trendpreneur

The afternoon kicks off with a talk about data and customer understanding from Beyond Analysis. The entire afternoon’s talks are behind the jump – usual disclaimer, these are unedited notes, etc.

- looking back we can predict the future – appropriate to supermarket etc, lots of customers, lots of regular interaction – but what if you’re not? what if you see your customers infrequently?
- understand consumers when they’re not in the store
- infrequent/irregular transactions > hard to draw conclusions, insufficient data.
- but looking at behaviour data -> what customers are really doing. actions > words. brand perception doesn’t always line up with market share (chewing gum).
- questionnaires channel consumer into a specific frame of mind. [tweet; this isn't how market researchers do it any more]
- but what if we don’t have the consumer in the room? -> smm
- [... more stuff that isn't new, people use social media, it's growing, etc]
- dell example
- orange. saw no association between a ‘problems’ website and internal customer service. analysed it all. magic.
- comcast [cares].
- challenges; high level of human input even in automated tools – labour intensive. AI engines – orange the company vs orange the fruit. How much credence to give to findings. Context – blog with 1 important reader…
- no data set can help you fully understand your customer when considered in isolation. link them together – context + foresight = true insight. benefits of 360 view are seen by a lot of companies.
- social media = data collection tool as well as comm channel. note danger = hyperbole around SM may prevent customers using it in a meaningful way.
- B&Q – pulls together 20k data sets from organisation to inform how company is growing. SMM is one of the data sets. context matters.

- Q- single most important thing brand should be doing – think about how social media sits among other data within the organisation > how it’s going to add to the picture. “is it on a par with other data?” used correctly, can be more powerful.
- Q- ROI of analysis tools – Within overall context they’re reasonably priced. ROI is function of whether organisation set up to correctly use the tool.

Next up – Movember participant Giles Palmer on data. CEO Brandwatch.

[story about company past - had an idea, built something, etc]

what do SMM tools do
- gather search analyse present collaborate integrate.

gather
- how do you get all the information?
- having access to the data is an extremely difficult data challenge. they don’t buy in data. buying in is pretty smart, isn’t cheap to do it inhouse. use data aggregators. even if you buy it you have to store it – challenges of distributed data. it’s a massive exercise.

search
- it’s not the same as normal search. things like proximity search etc. google and friends rank by relevance – don’t have to worry about the long tail, less relevant hits buried deeper in. but in SMM you do care about long tail. there is no unimportant data. build up search strings – boolean operators – example of amazingly complex Shell string (mostly -”nut shell”, -”shell-shock” etc).

analyse
- structured [no. comments, no. links, author name..] and unstructured data
- challenge – doing it every day – every page has a different structure..
- unstructured data is even harder – text means different things to different people. sentiment and topic/theme analysis. over-claiming/jargon that goes on in this area.

sentiment
- sample 400k web pages, 1k topics, analysed by humans: 58% neutral, 28% pos, 14% neg.
- top brands in october – google, MS, sony, ebay, BBC (sentiment*volume); brighton, hershey, x factor, talent shows, fedex (sentiment avg)
- people don’t agree with each other all the time. in their tests – over 1k items, people agree 85% of the time. [this is inline with academic results]
- how do you measure sentiment of millions of web pages (40-50m every day)
- crowdsource? expensive. slow. inconsistent. calculations at 30 articles an hour, £2000 for 10k articles.
- machines? kind of. not as good. machine learning. [not NLP? seems to refer only to bayesian type classification]
- hardest bit isn’t machines doing SA – it’s deciding which words/sentences you send to the machine [feature set]. individual article vs forum posts- which words refer to the subject you’re trying to sentiment-analyse? bigger impact on sentiment analysis than the classification algorithm.

stats and recommendations
- 650k mentions in 60 industries – the language per industry varies wildly
- 94% swedish videogames ["fluke"], 50% portuguese telecom
- target 75% but “bloody difficult” – if you classified everything as neutral you’d get ~60%!
- for small volumes crowdsource, large volumes use machines but look at most important mentions _manually_

Other analysis
- topic analysis, network maps and influence
- topic clouds, network visualisation (they don’t do it), influencers (“bloody difficult”)

Collaborate & Integrate
- passing data within and without organisation – agencies doing it right now. collaboration around data becoming more important, as is integration throughout the campaigns etc [see earlier presentations]

free giveaway! www.brandwatch.net/4 – 50 free beta testing accounts for a month.

[side note - people confusing semantic (understanding) with sentiment (tone) analysis]

Q&A – real challenge re sentiment is not algorithms but which words actually talk about you. [relevance.]
- Q- access to data – will we get more or less? Probably less (murdoch) > cost of tools increases.
- Q- solutions for very large amounts of data? distribute and “buy shitloads of servers”. can ditch useless data. stick it on the cloud? done the maths, and it’s not cheaper. it will be, but not yet.

Now – Brad Little from Nielsen BuzzMetrics on free vs paid.

- setting the scene – social media – millions of google results. 364 people on twitter call themselves “social media guru”.
- there are a *lot* of people monitoring buzz.
- why so many approaches?
- different objectives = different tools. no one type of social media.
- different investment levels drive various approaches.
- ability to listen in this way that’s new – not WOM itself. Do you *really* want to get to know your customers? It’s often about stock price, not action.
- DIY/Free tools – they’re free!
- Software – probably higher quality, more data, do more things
- Analyst – people + data – quality and expertise
- Consultant – Relationships – great access to clients. Actionability – what do we do with data.
- many free tools. forrester wave listening platforms Q1 2009 – looked at offerings.
- weakness:
- DIY – partial view
- Free – limited scope
- Software – reduced accuracy
- Analyst – speed, cost
- Consultant – using tools from 1-3

steps involved
- data collection > analysis > implementation [didn't want to repeat earlier talk]
- example – needed to harvest EA forum for a game project
- do we measure a tweet the same way as a blog or forum post
- how do we clean the data – GIGO
- people vs technology. local, multilingual, centralised teams? keywords or logic? “gossip AND girl”. massive query needed to isolate. markets of languages – just because you collect data doesn’t mean you can properly analyse it.
- implementation – recommendations & strategy.

Dyson example – air multiplier.
- what is actually being measured [examples of coverage]
- tech support issue, whether dyson is standing behind its products – analyst can see themes far easier than computer.
- mixed sentiment words in negative review [love, clever].

featurebabble – would take more than one fulltimer to compare these.
tools locked in ‘feature war’ – distracts marketplace from advancing.
tools will largely commoditise. differentiation in data quality / breadth, services provided, and expertise implementing actions.

-what are most of these tools trying to do?
- cover lots of data
- relevant and clean informatino – to save time
- manage process of workflow – save time, max effectiveness
- customise end output & result – user control
- liberate content – uncover what you need without restrictions
- support – value++, realise other benefits.

who is more influential?
- influence isn’t just quantitative metrics (x number of followers etc) – it’s also about what they’re saying. advocacy as metric, not influence.

FAQ
“can’t we get this stuff for free” – not really as good, less data, different output
“your data’s rubbish because google has way more results” – spam, redundancy, etc. apples vs oranges.
“compare providers by running a search and seeing who has most buzz” – that’d just incentivise companies to include spam! different coverage etc.
“a trial’s a good way to compare services, right?” – no it’s rubbish. tool provider – trials a matter of numbers – sophisticated queries etc require human time investment, customisation, effort. not just easiest to use right off the bat, got to set it up to get max performance [argument against freemium model here]

social media process: listen, learn, execute. [same stuff as earlier]

when unlocking value of tools please remember
- all data is not created equal
- dashboards have strengths but don’t answer every need
- combine research methodologies – listening *and* asking
- active client participation is key ***
- actionable insights = when tools (tech) combine with good local market researchers (people).

“buzz is up 5X” is a fact not an insight.

final recommendations
- determine stakeholders
- what you want to get out of it and what you might want to do with it
- look under the bonnet, don’t just talk to salespeople
- understand difference between monitoring, researching and strategy
- listen learn and then engage
- event, issue, launch or specific (anything) => more interesting research than general buzz.

WOM and SM are people based endeavours, and so is the research.

@bradleyjlittle

- Q- Someone’s influence (quantitative) *is* an indicator of content quality. -> Understanding how they talk about something is important – need a “BS meter” to distinguish quality/authority.

Panel – Jos Smith chairing “What’s wrong with SMM?”

Nick Koudas – Sysomos. Emmanuel Vivier, Asi Sharabi, Mark Rogers.

Nick:
- Spam. 65% of information in blogosphere is spam – noise > signal.
- sentiment analysis – more time = better. trade off.
both of the above are key technology challenges.
- pricing.
good; adaptive, global, granular focus, exhaustive coverage, interactive.

Emmanuel:
- cost to evaluate platforms – make it easy to try out and preach solution to clients. should Just Work. are you a software solution or a consultancy.
- if you don’t put people on top of good solutions you’ll not have the best answers even if you use the best tools.

Asi:
- umbrella with holes in
- unsustainable fishing
- tools – small print – need plenty of analyst time.

Mark:
- Unknown unknowns – how do you find the stuff you don’t know – most of it serendipity at the moment.

How can the industry make it easier for clients wanting to use SMM?
- mark – show how it really relates to money in the door. the winner will be able to show a lovely graph correlating SM with sales.
- within every big org there needs to be someone with a sophisticated understanding of what it all means – but it’s a lot of work. need some good people both sides.
- nick – ease of use of tools. if you say ‘lets do social media’ need clear business metrics. sometimes that’s not analytics but insights.
- emmanuel – not just a black box – need to know how it works and how to customise it.
- ann longley from audience – research – having done it manually using free tools and now using paid tools – in spite of limitations, tools have a great advantage. trick is being cost-effective, having clear brief.

spam
- what is it? malicious advertising, self promotion. big problem. twitter, most content produced by 5% individuals. most of that information is bots. ongoing battle. the moment there’s a defence, there’s a new technique.
- “if it can’t be spammed, it’s not social media”. everyone’s going to want to fake conversations. most spam right now about playing google, but that’s changing (coming to get us). pattern matching – keywords – vs links by real people / links by spammers and infer.

resources
- fulltime post – ’social media monitoring dude’ inhouse. agency side takes a while to strike balance – internal vs client resources.
- a lot of over claiming, ubiquity of google problematic.

final comments
- listening is only part of the problem. don’t spend all your money on the tools.

After a tea break it’s We Are Social’s Robin Grant on how they helped Skype.

‘facilitating conversations for its clients will become the new role of an agency’ forrester

broke SM strategy down into three areas.
- corp blog had turned into one-way press release delivery system.
- took blogs from one-way to a genuine conversation platform, made sure everything on topic.
- made sure they were responding to comments, surfacing and documenting stories. put product managers in front of the public.
- trust – people trust other people, not adverts. impacts purchasing behaviour.
- online conversations do drive sales – Gruhl & Guha paper. 2 day lag between conversation and sales.
- we can start conversations – sent a phone to Loic – but that’s a small drop in the ocean. background chatter – respond to it. [but sometimes corp comments on blogs can go down the wrong way]
- create moment of positive WoM – skype reply to random help request ‘does anyone know how to…’ ‘great Twitter moment, Skype blogger sends me an answer’. rep++.

how did they do it?
- none of the monitoring tools are perfect (“none are that great”) – looked at them – wanted every single conversation so used publicly available tools and built system on top of that. “need a human being reading every one of those conversations” > triage process. yahoo pipes, etc.
- identified key communities and individuals
- set up keyword alerts (lucky, specific keyword)
- triage – make sense of conversation. filter for items that can be meaningfully responded to. initially low bar – but as confidence increased, so did participation – there is no bar. as soon as you remove ‘faceless’ aspect of corporation people less likely to rant and rail.
- assign most appropriate response – online ticketing system.
- Review monthly – dashboard for insights. overview. alerts, twitter, getsatisfaction, categories over time, trends over time.
- categories, priorities designed to fit with skype’s needs

effective even in times of crisis
- crisis when this system set up – china spying news story
- spread like wildfire
- usual approach – terse press release
- got CEO to write a blog post
- went back out to conversations on social media and left a response
- that then got referred to in future posts
- sponsored adwords ’skype+china’ – relatively unnecessary – skype’s own blog got #1 search result within 24hrs.

Q&A
- Q- who’s “we”/who’s doing what? – It’s a transitionary process, initially outsourced to we are social, but over last 18 months that’s mostly gone inhouse. now doing it all internally – takes 30-45 minutes in morning and evening.
- Q- not realtime then? – on a normal day.. no. product launches, announcements -much closer eye. [but they can't spot crises within minutes?]
- Q- what’s PeterAtSkype’s fit within the company? – ex we are social employee now inhouse at skype. someone who really understands & is passionate about social media.

Marshall Sponder – Impromptu panel
Future of SMM
webmetricsguru.com

- social media platforms today are immature – focusing on quantitative metrics not actionable data.
- forrester wave – only the largest platforms were listed – >$10m revenue and x enterprise clients. many of the interesting technologies excluded.
- nobody’s defined:
- what conversations are > how conversations are measured.
- what’s really positive, negative and neutral [er - we don't?]
- how listening platforms integrate with data from other systems – call centres, POD, analytics… data still silod.
- meaningful consulting services to help businesses engage with data.

sentiment analysis today is too much like quantum physics
crimson hexagon vs techrigy, sm2 etc. get something different out. what is the reality? the truth?

today 99% of the time, social media takes a long time before results
- so what should we monitor to show success
- no clear indication what to measure, what to expect after 3/6/9/12 months..
- no clear idea how much to spend
- no standards on what SM is and how it should be measured

at least in web analytics – might not agree how we measure a visit – but we have consistent metrics- bounce rate, etc.

most platforms today not capable of advanced semantic analysis/meme clustering. [not really semantic analysis but more meme/attitude categorisation]

awful geolocation capabilities – out of 312 blog entries located in NY, only 30 were from bloggers in NY [Alterian] – more luck with blogger directory, not tool.

- overlap free vs paid tools. SMM solutions wiki.
- No standards for social media and little desire for them from vendors.

so what does the future hold?
- meme tracking common capability of select paid & free tools. along with ability to act on data. crimson hexagon leading now. would be good to pull out twitter handles – drill down.
- integration of social media with web analytics, crm, search. google could accelerate this, merging SMM into google analytics.
- new SM teams formed, merges etc
- social search fuel development of new SMM capabilities. sidewiki comments, google caffeine > real-time search results, google waves [lots of google....]
- businesses will adopt SM via real time search results – they’ll have no choice. (smm as new SEO)
- as a field? look at web analytics recent history. merging SMM with business intelligence
- maturation of common set of definitions. thisisbeta.influencescorecard.com. seeing a need for standards.
- open source modular tools > assemble and customise our own monitoring platforms. current lock in even though everyone’s using the same data.
- facebook data
- real time alerts in facebook
- social media budgets more closely aligned with paid/organic search budgets (due to move of RT search into google, bing)
- monitoring analytics for SM – part of marketing budgets in many corps.
- first true keyword tools to be released – conversations not keywords – tools will be released that allow optimisation of social media similar to SEO [this seems a bit left field.. social media doesn't work on keywords, it's people driven]
- event horizon – are we too close to SMM to predict?

slides here

Panel – The future of social media monitoring
Neville Hobson, Marshall Sponder, Matthaus Krzykowski, Philip Sheldrake and Andrew Grill

- how do brands who’ve been programmed how to market act in the ‘real life’ of social?
- social media will be part of lifeblood of company. recruiting, growing company, strategic planning tool.
- social media subset of ’social web’ – the network itself is interesting, once we break away from proprietary – internet of things
- new brand of SMM tools – two startups with 95% success rates on specific industries. lot of innovation in silicon valley. twitter will get into the game. social gaming is big. hollywood model is doomed- tweet on a friday and nobody will go on saturday.
- more emphasis on business intelligence – it’s being absorbed. social media is currently separate/added in – it’s going to become integrated. semantic web going to solve a lot of problems. microformats stuff has to happen. standards body. [ed- how do you standardise the concept of relevance.....]
- lot of work is cleaning data – if you can eliminate the part that isn’t about your clients.
- accuracy/verticals – 95%
- automated sentiment isn’t there yet => manual classification with expert analyst familiar with media – you can’t train bloggers – can’t do everything automatically. dealing with mixed-sentiment posts.
- will brand mass monitoring become automated hell?
- self-regulating – social media start to regulate it – habitat example. it can happen but will be squashed.
- current tools – problems. even vendors criticised have been very generous. the infrastructure to bring them together hasn’t happened. problem is usually with procurement – organisations don’t know how to procure – so end up with not-quite-right solution that compounds – no wonder things don’t work. need people to bridge that.
- products will get better – that’s how technology works
- problems are overrated. services are good enough. get 3 hours job down to 10 minutes is the crux.
- “that happened a year ago in the US” – cycle times getting quicker. industry needs to listen. if it makes sense we’ll do it.
- similar problem with data analytics – understanding question + matching capabilities of tool to it.
platforms aren’t mature.
- “we didn’t have half of this stuff 2 years ago” – we’re not there yet either. some of the free stuff is really good.
- impact of google wave. – “what is it”
- if we shut down facebook and twitter then all our businesses cease to exist / what if twitter launch their own SMM service and charge us for data. lot of insecurity in market. risks.
- open v proprietary – origin of web free – now everything’s proprietary – walled gardens. one-many pull system will take its place. EU privacy rights.
- language and culture. important – that’s why we use humans.
final comments
- listen, learn, engage, integrate
- fight for your individual rights
- get a job in social media monitoring