NINA KANIN: I saw one of our clients.I was actually sitting with them in the room,and we showed them the dashboard for the first time.And there was such joy in actually seeingwhat your consumers are using.And it's very intimate.You can see hands.Sometimes you can see children in the background or dogs.And so I think building as much of that in to the reportitself along aside the findings and the stats is crucial.

  • 10:28

    NINA KANIN [continued]: When it comes to describing the modeling that we do and the AI,I think, again, visually-- because for most people whodon't have a hard core research background,you say "cluster" or "log" and you've lost them.So I think it's all about painting a picture.

  • 10:41

    OLIVER MAY: So the advantages of applying a technology approach,and specifically machine learning--and within machine learning, something called naturallanguage processing--is it actually allows us to analyze large volumes of textat scale.So a basic example is that at the start of all

  • 11:01

    OLIVER MAY [continued]: of our surveys, we generally ask the question,how are you feeling today?And sometimes people respond with a single word.They might say, good.Or other people will respond with multiple paragraphs,telling us all about their day, how they're feeling,what's happened to them during their day.Now, you can imagine if we're getting 10,000 responses

  • 11:23

    OLIVER MAY [continued]: like that, that's a lot for a researcherto actually read through.So what we do is we actually pass all of that text datathrough machine learning algorithms.And the process that those algorithms then go throughis to what we call code up those responses.Now, what I mean by that is we actually have a set of emotions

  • 11:45

    OLIVER MAY [continued]: that we have sitting in our database.And the machine is intelligent enoughto be able to interpret the context from the responsesand actually tag those sentences or words with specific emotionsfrom the database.So that's a very powerful technique.And that is a way in which we actuallytranslate this large volume of rich qualitative data

  • 12:09

    OLIVER MAY [continued]: into something that can then be analyzedin a very quantitative way.Imagine, again, you've got those 10,000 responses.I can now say things like, 40% of those responseswere tagged with the emotion "happiness."So suddenly you're turning somethingthat's very rich but difficult to digest into somethingthat is quantifiable.

  • 12:30

    SAM LOWE: Capturing things in the momentallows you to not be at the mercy of recalled behaviorand claimed behavior.I think the movie industry worked outdecades ago that they were better off when they had a testscreening of a new movie to actually film the audience'sreactions to it than they were to just ask them what they

  • 12:51

    SAM LOWE [continued]: thought when they walked out of the screening, or at leastto do both, because when you walk out of the screeningyou don't remember some things.Some things that you actually did respond well to maybe youdon't believe that that's somethingthat you want to share.Maybe you choose to phrase yourself in an ambiguous way.So capturing things at the moment, as well as

  • 13:13

    SAM LOWE [continued]: the claimed behavior, gives you far more data to work with.You've got to have, then, the computational powerto deal with that huge increase in data,but it gives you far more data to deal with.And the other thing is if we can move away from pre-categorizinghow people can respond to you, thenwe can suddenly understand-- we can find out things that we

  • 13:35

    SAM LOWE [continued]: didn't know to look for.Now, if I want to go and find outwhat the opportunity for a new product is,I can choose to go and look for where there'sopportunities for 10 hypothetical productsthat I've made in my mind.But it's far better to go and gather the data naturally,and then find out what is the one that jumps out.It's far more cost effective, and it's

  • 13:55

    SAM LOWE [continued]: far more powerful than having to search for everythingthat you might need to look for.So that's that idea of using open text and natural languagesand images and videos, which we're allvery comfortable within in our daily livesnow thanks to the difference that mobile and social mediaand the internet has made.That gives us the ability to find

  • 14:15

    SAM LOWE [continued]: these things which we otherwise wouldn'thave been able to find.And now, thanks to machine learning,we can actually then explain them after the factrather than having to explain them before we start.

  • 14:25

    NINA KANIN: Generally, when we geta new kind of ad hoc business problem to tackle,that will come in and I will work with our growth teamto help frame that problem and kind of helpdesign the solution.But once we've decided that's a go,we take that to our data and operationsteam and our research team to help start

  • 14:45

    NINA KANIN [continued]: to kick that off and structure all of the back end solutionswe have to make that happen.So there is a lot of planning around management of our Beesto make sure that we're getting the right kinds of consumers.There's a lot of data quality.There's a lot of checking of the submissions that come in.And also actually building and customizing the dashboarditself.In addition, we tag one of our data scientists to come in,

  • 15:08

    NINA KANIN [continued]: and we work with them on actually findingthe relevant clusters and what are those demandoccasions that we're going to unearth in the study.And at the very end, sort of it becomes a dual responsibilityof the research team and the strategistto tag team that report and make surethat it's extremely actionable before we give itback to the client.

  • 15:30

    OLIVER MAY: So in terms of the type of requeststhat we generally get from our clients,they can be quite varied.We do a range of different projects.Typically, with a new client, we start offwith a small brief to give them a taste test of what Streetbeesis all about.With some of our larger clients, we'rereally working on much more important, strategic pieces

  • 15:51

    OLIVER MAY [continued]: of work for their businesses.So to give you some examples, we work with a lotof large FMCG clients.And we have a particular expertise in food and beverage.Some of the work that we're doingfor some of those large clients at the momentis really trying to define their new product developmentroadmap.So a big project that we're running at the moment

  • 16:13

    OLIVER MAY [continued]: is actually what we would call an innovation finder.And just to give you a sense of howthat might work in practice, what we dois we collect food and beverage momentsfrom thousands of consumers in different countriesaround the world.We capture that together with allof that emotional and contextual information

  • 16:33

    OLIVER MAY [continued]: that I described as metadata.And what we do is we then apply the machinelearning to identify clusters of occasionswhere the people in those occasions are buying,consuming, and behaving in similar ways.And that's really the key insight or the key kindof distillation that the clients are looking for

  • 16:55

    OLIVER MAY [continued]: to be able to identify these clusters of occasionswhere behavior is similar.What that leads to is an analysisof each one of those occasion spacesand a discussion with the client about what that means in termsof not only products that they could pitch and developfor those specific occasions, but also

  • 17:15

    OLIVER MAY [continued]: thinking about how we would actuallymarket to those particular groups of people.What sort of language is going to be used in order to reallyresonate with those people?And also, where do we find those people?Are they millennials who are typicallyspending a lot of time on Instagramand other social media apps?Are they older groups of people for which

  • 17:35

    OLIVER MAY [continued]: above the line marketing, TV advertising,all that kind of thing is going to be more effective?So it's a very specific approach wherewe work with you as a client to help you understandnot only what are the new products that you shoulddevelop, but also a comms strategy,and also defining where you should actuallybe placing those comms.The skills that we look for for people

  • 17:58

    OLIVER MAY [continued]: that we employ in this business very much depend on the rolethat you're applying for, obviously.But I think one thing that I find very interestingis that I think more and more the world is shifting awayfrom traditional forms of educationand shifting more towards really practical applications

  • 18:21

    OLIVER MAY [continued]: of your skills and really an abilityto demonstrate those in interviews.One of my roles, which I didn't touch on before,is actually the Head of People.So I run the talent and the HR function here at Streetbees.So I spend a lot of time sitting in interviews.And I do have some quite specific thingsthat I look for.

  • 18:42

    OLIVER MAY [continued]: Absolutely, one of the key thingswould be a demonstration of not only you'vehad a great education, but also haveyou thought about applying some of those skillsoutside of maybe traditional education or internshipprograms, those kind of things.For example, maybe you're coming to apply to Streetbees

  • 19:02

    OLIVER MAY [continued]: as an app developer.What sort of projects have you taken upon yourselfto actually develop outside of your course?Because for me, really showing me a fantastic portfoliois probably a lot more powerful than tellingme what grades you got in your university degree.I think over and above that, I think personality factors arereally, really important.

  • 19:23

    OLIVER MAY [continued]: Working in startups is hard.I know there's a big perception out there sometimesthat startups are very trendy and a fantastic place to work.The reality of startups is that you'regoing to have to work a lot harder in a startupthan you will in a traditional corporate environment.The quid pro quo with that is you'regoing to be given an awful lot more responsibility.

  • 19:46

    OLIVER MAY [continued]: There won't be that safety net, which I thinkis fantastic for people who are looking to really pushtheir career fast.The breadth of responsibility that you're going to getand the visibility that you'll have across the entire businessis a lot broader than you would getin what would be typically a more narrow rolein a corporate environment.And so there's a lot of plus factors there,

  • 20:07

    OLIVER MAY [continued]: but it is going to be tricky.It is going to be tough.And I think, in terms of personality factors,I would then look for tough situationsthat people have been through in their lifeand how they've really worked through those situationsas a demonstration of, can I performin this sort of environment?

  • 20:25

    NINA KANIN: Being able to think on the fly.And not just have answers, because most the time whenwe're having a first conversation with a client,they're not going to have the answers yet.That's why they want to do the research.But being able to ask questions on the go and kind ofthink with a client in the meeting.Be comfortable in those settings because weare at the size where we are always in front of clients

  • 20:47

    NINA KANIN [continued]: and we're very much involved in their business.Having a very hybrid mindset.Like I talked about before with qual and quant,it's really important because our data straddles both worlds.You have to feel really comfortable getting nittyand gritty with both.And I think having the "yes, and" attitudebecause we're still inventing this and we're figuring it out

  • 21:10

    NINA KANIN [continued]: together.

  • 21:11

    SAM LOWE: I would encourage everybody,even if they're not computer scientists, to certainlyget involved in machine learning and find out what's now becomepossible in machine learning.I believe it's the biggest--it's going to be the biggest generational changein technology in the next decade or two.It's going to have as big a change on us

  • 21:32

    SAM LOWE [continued]: all as the internet has had in the last two decadesand mobile has in the last decade.So you don't need to be a computer scienceor a mathematician to do that.You can get involved in it from the social side.There's an awful lot of work on the ethics of machine learning.There's an awful lot of work on the data

  • 21:53

    SAM LOWE [continued]: and the meaning of the machine learning and how it gets used.So I think it's such a fertile area that everybody shouldbe finding out about it, because Ithink it's going to affect all industries that people go into.[MUSIC PLAYING]


Streetbees Co-founder, Oliver May, Client Strategy Director, Nina Kanin, and Chief Technology Officer, Sam Lowe, discuss the conversational research methodology and machine learning Streetbees uses for large-scale market research.

Looks like you do not have access to this content.

Using Conversational Methodology for Market Research: Streetbees

Streetbees Co-founder, Oliver May, Client Strategy Director, Nina Kanin, and Chief Technology Officer, Sam Lowe, discuss the conversational research methodology and machine learning Streetbees uses for large-scale market research.

Copy and paste the following HTML into your website