《Insight Platforms:AI研究与分析洞察平台指南(英文版)(65页).pdf》由会员分享,可在线阅读,更多相关《Insight Platforms:AI研究与分析洞察平台指南(英文版)(65页).pdf(65页珍藏版)》请在三个皮匠报告上搜索。
1、The Guide to Arti?cialIntelligence forResearch&AnalyticsMike StevensThe Insight Platforms Guide to AI for Market Research&AnalyticsTable Of ContentsAbout this ebook3About Insight Platforms4About the Author5Part 1:What is Arti?cial Intelligence Anyway?Part 1:What is Arti?cial Intelligence Anyway?6 6L
2、ighting the Way7A(Very)Brief History of AI10Machine Learning11Natural Language Processing13Computer Vision14EmotionAI14Cheat Sheet15Part 2:23 Practical Uses for AI in Research&AnalyticsPart 2:23 Practical Uses for AI in Research&Analytics16161.Converting Speech to Text172.Automatic Translation203.An
3、alysing Text214.Social Media Listening245.Analysing Customer Experience Feedback266.Emotion Analytics287.Conversational Feedback(Chatbots)298.Big Qual339.Natural Language Analytics3410.Automatic Content Generation3611.Recognising Content in Images and Video3812.Recognising Faces411Insight PlatformsT
4、he Insight Platforms Guide to AI for Market Research&Analytics13.Virtual/Mixed/Augmented Reality4214.Analysing Vocal Tone4315.Analysing Eye Movement and Facial Expressions4416.Biometric Feedback4617.Data Visualisation4818.Price Optimisation4919.Attribution Analytics5020.Customer Journey Analytics512
5、1.Segmentation5222.Personalisation5423.Enhanced Prediction54Part 3:A Research&Analytics Blueprint for AI SuccessPart 3:A Research&Analytics Blueprint for AI Success55551.Improve your Business-as-Usual572.Build New Client-Agency Models593.Embrace Embedded Insights604.Always be Learning625.Do What AI
6、Cant63Final Thought642Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsAbout this ebookThis guide aims to demystify AI,show how it is used today,and suggest tangible steps forinsight teams to make the most of it.It is broad,and not particularly deep.It focuses on pract
7、ical applications,not theory.Part 1Part 1 is a blu?ers guide to some basics of AI.Remember when you were in your teens,and people started talking about new music youhadnt heard?You didnt feel right until you had listened to the track everyone was talkingabout.Reading this section will be like listen
8、ing to that track:youre not going to be an instantsuperfan,but youll be able to hold your own next time it comes up in conversation.Part 2Part 2 is the longest section.It covers 23 di?erent applications of AI in research&analytics.These are all things that arehappening today,some of which you will b
9、e familiar with.This section also includesexamples of technology platforms and links for you to?nd out more.Part 3Part 3 is a short roadmap for maximising the AI opportunity.If you work in research&analytics for a brand or an agency this section has 5 actionitems to help you prepare for the changes
10、AI will bring.3Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsAbout Insight PlatformsInsight Platforms is a learning resource for buyers in product research,consumer insight,social intelligence,customer experience and digital analytics roles.It is the only site dedic
11、ated to software and data solutions for research,with comprehensivedirectory and expert content help users learn about and choose the right platforms.In the Platforms Directory,users can search or browse to?nd software providers,datasources and managed services.They are able to?nd and compare more t
12、han 700 suppliersin 30 categories,from A/B testing to Visual Analytics.The Blog includes how-to guides,independent reviews and strategy articles written byagency,consultancy and enterprise experts.The Resources section includes exclusive content and learning materials for registered users:e-books,mi
13、ni-courses and webinars.4Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsAbout the AuthorMike Stevens is a leading advisor,writer and speaker on theintersection of technology and research.He has over 20 years international experience with research,software and consult
14、ing?rms including Vision Critical,wherehe led the EMEA region,and Kantar,where he managedregional business units and global accounts.His consultancy?rm,What Next Strategy&Planning,providesinsight expertise,transformation help and training tocorporate insight teams,agencies and software companies.He
15、is also the Founder and Editor .You can contact him by email,follow him on Twitter or connect with him on LinkedIn.5Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsPart 1:Part 1:What is Arti?cialIntelligence Anyway?6Insight PlatformsThe Insight Platforms Guide to AI f
16、or Market Research&AnalyticsLighting the WayIn the last 24 hours,how many times have you bene?ted from arti?cial light?How many lightswitches have you?icked?How much have you achieved that wouldnt have been possible without those glowing?laments,tubes and LEDs?No idea?Me neither.Its not something yo
17、u think about very often.Arti?cial light is just there.Its part of the fabric of our existence.Its so embedded in our livesthat it has become invisible.Were only truly aware of it when its absent:a bulb blows,the power goes down or were inthe countryside at midnight.But it wasnt always like this.Abu
18、ndant,cheap arti?cial light is a comparatively recent phenomenon,and its economicimpacts rarely get much attention.If you lived in mediaeval England,seeing after dark was the preserve of the extremelywealthy.One million lumen-hours of candlelight-thats roughly a year of lighting a singleroom for a f
19、ew hours each day-would have cost nearly 40,000(USD$50,000)in todaysmoney:Source:Seven Centuries of Energy Services:The Price and Use of Light in the United Kingdom(1300-2000),Fouquet&Pearson,Jan 20067Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsAs candle-making be
20、came more e?cient,this dropped to 15,000 by around 1450-where itstayed for the next 350 years.And then came the 19th century:gas lighting was introduced in the early 1800s and electriclighting came in after 1870.The unit cost of arti?cial light plummeted,and adoption grewexponentially.By the dawn of
21、 the 20th century,the light bulb was in factories and homesthroughout the country.Today,a million lumen-hours of a typical 10W LED bulb will cost between 1 and 2 inEngland.So what does all this have to do with arti?cial intelligence,market research and analytics?Stay with me.Look again at the chart
22、above.Around 1800,there is a critical in?ection point-after which,there is exponential growth inthe use of arti?cial light.Being able to lengthen the day triggered massive economic,technological and socialinnovation.Factories could run for longer and be more productive.Workers could educatethemselve
23、s by reading after their shifts.8Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsToday,we are at a similar in?ection point in the application of AI to consumer insight,userresearch and customer experience management.Over the next few years,the adoption of arti?cial in
24、telligence will drive massive growth involumes of feedback,research and analytics:Today,we are at the point where the lines cross.This is researchs light bulb moment.AI will transform the research&analytics ecosystem by:embedding researchembedding research capabilities in software tools throughout o
25、rganisations helping non-specialistshelping non-specialists interpret feedback to make better,user-centric decisions turbocharging researchersturbocharging researchers by giving them more data,enabling smarter analysis andspeeding up work creating new rolescreating new roles in research,technology a
26、nd customer strategy fundamentally changingfundamentally changing what we think of as research.Its an exciting time to be part of this industry.9Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsA(Very)Brief History of AIArti?cial Intelligence generates hype and fear in
27、 equal measure.Were either heading for atechno-utopian future or one in which humanity is enslaved by Skynet.More prosaically,many researchers fear their own redundancy within the next decade.But what is it,really?AI has actually been around since the late 1950s.GOFAI(Good Old Fashioned AI,or Symbol
28、icAI)comprised a series of pre-programmed production rules(if-then-else statements)thatled to some useful real world applications.Many Expert Systems use this form of rules-based AI for medical diagnoses or controllingmanufacturing processes.But these narrow AI tools require exhaustive programming a
29、nd are not really intelligent:they just apply a set of rules much faster than a human can.Then-starting in the mid-eighties-computers were designed that could adapt their own if-then logic based on inputs and feedback.They were able to learn.These machine learning approaches underpin a lot of todays
30、 AI software.Since the early 2000s-thanks to massive growth in computing power and volumes of data-AI has evolved to bring us deep learning:new forms of advanced machine learning that usearti?cial neural networks to more closely model how human brains learn.The evolution of Arti?cial Intelligence10I
31、nsight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsMachine LearningThis is about as technical as this guide will get.Dont be put o?.So what is machine learning?Its when an algorithm takes input data,?nds patterns,learns from them and then appliesthat learning to make a de
32、cision.Simple.Most machine learning is used for classi?cation,regression or clustering,and there are fourmain categories of algorithm:Supervised Learning Unsupervised Learning Semi-supervised Learning Reinforcement LearningStill here?Good.Lets look at some examples.Supervised LearningSupervised Lear
33、ning is the process of feeding large volumes of training data to a softwareprogram so that it can run its own classi?cation or regression models more accurately.Imagine you want to train a computer to recognise a tumour from a brain scan.You mightgive it some basic rules,show it thousands of existin
34、g scans and give it feedback each time itcorrectly or incorrectly?ags an image as cancerous.This would be a classi?cation-based machine learning model.Or imagine you want to predict the change in value of a given stock market equity.You wouldfeed the model lots of historical information about the co
35、mpanys own performance,datafrom competitors,consumer con?dence,external factors such as the weather.and train itto predict the historical value.Once it is su?ciently accurate at predicting the past,you might be con?dent enough in itspredictions about the future.This would be a regression-based super
36、vised learning model.If you hear people talking about Nearest Neighbour,Naive Bayes,Decision Trees,LinearRegression or Support Vector Machines(SVM)they are probably referring to supervisedmachine learning.11Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsUnsupervised
37、Learning Unsupervised Learning is where there is no human guiding the computer.The algorithm?nds patterns in data by itself.There is no outcome variable on which to try to modelrelationships.Common uses including detecting unseen patterns in data,summarize dataand describing it.Imagine taking custom
38、er-level data from CRM records(spend levels,contact centreenquiries,products purchased,locations,demographics),pouring it into the computer andgetting output that shows groups of customers with a high propensity to buy certain producttypes.That would be a clustering-based unsupervised learning model
39、;segmentation and basketanalysis are common marketing applications for this.K-means and hierarchical clustering are common algorithms.Semi-Supervised Learning Semi-Supervised Learning is a combination of the previous two approaches.Sometimes itcan be costly to have a human label data and supervise t
40、he machine;semi-supervisedapproaches help to limit that cost.Reinforcement Learning Reinforcement Learning is a way of training algorithms to respond to their environmentusing a system of rewards and punishment.These are set in advance,and there is no humaninvolvement in the learning process.Driverl
41、ess cars and some marketing optimisation toolsuse this approach.And thats most of the complicated stu?out of the way.12Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsNatural Language ProcessingNatural Language Processing NLP-powers hundreds of apps and services we no
42、w take forgranted:speech recognition,machine translation,search engine crawling and indexing,those did you mean suggestions for mis-typed Google entries.It connects linguistics with the di?erent strands of AI so that computers can do useful thingswith language:analyse it in spoken or written forms,r
43、espond to queries from users and evengenerate output in coherent sentences.Natural Language ProcessingThe branch of NLP known as Natural Language Understanding powers software for analysingthe content and meaning of text data by picking out keywords,identifying entities(such asbrand names)and interp
44、reting sentiment.But language is tricky stu?.Were rarely conscious of it,but it contains lots of sub-systemswith their own internal structures and rules:phonology(sound patterns)morphology(symbols,characters,words)syntax(structural features like nouns,verbs,sentences,grammar)semantics(the meaning co
45、nveyed).For computers,blocks of language-in books,conversations,social posts,search queries orfeedback surveys-are designated as unstructured data until they can be converted intostructured data that they can do something with.Deconstructing and reconstructing language using learnable rules(for phon
46、ology,morphology,syntax and semantics)is the heart of Natural Language Understanding.13Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsComputer VisionTeaching computers to see is critical for the development of robotics,autonomous vehiclesand dozens of other applicati
47、ons.But its hard:seeing,interpreting and responding to visual stimulus takes up more of thehuman brain than any other distinct process.Think about all the steps involved when someone throws a ball towards you:the image of the ball hits your retina,which sends the signal to your brain your visual cor
48、tex analyses the image and compares it to everything else it alreadyknows then it classi?es the image as a ball and tells your hand to catch it.In fractions of a second.Recreating these processes is a very tricky programming task.Certain shapes and colours inimages can still bamboozle AI.One normall
49、y reliable algorithm stubbornly classi?es anabstract swirl pattern as a toaster.Another can easily recognise birds and bicycles;but a birdriding a bicycle is problematic.Despite these quirks,accuracy levels in computer vision are improving rapidly-thanks toexponential growth in training data.For Goo
50、gle,that comes from all those public images harvested for Image Search and thebillions of photos backed up to the cloud in the Google Photos app.For research,these improvements open up lots of possibilities.EmotionAIEmotionAI is a catch-all label,but its being widely adopted.It covers the user of ma
51、chine anddeep learning techniques to classify users emotional states and identify their non-rationalresponses.Most models use NLP and Computer Vision techniques,with models grounded inneuroscience or clinical psychology.14Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analyti
52、csCheat SheetGOFAI/Narrow AIRules-based software programmes to classify or predict something.Used in expertsystems such as medical diagnostic support tools.MachinelearningFeeding an algorithm lots of input data(training)until it produces the right outputdata.SupervisedlearningA human checks the outp
53、ut data,gives feedback to the computer and if necessaryre?nes the model to improve its accuracy.UnsupervisedlearningLetting the computer?nd its own patterns in the data.Deep learningA sub-set of machine learning,with more complex network algorithms that try tomimic the human brain.AGI(Arti?cialGener
54、alIntelligence)Computers with broad human-like intelligence that can think and reason like us.NLPComputers interpreting language.ComputerVisionComputers interpreting images.The SingularityThe point at which AGI is achieved:machines surpass human-level intelligence,teachthemselves to be ever smarter
55、and either make the world wonderful or turn us all intopaperclips.By 2045,apparently-according to Ray Kurzweil.15Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsPart 2:Part 2:23 Practical Uses forAI in Research&Analytics16Insight PlatformsThe Insight Platforms Guide t
56、o AI for Market Research&Analytics1.Converting Speech to TextAKA automatic transcriptionIn May 2017,Googles voice recognition algorithm hit 95%accuracy-roughly on a par withhumans.Who knew that we mis-hear roughly 1 in every 20 words?But voice technology is still in its infancy:Amazon Alexa,Google H
57、ome and Apple Siri are alljust getting started.Today,there are several useful applications in research&analytics for automatictranscription:documenting user interviews or group discussions transcribing customer service or helpdesk calls for CX analytics making large volumes of video content searchab
58、le.Tools such as Trint,Otter and Zpoken Transcribe are easy-to-use commercial platforms fortranscription.Just upload your audio?les and o?you go.Pay by the minute,or subscribemonthly if you have enough volume.Trint:an example of SaaS Transcription platform17Insight PlatformsThe Insight Platforms Gui
59、de to AI for Market Research&AnalyticsMany of these user-friendly transcription platforms actually rely on AI tools from the bigcloud providers-Google Cloud,Microsoft Azure and Amazon Web Services.They all have AIsolutions that developers can tap into.Their speech-to-text algorithms are trained on t
60、hehuge volumes of data processed by their voice assistants(Google Home,Cortana and Alexa).Any developer can access these tools via API.(API-for those too embarrassed to ask-Application Programming Interface.These are toolsthat allow di?erent pieces of software to work together and exchange data.They
61、 arefundamental building blocks of the internet and the way that all software connects today.)Google Cloud Speech-to-Text service:an example of a machine learning APIBy the way,if the thought of developers and APIs sounds a bit too technical,its worthpersevering:the cost di?erence between the APIs a
62、nd the polished commercial products canbe signi?cant.As of January 2019,60 minutes of audio transcribed costs around$15 with Trint or$1.44using with the Google API.If you have more than an occasional need,its worth it.18Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics
63、The video management and analytics platforms(Voxpopme,Living Lens,Big Sofa,Plotto)alsoget a mention here.These tools include options for auto-transcription of video content-inup to 74 languages in the case of Living Lens.Living Lens:an example video management platform with automated transcription19
64、Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics2.Automatic TranslationNow that you have perfectly transcribed content in 74 languages,how do you make sense ofit?Unless youre an extreme sort of polyglot,youll need some help to translate those interviewtranscripts,open
65、-end survey responses and comments from review sites.Here again,the big cloud players have good options.Google Translate used to be hilarious:it was fun to convert a simple sentence into Chinese,then turn the translation back into English as something utterly mangled.But that happens less and less t
66、hese days,and it stems from 2 key factors that drive all AIadvances:huge amounts of data,the rocket fuel of AI:with nearly half a billion daily users,GoogleTranslate has gathered masses of feedback for its learning model smarter algorithms:in late 2016,the Google Neural Machine Translation systemlau
67、nched,and the quality of output improved measurably for all languages.Were not laughing any more.Google has lots of data and smart AI engineers,but its not your only choice.Microsoft Azureand AWS o?er translation APIs,but there are also specialist translation platforms providedby Smartling,DeepL and
68、 others.Smartling:an example automatic translation platform20Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics3.Analysing TextNatural Language Understanding-text analytics-is fundamentally changing how researchgets done:Surveys get shorter:if we can analyse open-end re
69、sponse better,why force people toanswer an exhaustive set of pointless rating scales?Let them say what matters to them,in their own words,in much less time.Data sources expand:researchers used to be con?ned to asking questions;now we canlisten meaningfully to what they say in social media,online rev
70、iews and call centres.Qual gets bigger:this sounds oxymoronic,but text analytics means we can now manageand interpret online discussions between hundreds of participants rather than just ahandful.Boiling it down to basics,text analytics software includes two core features for researchpurposes:keywor
71、d or entity extraction sentiment analysis.Keyword(or topic)extractionKeyword(or topic)extraction is relatively straightforward:even basic software can pick outitems from a body of text and track changes in volume over time.Entity extractionEntity extraction(AKA Named-Entity Recognition)is a little m
72、ore complex.This classi?esnamed entities into pre-de?ned categories such as the names of persons,companies,placesetc.This is particularly important if,say,you want to analyse reviews to understand howpeople talk about yours or your competitors products.There are existing models for entity extraction
73、(eg classifying automotive brands andmodels);but if you work in a niche category you will need to train a custom model to getsensible results.Sentiment analysisSentiment analysis tries to work out opinions expressed in text,with output expressed interms of polarity(whether the opinion is positive or
74、 negative).There are dozens of tools for general purpose text analytics,and you can?nd a long list ofsolutions in the Insight Platforms directory including Lexalytics,Odin Text,Decooda andothers.21Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsGenerally,there are thr
75、ee ways you can use these tools:through a web interface with an Excel plug-in via API.Gavagai Explorer is one tool that works through a web interfaceweb interface.You can upload CSV?les oftext or link directly to a SurveyMonkey project.Gavagai Explorer:an example of web interface for text analyticsM
76、eaningCloud has features for sentiment analysis,topic extraction and text classi?cation,and can be used in any of the three ways outlined above.22Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsMeaningCloud:an example of an Excel plugin with text analytics featuresFin
77、ally,a large number of API solutionsAPI solutions can plumb text analytics directly into otherwork?ows.MonkeyLearn is a solution accessed mainly through dedicated APIs and Zapier.It includesstandard models for sentiment,emotion and product classi?cation,as well as a full range ofcustom models.Monkey
78、Learn:an example of API-based text analytics23Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics4.Social Media ListeningSpoiler alert:this application and the two that follow(CX analytics,emotion analytics)arevariants of text analytics.But they are large and speci?c eno
79、ugh to be called out separately.Social listening is the process of analysing content in social platforms to understand topics,identify keywords,track mentions and measure sentiment.Data can be scraped(ie extracted from online sources)for one-o?projects using tools likeDexi or the skills of a program
80、mer;or most listening platforms have continuous feeds of datafrom a range of di?erent sources.Those sources are no longer purely social.Data can come from review sites like TrustPilot,e-commerce,news,blogs,forums,YouTube comments and Reddit-as well as Twitter,Facebook and Tumblr:List of data sources
81、 available in the Pulsar platform24Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsOnce the data is in the listening tool,it can be processed using the same NLP techniquesused in text analytics to.identify trends in a category measure brand mentions understand sentime
82、nt about speci?c topics or brands.Many listening tools are also tightly integrated with content publishing tools,and this isimportant to highlight:AI is helping to embed research and analytics functions insidemarketing platforms.Social management tools like Sprinklr,for example,are as much insight-a
83、nd-action tools formarketers as they are analysis tools for researchers.The Insight Platforms directory contains more than 50 social listening tools includingCrimson Hexagon/Brandwatch,Social Bakers,Meltwater and Pulsar.Pulsar:an example social listening platform25Insight PlatformsThe Insight Platfo
84、rms Guide to AI for Market Research&Analytics5.Analysing Customer Experience FeedbackCX analytics uses NLP techniques to understand Voice-of-Customer feedback for optimisingthe customer experience.Feedback sources were once purely survey-based,but can now include comments posted insocial media,forum
85、s and blogs;support tickets and transcripts of conversations withcustomer service teams;as well as more structured input from ratings&reviews,feedbackforms and the good old CSAT or NPS survey.The data has three main components1.A rating or score2.Explanatory comments3.Contextual data about the custo
86、mer and their purchase/behaviour/support request.Most CX analytics tools now have ready-built models for text classi?cation and sentimentanalysis in big categories(eg air travel,hotels,restaurants).But in most cases,these modelswill need to be adapted to the nuances of sub-categories or geographies;
87、and for nichemarkets,custom models will need to be built from scratch.There are two broad types of CX analytics tools for processing VoC content:Integrated collection and analytics platforms like Qualtrics,Medallia,Clarabridge,InMoment and MaritzCX.Standalone analytics platforms like Chattermill,The
88、matic,Wonder?ow,ipiphany andSentisum.This second group of tools dont have built-in survey capabilities;instead,they integrate datafrom di?erent sources;analyse and combine it;and visualise it in dashboards for analysisand action planning.26Insight PlatformsThe Insight Platforms Guide to AI for Marke
89、t Research&AnalyticsChattermill uses NLP to analyse customer feedback from surveys,CRM systems,supportplatforms and product reviews.Themes and sentiment are visualised in dashboard reports.Chattermill:an example of a CX Analytics platform27Insight PlatformsThe Insight Platforms Guide to AI for Marke
90、t Research&Analytics6.Emotion AnalyticsNLP models have been developed that focus speci?cally on decoding the emotional contentof language.Adoreboards Emotics platform,for example,analyses expressions of emotionto understand customer experiences.HearbteatAItakes text input from any source-survey open
91、-ends,qual transcripts,customerfeedback,product reviews-and classi?es it using universal emotion categories(10 primaryand up to 100 secondary emotions).Heartbeat AI:an example of an Emotion Analytics platform28Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics7.Conversa
92、tional Feedback(Chatbots)Beyond understanding,NLP has several other branches.Conversational interfaces andNatural Language Generation are two such areas that are beginning to have a major impacton the way that research data is collected,queried and reported.ChatbotsChatbots are being adopted as web
93、interfaces for sales,customer service and technicalsupport.They o?er cost savings and-in some cases-better speed and performance thanhumans can provide.That might seem surprising:some chatbots today are frankly terrible and deliver a shockingcustomer experience.But-like Googles translation algorithm
94、s-they are improving fast and will continue to do so:the more questions they are asked,the more data they have to train and re?ne their model,the more functional they become.One of the main chatbot uses in research is gathering feedback data.In this,they o?erseveral bene?ts over traditional online s
95、urvey methods:chatbots can be embedded in other applications(sales,service etc)so the feedbackprocess is more connected to the actual user/customer experience the same interface can capture both structured(quant)and unstructured(qualitative)responses,with the potential for a more natural?ow of quest
96、ions and answers as well as websites,Facebook Messenger,Kik and other chat apps can host researchchatbots,making it easier to reach certain audiences-younger consumers who dontuse email or respond to pop-up website surveys AI can reply intelligently and encourage deeper,more considered open-endedres
97、ponses from participantsIf you want to build a chatbot for research or CX feedback,you have two options:1.Adapt or build your own chatbot2.Use a dedicated chatbot for research or CX.If you want to create your own chatbot,solutions aimed at sales and service use cases(egfrom Drift,LivePerson and Hubs
98、pot)can be adapted to run simple surveys.29Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsCustom bot-builder platforms such as Botsify(for website and Facebook Messenger bots),Collect.chat(for websites)or Chatfuel(for Messenger)can also be used to build moreresearch
99、or CX feedback bots from scratch.These tools are designed for marketers,not developers-and do not require codingknowledge to get started.Collect.chat:an example chatbot builder platformIf adapting an existing tool doesnt work,there are now several dedicated chatbot platformsfor research and CX.30Ins
100、ight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsSurveySparrow has a wide range of question types,survey templates and question logic withoptions to embed chatbots on websites or in other software like Slack,Intercom andMailchimp.SurveySparrow:an example chatbot research
101、platformRival Technologies enables chatbot surveys to run in a web browser or be embedded inmessaging apps,and can be used to create both long and short term mobile-?rstcommunities.Rival:an example chatbot research platform31Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Anal
102、yticsWizu focuses on customer experience feedback,with templates for NPS,CES and CSATsurveys and integrations with CRM platforms.Wizu:an example chatbot research platformUltimately,these text-based chatbots are part of a broader evolution of intelligent assistants.Over time,they will be enabled thro
103、ugh voice services such as Alexa,Google Home andCortana.Rival Technologies and Rant&Rave,a Voice-of-Customer platform,have both experimentedwith building Amazon Alexa skills for conversational surveys.32Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics8.Big QualNLP too
104、ls allow us to manage,analyse and respond to large volumes of unstructured textdata.These techniques can be used to manage online discussions with large groups ofparticipants.Compared to a human moderator,Big Qual tools are much faster at summarising contentand extracting keywords.The Groupsolver an
105、d Quester platforms support semi-structured discussions with largegroups of consumers for brainstorming,concept development and idea screening.Remeshuses a range of AI techniques to gather feedback from up to 1000 participants overa sixty-minute session.Responses are analysed on-the-?y,clusters of s
106、imilar opinions arevisualised,and moderators can then focus on the most relevant topics or promising ideasand probe deeper in real-time.Remesh:an example big qual platform33Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics9.Natural Language AnalyticsJust as chatbot int
107、erfaces are changing how data is collected,so they are starting to evolvehow we conduct analysis.Microsoft Power BI is a business intelligence platform for visualising data in dashboards andreports.Its Q&A feature allows users to question their data using natural language;suggested visualisations ar
108、e then displayed in response.It even runs in the smartphone app:Microsoft PowerBI:an example analytics and visualisation platform with natural language query interfaceGoogle Analytics and other digital analytics platforms also have similar features.These virtual assistant capabilities are becoming m
109、ore widespread in research and analyticsplatforms.Lymbyc,Course 5 Discovery,NAVIK ResearchAI and Market Logic Software have aconversational query feature to help make their knowledge management platforms moreeasily searchable.These tools combine disparate sources of insight-presentations or reports,
110、surveys,transcripts,audience or shopper panel data,social content-with a natural language searchinterface.34Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsBloom?res knowledge management platform also incorporates natural language querycapabilities in its Scarlet AI e
111、ngine.In addition,it uses a variation of Googles PageRank algorithm to?nd content based on bothcontext and authority relative to other data;uses autocomplete suggestions to help guideusers;and recommends other content related to the users search query.Bloom?re,an example knowledge management platfor
112、m with natural language analytics35Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics10.Automatic Content GenerationNatural Language Generation is the process by which software uses input data to create acoherent narrative.NLG models have been used to write product desc
113、riptions on websites,poems and even scripts for Hollywood movies.The Associated Press used to employ writers to summarise the quarterly?nancial reports ofpublic companies.Today,NLG software creates those stories automatically,populatingtemplates using the underlying data.The results are indistinguis
114、hable from those created byjournalists.Research agencies employ sta?to synthesise data from standardised projects(like ad tests)or continuous trackers.They then spend time writing commentary for reports andpresentations.Much of this work can now be automated using the same tools adopted bythe Associ
115、ated Press.Narrative Science and Wordsmith by Automated Insights are software tools that generatenarratives from data.They integrate with analytics and business intelligence tools likeMicrosoft Power BI,Tableau and Qlik to write headlines,summaries or full stories.Zappi,the research automation platf
116、orm,uses NLG to write commentary on its productdevelopment and communication testing tools.Zappi:an example of Natural Language Generation for research reporting;eagle-eyed readers will alsospot the automatic Google translation feature.36Insight PlatformsThe Insight Platforms Guide to AI for Market
117、Research&AnalyticsNatural Language Generation features are also being applied in qualitative research.2020 Researchs Qualboard 4.0 platform now includes a Smart Reply feature for qualitativemoderation,trained on a million moderator-respondent interactions collected over 10 years.Like messaging apps,
118、it interprets the content of a post and suggests follow-ups and probesto save the moderator time when replying.2020 Qualboard 4.0,an example online qualitative research platform with text analytics and NLG features37Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics11.R
119、ecognising Content in Images and VideoFinding and identifying objects in images has several uses in research.Images posted to social media can be decoded to understand how often and in what contextindividual brands are used.Picasso Labs can identify objects in images from any public source to unders
120、tand howconsumers and in?uencers post about brands,categories or topics.Picasso Labs:an example of computer vision applied to social media researchSome social listeningsocial listening platforms such as Pulsar and Crimson Hexagon have also added thiscapability.38Insight PlatformsThe Insight Platform
121、s Guide to AI for Market Research&AnalyticsOur own photos can reveal a lot about our lives,behaviours and values.Pixoneye is a recentstartup whose opt-in smartphone app can analyse the content of a users photo library toadd richer pro?ling to segments.Pixoneye:an example visual analytics platformSho
122、pper diariesShopper diaries have been transformed in recent years through the use of smartphoneapps.These tools can recognise photos of packaging and classify the product or brand;andlong-term shopper panels such as Infoscout use visual analytics to decode the content ofscanned purchase receipts:Inf
123、oscout Omnipanel:an example shopper panel using computer vision for receipt scanning39Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsAnd computer vision is transforming category managementcategory management.Trax Retail uses millions ofimages of POS and shelf display
124、s from?xed in-store cameras and a crowdsourced panel ofsmartphone users.The AI recognises which products are being replenished,and combinesthis with Nielsen shopper panel data to build a fuller picture of category performance.Trax Retail:an example of computer vision applied in-store for category ma
125、nagementThe same technology that recognises objects in static images can be applied to video.Fixedcamera and smartphone auto-ethnography projects can generate enormous quantities ofvideo:computer vision AI helps to focus and speed up researchers analysis work.Living Lens,the video management and ana
126、lytics platform,uses object recognition softwareto identify and classify items within videos;researchers can then search for speci?c objectsto?nd the clips in which they appear.Living Lens:an example of computer vision applied to video analytics40Insight PlatformsThe Insight Platforms Guide to AI fo
127、r Market Research&Analytics12.Recognising FacesResearch applications for facial recognition are emerging slowly,as there are serious privacyand compliance issues-especially under GDPR in Europe and similar regimes indevelopment.Some current applications include:using AI to identify shoppers age and
128、gender from in-store cameras and combiningthat data with people-counting software(Ipsos Retail Performance)identifying repeat respondents in qualitative research to minimize fraud(Ipsos India).41Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics13.Virtual/Mixed/Augmente
129、d RealityResearch has used digital simulations of products,?xtures and even whole stores for sometime.Advances in VR headset design have brought more immersive software like InContextSolutions for mixed reality simulations.Augmented Reality technology has massive potential for researchers as it o?er
130、s the scope toask speci?c questions in context at the point of experience/purchase.Gorilla in the Room is an early innovator in this space,and a specialist provider ofAugmented and Virtual Reality research tools.Its software can be trained to recognise speci?c products,brands or ads using a smartpho
131、necamera.For consumer panels or diary studies,these real world triggers can launch a digitalsurvey to capture feedback in situ.Gorilla in the Room:an example of an augmented reality research platform42Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics14.Analysing Vocal
132、ToneRecordings of conversations sales discussions,customer support calls or even researchinterviews can be analysed using AI to identify emotions from both intonation andbreathing patterns.Beyond Verbal analyses this data for emotional valence,arousal and temper.inVibecombinesacoustic analytics(clas
133、sifying valence and arousal in a speakers tone ofvoice)withlanguage analytics(using NLP to decode the sentiment of transcribedconversations).inVibe:an example of vocal tone analytics43Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics15.Analysing Eye Movement and Facial
134、ExpressionsEye trackingEye trackingwith glasses(such as Tobii)has long been used by retailers and brands for POSplanning,and more recently for UX design and testing media with hi-res under-screencameras such as Gazepoint.Online eye tracking solutions using webcams include Sticky,Eyes Decide,Emotion
135、ResearchLab,Crowd EmotionandA?ect Lab.Facial codingFacial codingcaptures a viewers face as they watch an ad,a trailer or even longer-formvideo.Changes in micro-expressions are mapped using machine learning models to classifythese expressions into one of a handful of core emotions(anger,disgust,fear,
136、happiness,sadness,surprise or some variant of these).The outputs are usually a series of trace lines that show peaks or troughs of emotionsmapped against time-stamps in the video:Realeyes:an example facial coding platform44Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analyt
137、icsEmotion Research Lab,Realeyes,A?ectLaband A?ectivaall have facial coding solutions thatcan run over the internet using webcams.Advances in AI have helped to link eye tracking results and facial expression analysis withother data sources to build composite models of emotional engagement.Crowd Emot
138、ion links eyes,face and reaction times(via implicit response tests)to build acombined model of media engagement for advertisers and video content creators.CoolTool combines eye tracking,facial coding and click tracking behaviour to build acomposite picture of mobile user experience.Its UX Reality to
139、ol can be used to testsmartphone apps or mobile web experiences.45Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics16.Biometric FeedbackDevices cans be used to measure pulse rate,brain activity or electrodermal activity to buildmodels of emotional states.Where these me
140、thods used to be the preserve of wealthymarketing teams or university departments,they are now much more accessible.EEG(electroencephalogram)measures of brain activity typically capture cognitive state(asleep drowsy low engagement high engagement)and workload(boredom optimal information overload).Em
141、otiv provides both EEG headsets and software for analysingresults:Emotiv:an example biometric hardware and software platformGSR(Galvanic Skin Response)measures changes in sweat gland activity that are telltale signsof emotional arousal.Sensors like Shimmer can detect this,and software such as iMotio
142、nscan integrate biometric feeds from a range of di?erent sources.46Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsMindproberhas taken this online with its panel of biometric sensor-equipped respondentsand self-service tools for project management and analysis.It can
143、be used for testing ads,video or even live broadcasts.Mindprober:an example of an online biometric research platformThe way we touch our smartphone screens can also be modelled to infer our emotionalstate:are we stabbing the screen aggressively or stroking it a?ectionately?Emawww and Chromo both hav
144、e solutions that do this and can be embedded intosmartphone apps to provide developers with additional user feedback on their products.47Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics17.Data VisualisationData visualisation doesnt rely on AI;but the opposite is incre
145、asingly true:AI needs creativevisuals for telling complex stories to simple humans.Many platforms now display data in ways that clearly guide the researcher-or other non-specialist user-to the most relevant?ndings.This visualisation from the Qualtrics platform,for example,tells the user whats happen
146、ing with an automated summary(Check-in iscorrelated with NPS)and reinforces the message with visual chart components:Qualtrics:an example of guided visualisationAI techniques will increasingly be used to?nd the insights;visualisations will be more andmore necessary to translate those insights for bu
147、sy researchers and analysts.48Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics18.Price OptimisationMany markets operate dynamic pricing models:they vary pricing according to a buyerssegment,their purchase history,the time of day,the scarcity of inventory or dozens ofo
148、ther variables.In one famous case,travel company Orbitz showed how Mac users were willing to pay a$20-$30 premium over PC users.And weve all felt the pinch when Ubers dynamic pricing kicks in late on a Saturday night.Arti?cial Intelligence can take hundreds or thousands of variables to model the opt
149、imal pro?tor revenue maximising price point.For example,Perfect Price uses supervised machine learning and reinforcement learning togenerate prices on-the-?y for categories such as air travel,hotels and car rental.Perfect Price:an example dynamic price optimisation platform49Insight PlatformsThe Ins
150、ight Platforms Guide to AI for Market Research&Analytics19.Attribution Analytics“Halfthe money Ispendonadvertisingiswasted;the trouble is I dontknowwhichhalf.”John Wanamakers quote is more famous than the man himself,and until recently wasaccepted as a Marketing Fact of Life.But AI hopes to change a
151、ll that.It can process thousands of variables and millions of data points;use machine learning andregression to understand the impact on customer behaviour;and create models to explainthe relative contribution of di?erent touchpoints,experiences and campaigns.Many AI tools even take this learning an
152、d apply it by adjusting online campaigns in real time.Albert.ai carries out attribution analytics autonomously and automatically tweaks targeting,media buying and digital execution on-the-?y across email,mobile,social,search and display.Albert.ai:an example of AI-based attribution analytics50Insight
153、 PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics20.Customer Journey AnalyticsCustomer journey mapping is a well-established approach for visualising the end-to-endcustomer experience.Customer journey analytics takes this a step further by?lling the map with data.Everycustome
154、r interaction-across channels and over time from millions of touchpoints andevents gets modelled into journeys to help understand,analyse and shape the customerexperience.The aim is to give marketers and CX managers a way to prioritize the right touchpoints andexperiences that will grow revenue or r
155、educe churn.Pointillist is one such journey analytics platform.It uses machine learning and predictiveanalytics to test hypotheses and model future behaviour.Pointillist:an example customer journey analytics platform51Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics21
156、.SegmentationUnsupervised machine learning models can identify clusters of users in large data sets:e-commerce transactions,CRM records,location,subscription or survey data.These models do not need to be built around a priori hypotheses;they can be used toanalyse records of user behaviour,spending,p
157、reference,demographics or anything else and?nd the commonalities that occur naturally.When combined with good domain knowledge,the resulting segmentations can show upnew opportunities for communications and customer management,The A?nio platform uses machine learning to build custom audience segment
158、s bycombining and analysing?rst and third-party data.A?nio:an example segmentation analytics platform52Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsFrontier7 is a primary research management and analytics platform that uses unsupervisedmachine learning for automati
159、c customer segmentations.Frontier7:an example segmentation analytics platform53Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics22.PersonalisationAI-based personalisation is a key tool for social,ecommerce and media platforms.Complex models drive Amazons product recomm
160、endations,Facebooks people you mayknow suggestions and even the targeted movie posters displayed to users on the Net?ixhome screen.Natural Language Generation is used to create tailored content for email communicationsand programmatic ad campaigns.Platforms such as Pure Clarity help marketers to del
161、iver these experiences;they also providea rich source of insight and analytics for researchers and data professionals.23.Enhanced PredictionThis last category really is a cop-out.You can make a case for saying that ALL arti?cial intelligence is about prediction,and thereare far too many applications
162、 and examples to do justice to here.This is just a small selection:Research agency Strategir has built a proprietary AI solution that can transform 200survey respondents into a virtual shopper panel of 2 million consumers-to supportmuch more granular and accurate modelling of purchase behaviour.A fu
163、ll case studybased on Strategirs model is available here.Retail sales data,social buzz and other indicators are combined into predictive modelsby tools such as Trendskout and Trendscope from Black Swan Data.The most visually engaging elements of a web page,campaign or POS display can bemodelled usin
164、g AI without the need to conduct fresh A/B tests or primary research;tools such as Eyequant and Dragon?yAI provide this A?ectlabs platform has millions of data points from historic facial coding,eye trackingand brainwave mapping research projects.This data is powering a machine learningmodel that wi
165、ll be able to predict emotional engagement with new content withoutneeding fresh respondents.54Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsPart 3:Part 3:A Research&Analytics Blueprintfor AI Success55Insight PlatformsThe Insight Platforms Guide to AI for Market Res
166、earch&AnalyticsSo what do you do about AI now?This?nal section maps out 5 strategies for insight teams and agencies to maximise the AIopportunity.These new technologies will eliminate some old roles and create some new ones;they willmove certain types of work into clients and out of agencies;and the
167、y will bring newopportunities and challenges on both sides of the fence.Navigating the world of AI wont be straightforward;but there are clear steps you can take toincrease the chances of success.56Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics1.Improve your Busines
168、s-as-UsualAI for research is evolution as well as revolution.In the short term,one of AIs biggest impacts will be to improve current ways of working.There is much that is being disrupted in research;equally,there is much that will continueand evolve gradually.In-depth interviews,online surveys,acces
169、s panels,reports and presentations:these will allcontinue,and in most cases continue to grow.AI tools will enhance these methods byreducing timescales,improving data quality and cutting overheads.Some simple examples include:Transcribing and translating interviews:Transcribing and translating interv
170、iews:despite the scale of online research,a lot still takesplace face to face(social,qualitative,usability,in-home,B2B).Text-to-speech applications willhelp transcribe discussions on the?y or very quickly afterwards.This will reduce the costs and timescales of human transcription;make the content of
171、interviews searchable and databasable;and allow text analytics to?nd and quantify themes,supporting the work of the core researcher.Reducing fraud on survey panels:Reducing fraud on survey panels:panel fraud is a major challenge for online surveys(botfarms,individuals with multiple accounts,dishones
172、t respondents etc).AI tools can helpmitigate these risks and improve data quality.Panel technology?rms P2SampleP2Sample and VIGAVIGA have built machine learning models to recognisefraudulent behaviour.The algorithms use a combination of historic survey responses,digitalbehaviour data and known indic
173、ators of historic fraud to improve quality.Improving predictive models:Improving predictive models:using surveys to forecast the uptake of new products orservices has long been a staple of consumer market research.Combining these techniqueswith AI models can improve their predictive power,as Strateg
174、ir has demonstrated.57Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsPractical StepsPractical Steps1.Identify the lowest hanging fruitIdentify the lowest hanging fruit.Go through the 23 applications above and give eachone a score out of 5 on 2 measures for your team
175、or business scale of impact andease of implementation.Multiply the scores together(max 25)to give you a prioritisedranking of opportunities.Start on those with the highest score.2.Set your goals.Set your goals.You have 3 dimensions to play with cost,return and timescale.Thecost side of the equation
176、needs to include the time you invest(dedicate at least 20%ofan FTE to manage this process),the tangible software costs and the learning costsattached to rollout.The return should be both the dollar amount saved and a proxy forthe value generated(stakeholder impact,new business won etc).And you need
177、to setmilestones for achievements at 3,6 and 12 months.3.Hold the beauty parades.Hold the beauty parades.Chances are,youre already besieged with tech vendors tryingto pitch you.So make the most of it:invite the AI solution providers in,share your usecases,get them to prove their value.This is the fu
178、n part,and you will learn a lot.Enjoyit.58Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics2.Build New Client-Agency Models Agency engagement and remuneration models are broken.Nobodys happy:clients feelunder-served and agencies dont make enough money.And it looks set
179、to get worse in the age of AI as client teams bring more activity in-house andagencies struggle to?ll their revenue gap.But it doesnt have to be this way.The best agencies will play essential roles even as AImatures:they bring external perspective,deep domain expertise and burst capacity essential b
180、ene?ts for client teams.Modern ways to procure and manage agency expertise include true retainers to lock inrelevant expertise,customer success(technology+support)models for in-house platformsand researcher marketplaces for occasional expertise or extra bodies on short-term projects.Practical StepsP
181、ractical Steps1.Client teams:get creative with your strategic agenciesClient teams:get creative with your strategic agencies.There are three of you in thisrelationship now-you,the agency,and your AI platforms.You need to bring those keypartners properly inside the tent to co-create new work?ows and
182、commercial models.You need to be transparent about budget;they need to be transparent about how theymake/lose money;and you need to build joint approaches to working with shared AIplatforms.2.Agencies:focus your value proposition.Agencies:focus your value proposition.You need to deliver what AI cant
183、(yet):contextual insight,engaging narrative,deep expertise.Ten years ago,there was plentyof money to be made collecting survey or focus group data and writing basicPowerPoint reports.Those days are over.3.Both:make longer term plans.Both:make longer term plans.That sounds tricky when everyone parrot
184、s truisms aboutthe extreme pace of change.But strategic client-agency relationships need on-going 24-36 month horizons.Without that,there is not enough joint con?dence to make theinvestments that are needed.59Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics3.Embrace E
185、mbedded InsightsMore days that are over:those when every research project was run through an insight teamor outsourced to an agency.Software has put research tools directly in stakeholder hands,and AI will massivelyaccelerate that trend.Every department is expected to be user-centric/consumer-driven
186、/customer-?rst.What does this mean in practice?It means that they get their own speci?c insights about those customers engage with users directly rather than through a research team build that customer feedback directly into their own work?ows.AI will put more insight tools in the hands of non-speci
187、alists everywhere.They wont need tobe experts in research and analytics:the software will bring it to them.This embedded insight model means that far more research will be done outside researchteams than within them.Here are some examples where embedded insight is happening today.Customer Experience
188、 teamsCustomer Experience teams:many CX teams now manage their own Voice-of-Customerprogrammes without ever engaging their research colleagues.Enterprise platforms like Medallia,Qualtrics and InMoment gather customer feedback atevery touchpoint and include AI features for segmenting customer groups,
189、sentimentanalysis and alerts.Startups and digital businesses can use tools like Wootric,AskNicely and Retently to capturemetrics like NPS and carry out text analytics of verbatim feedback.Product Management teams:Product Management teams:digital product teams need a constant stream of feedbackabout
190、users:how many,how often,where they click,their journey?ows,whether theyconvert.Users on-site or in-app behaviour is captured through tools like Heap and Google Analytics;increasingly,this what data is being augmented with why data in the form of user feedbackin platforms like Apptentive,Usabilla an
191、d Mopinion.60Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsIntegrated platforms like Hotjar include surveys,forms and user recruitmenttools as well asin-page analytics,heatmap generation and sessioncams.AI is used to combine data sets,model preferences and trigger t
192、argeted requests for feedback based on behaviour.UX Design teams:UX Design teams:designers can now go straight from prototype draft to user feedback in afew clicks.In a moderated online test,the designer can see both video of the participant and theirdevices screen.Recordings can be transcribed and
193、analysed using both NLP and computervision.Adobe XD is design software that integrates with UserTesting;and prototyping platformMarvelintegrates directly with Lookback.Practical StepsPractical Steps1.Get over it.Get over it.The?rst step is acceptance.Insight teams and agencies need to stop gettingan
194、noyed when stakeholders do their own research.Stop disparaging it as quick anddirty or insinuating that its not proper because they didnt involve you.2.Support and enable.Support and enable.You have the expertise in research design and analysis so shareit.Be advisory rather than executional.Hold bes
195、t practice workshops,write trainingguides and try to in?uence the choice of tools.61Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics4.Always be LearningIf youre reading this far,Im probably preaching to the choir.AI represents a step change in the tools and data sourc
196、es available to researchers.It poses amajor learning challenge for an industry that under-invests in training&development.One recent report by Ray Poynter of NewMR suggests that 1 in 4 researchers get no trainingin any given year;and that only 1 in 5 get more than 2 days a year.This is woefully insu
197、?cient if researchers want to thrive in the age of AI.Practical StepsPractical Steps1.Dedicate time to learning.Dedicate time to learning.For you personally and for your teams.Best practiceguidelines vary,but plan for a minimum of 2 hours per week.2.Re-think how you do it.Re-think how you do it.Too
198、much training is triple-F:face-to-face,full days and quicklyforgotten.Micro-learning like Googles Whisper Courses can be far more e?ective.Online platforms like Udemy have hundreds of self-paced courses to help with AI-related topics for researchers including machine learning,data science and conver
199、sionoptimisation.62Insight PlatformsThe Insight Platforms Guide to AI for Market Research&Analytics5.Do What AI CantComputers cant do empathy and they struggle with context.For all AIs achievements,applications remain narrow models are good at one speci?c thing and useless at everythingelse.AI will
200、increase the volumes of data available to everyone working with insight and analytics;it will also drive a parallel increase in the need for real human insight and relatablestorytelling.The data will help to measure,diagnose and plan at speed and scale;creative insights andmoments of understanding w
201、ill come from individual,qualitative connections.Specialists with research skills in these areas will be in high demand.For example:Digital products rely on streams of performance data,but qualitative user interviews,observations and feedback are at the heart of the design process.Understanding peop
202、le andtranslating those insights for engineers is a critical skill for user experience researchersuser experience researchers.Context is everything.Platform analytics can generate thousands of rows of data for a singlecustomer using a single app.But it cant see what else is going on in the users lif
203、e or how itmight impact their behaviour.EthnographersEthnographers are experts at making these connections.Managers need to mingle with the people who pay their mortgages to avoid the idea ofcustomer becoming an abstract notion or a series of data points in a dashboard.QualitativeQualitativeresearch
204、ersresearchers will be needed to facilitate customer closeness sessions to connect executiveswith the humans they sell to.Practical StepsPractical Steps1.Get out from behind the desk.Get out from behind the desk.AI will put more and more data on your screen but itrisks separating researchers and ana
205、lysts even more from the people they hope tounderstand.Spend time with users,watch them,listen to them.2.Use anecdote and illustration as much as dataUse anecdote and illustration as much as data.Insight leaders drive change bycombining evidence and metaphor;let the AI deliver your evidence,but make
206、 usecreative metaphor to land your story.63Insight PlatformsThe Insight Platforms Guide to AI for Market Research&AnalyticsFinal ThoughtWhen looking at any new and potentially disruptive technology,its always worthkeepingAmaras lawin mind:We tend to overestimate the e?ect of a technology in the shor
207、t run and underestimate thee?ect in the long run.AI will change research&analytics dramatically even if you havent felt its e?ects so far.Embrace it and evolve;try not to fear it.If you found this e-book useful,make sure you If you found this e-book useful,make sure you register for a FREE Insight Platformsaccountregister for a FREE Insight Platformsaccountand and stay in the loop.stay in the loop.And if you want to discuss anything in this ebook or need help?guring out AI,send me anemail or message me on LinkedIn.64Insight Platforms