AnalyticsCube

Tracking bags, personalizing offers, boosting loyalty, and optimizing operations are all goals of a renewed data-driven approach by major airlines.

When a customer checks into a flight with United Airlines UAL -0.87% , there is typically an array of potential add-on offers to navigate through: flight upgrades, access to the airline’s United Club, and more.

Under United’s old “collect and analyze” approach to data, the airline would use information about customers’ choices about those items, in aggregated fashion to “see what the most successful products were, and market with those [insights] in mind,” said Scott Wilson, the company’s vice president of e-commerce and merchandising.

That approach has changed. As of the beginning of this year, “collect, detect, act” is United’s new data-focused mantra, and it’s changing the way the airline serves its customers.

“Now we look at who the customer is and his or her propensity to buy certain products,” Wilson explained. More than 150 variables about that customer—prior purchases and previous destinations among them—are now assessed in real time to determine an individual’s likely actions, rather than an aggregated group of customers.

The result, delivered in about 200 milliseconds later, is a dynamically generated offer tailored to the individual. Its terms, on-screen layout, copy, and other elements will vary based on an individual’s collected data. For United, the refined approach led to an increase in year-over-year ancillary revenue of more than 15 percent, he said.

‘Airlines evolved big data’

Welcome to the big data era in the airline industry, which in many ways was one of its earliest participants.

“Airlines are awash in data, much of it unstructured,” said Bob Mann, an industry analyst with R.W. Mann & Co. But only recently have airlines been able to use big-data techniques “to solve, among other objectives, how to recognize and enhance customer value, and how to cultivate high-value customers,” he said.

“Airlines have always been very good at collecting data, but they haven’t always been good at using it,” United’s Wilson said. Now that the costs of storing and processing data have dropped—even as airlines collect more and more of it—it’s becoming easier for a company to act on it. At United, roughly a terabyte of customer data is floating around at any given time within its systems. “We don’t keep it all,” Wilson said. “We have to be selective about what we grab.” For the data that is selected, a real-time decision engine does the crunching to turn it into something useful.

It starts at the baggage carousel

One area in which the effects of big data technology are visible is in the handling of customers’ luggage. “We have over a number of years invested millions of dollars in baggage tracking,” said Paul Skrbec, a spokesman withDelta Air Lines. “That was one of those core, behind-the-scenes services for our customers.”

Millions of bags are checked each year with Delta DAL -1.46% —a total of 130 million are projected for 2014, Skrbec said—and “every customer has had the experience of boarding a plane after checking their bag and wondering if it was there.”

Through the use of hand-held baggage scanners used at passenger check-in, “we’ve had all this tracking data available,” Skrbec said. But “one of the things we realized about two years ago is that customers would benefit from having that information.”

Which is why Delta was the first major airline to launch an application allowing customers to track their bags from their mobile devices, he said. Spanning the iOS, Google Android, BlackBerry and Windows Phone mobile operating systems, the free app has been downloaded more than 11 million times.

In search of new revenue streams

It’s a similar story at Southwest Airlines  LUV -1.43% , which is using big data to determine which new customer services to implement.

“Southwest uses aggregated, anonymous customer data to promote products, services, and featured offers to customers on multiple channels, devices, and websites including Southwest.com,” said Dan Landson, a company spokesman. “By observing and looking into customer behaviors and actions online, we are better suited to offer our travelers the best rates and experiences possible. We also use this data to support the evolving relationships with our customers.”

For example, “we look at the city pairs that are being searched to help us determine what type of service we should have on a specific route,” Landson said.

The payoff? “Our customer and loyalty segments grow year-over-year,” Landson said. “We believe that intelligent, data-based targeting has a lot to do with that growth.”

‘$1 million per week’

The benefits of a data-focused approach may be easy to understand, but execution is another matter entirely. For most airlines, the first problem lies in “bringing together all sorts of disparate silos of passenger information—booking information from transaction systems, web and mobile behavior (including searches, visits, abandoned carts), email data, customer service info, etc.—to create a single, consolidated view of the customer,” said Allyson Pelletier, vice president of marketing with Boxever, which offers a marketing platform focused on putting big data to work for the travel industry.

“Armed with this information, and the resulting insights, they can then take specific action that helps them convert more visitors on-site, secure more revenue, or increase loyalty across any channel,” Pelletier said.

At Norwegian airline Wideroe, for example, a single customer view “enables agents in the call center to understand the full history of the customer—not just the customer service history, but also their recent visits to the website or promotional emails they’ve opened,” she explained.  “After they solve the customer service issue at hand, they’re in a powerful position to then recommend the most appropriate ancillary service—driving add-on revenue—or offer a complimentary upgrade, thereby driving loyalty.”

Insights garnered from a single customer view can also drive personalized messaging into various communications channels, and email is a popular starting place, Pelletier noted.

“One of our largest clients in Europe uses Boxever to understand abandoned carts and then trigger personalized emails to the abandoners,” she said. “They reported back subsequent bookings of $1 million per week from these communications.”

Boxever also cites a 21 percent reduction in customer-acquisition costs on paid media “by understanding who the customer was, where they came from and whether or not they were already a customer,” said Dave O’Flanagan, the company’s chief executive. “This way they could start to move those customers away from expensive acquisition channels to retention channels, like email, which is much cheaper.” There is also potential for a 17 percent uplift in conversion on ancillary cross-sells, such as adding hotel or car to a booking, he added.

‘Few companies are really leveraging big data’

Exciting though those benefits may be, there’s an even bigger pool of potential payoffs remaining untouched. “Surprisingly few [airline] companies are really leveraging big data today,” O’Flanagan said.

Indeed, “I’ve not seen a single major airline with an integrated ‘big data’ business solution, nor an airline with a plan to integrate such a program,” said Richard Eastman, founder and president of The Eastman Group, which builds travel software.

That depends on how one defines big data, however. “The airlines will tell you they ‘have it all’ without really knowing or understanding what ‘big data’ really is,” Eastman said. “Airline managements remain so focused on selling seats with their existing inventory systems that they have ignored buyer information needs as well as the tools that would enable them to reach out to buyers and travelers to serve those needs—let alone, reach buyers at decision-making moments.”

Marketing, flight operations and crew operations are all areas of rich opportunity, O’Flanagan said.

“I think there’s still a huge unmet need in the marketing and customer experience area,” he said. “Companies like Google are trying to be the ultimate assistant with technologies like Google Now. I think there’s a huge opportunity for airlines to create a helpful travel assistant that knows what I need before I do by combining data with mobile—helping people through airports, in-destination, right throughout the whole travel journey.

“Imagine a travel application that knows where I am, that I’m traveling with my family and that the weather is bad on our beach holiday. It could start to offer alternative itineraries close by that are family-friendly and not weather-dependent. These are truly valuable things an airline could do for me if they could use big data effectively and join the dots between me, my travel experience and environmental factors affecting that.

Originally posted via “For the airline industry, big data is cleared for take-off”

Source: For the airline industry, big data is cleared for take-off by analyticsweekpick

en you’re a massive international news organization, revamping your flagship platform is only part of the story. Adapting and consolidating operations to stay relevant in an ever-changing media landscape is the order of the day.

A key part to that story is obviously understanding what it is your readers like, what they actually read and what they really should know about. The Guardian, to help it achieve those goals, has built its own in-house analytics engine called Ophan. It’s a bit like Chartbeat or Parsely, or many other analytics platforms at its core.

Behind the scenes, it publishes about a quarter of a billion events per day and typically the lag before something shows up on the dashboard is somewhere between three to five seconds. But it hasn’t always been such a powerhouse.

The project grew out of a hack day, but the team couldn’t bring themselves to turn it off at the end of the weekend. So they just left it on, running in the background. Three months later, The Guardian’s director of architecture, Graham Tackley, decided to devote some real time and attention to what it could achieve. Tackley and Chris Moran, The Guardian’s digital audience editor, quickly saw that by measuring site data more closely, the organization as a whole could potentially benefit.

“It’s like understanding how journalism works digitally […] the ideal is that we should all understand that. And that eventually kind of codified into the idea of democratizing that data within The Guardian. All of those sub-editors that were working without guidance, essentially. We wanted them to be able to see it for themselves,” Moran said.

When it was ready, they opened it up to the whole company, to anyone that could potentially benefit from additional data insight. Since switching it on, Ophan has gone from zero to more than 950 monthly active users within the organization, Moran exlained.

The key thing about that organic growth [is that] you don’t get 950 people [to use a tool like Ophan] by buying in a tool and going through some massive training process,” Moran said. “The way you get that is by making the tool useful to the individuals within this building, so everything it has grown into has been as a direct result of within editorial and beyond – but very much focused around editorial and people saying ‘I’ve got this problem, could Ophan help with this?’”

The answer to ‘why do we need Ophan?’ isn’t ‘because data,’ it’s because it can help us do our editorial jobs, even if that’s just as simple as showing some section editors that they might be wasting their time micro-managing the front page and should instead be focusing on an article level on how to move people along to other pieces of content. That in itself helps them prioritize their time.

A frequently cited concern by editorial when talking about data is the fear that if you only pander to readers and a wider audience, you end up with content that falls somewhere between asinine and anodyne. However, perhaps unsurprisingly, Moran disagrees.

Page views are much maligned, people say why use page views because it just leads to clickbait – but actually, if what you’re trying to do is judge how well you’ve promoted a piece of content, it’s really effective. It’s not necessarily a sign of quality – as long as we all understand that, that’s fine – but it can tell you whether something is working.

The thought here is that if ‘page views’ is a useful metric in its own right, it somewhat removes it from the advertising process – clicks for feedback, rather than to drive revenue. Moran explains that while there’s no commercial aspect to his job at all. “We basically work on the principal that if we get our excellent journalism in front of the widest possible audience for each piece, that’s probably going to have a good commercial benefit – and surely that’s also an editorial aim?”

Beyond basics

Ophan isn’t just tracking page views. It’s also looking at things like median attention time, using the same sort of methodology as Upworthy or Chartbeat, to find out where users are engaging with each page. This goes some way to arguing against pure page view measurements, as the team can see whether they’re just clicks or whether people are actually engaging with the article. And what does ‘engagement’ consist of here?

“The way it works is that there has to be evidence you’re actively doing something on the page – it has to be in the foreground tab, and you have to be moving the mouse or scrolling, or clicking, or doing something like that,” Tackley said. “Every time you do that, the timer starts for five seconds.”

The median attention time detail of Ophan is split in terms of dispersal. It’s in 10-second chunks, and visitors are color-coded depending on how long they stayed on a page. People who left before 10 seconds are widely classed as ‘bounce’. This data can give an indication of how any particular piece of content is resonating with the audience, but different formats perform differently. A live blog behaves very differently than a regular article because people are refreshing it, so the attention time tends to get dragged down. They also tend to be pretty long in comparison to regular written content.

By looking at these figures, section heads, editors and journalists can see, broadly speaking, how long any article should take to read, how many people went beyond that and what the median attention time was.

What this doesn’t mean, however, is that there’s a magic formula for working out optimum article length, Moran says.

One of the reasons I worry about attention time as a metric around editorial content – particularly as an indicator of quality – because that’s a really tempting thing to do, to go ‘right, if we aggregate this data, we can find out the peak length of an article,’ but I just think that’s nuts. If people worry about page views as a metric leading to clickbait, attention time leading to you curtailing or lengthening your articles, I think is much, much more dangerous when you talk about editorial content.

There’s obviously a lot of buzz around attention time and various other things – at a macro level, it makes sense that if you have a lot of quality journalism, you’d expect people to spend more [time] on your site. But it’s also a very slippery metric…we’re learning what you can’t do at an article level is assume that time spent on page is a clear indication of quality, because it’s affected by so many other things.

The ‘Russell Brand’ effect

For anyone that works in publishing and has any understanding of the internet and social, it’s going to come as little surprise to hear that just dumping a URL onto the internet and hoping people find it isn’t a very good strategy. It needs promotion, and much of that is via social channels like Twitter and Facebook, as well as Google Search.

As such, Ophan has a whole bunch of icons and shortcuts indicating exactly where any one piece of content has already been promoted. If something hasn’t been, journalists and editors are encouraged to hassle Moran and his team.

“You can’t actually identify by subject what does badly, broadly speaking,” he explained. “The one thing those pieces that do badly have in common is that we haven’t promoted them. When you’re doing 500 pieces a day, it’s really easy for bits to slip through the cracks. If you’re the journalist who’s interested in it, or the sub-editor, or the editor, you can immediately get a sense of whether or not we’ve pushed this piece of content.”

One thing that doesn’t work, however, is teasers on social media. Moran has a serious distaste for social media ‘gurus’ who advise the use of them.

“It’s absolute horse shit. Every single time you use a teaser on social media is that you might get high click through, but it’s very unlikely as people don’t have the time, but what you definitely get is immediate bounce,” he said.

“The problem is people see Russell Brand going ‘I done this’ or ‘this is interesting’, and they all go ‘it got 500,000 retweets’ and everyone gets very excited, but Russell himself doesn’t know how many people bounced off that fucking article. Also Russell Brand is Russell Brand, he’s not a newspaper,” he added.

Exploratory data

Ophan provides the team with a level of granularity that turns it from analytics reporting into an exploratory tool. A nice example of this is a piece around the Nigerian elections it ran a couple of weeks ago. By looking at the data, the team could see that 35 percent of the total views came from within Nigeria, which is validating for the journalist and editor working on the piece.

More than that though, it also threw up an interesting data set from an unknown device type – it turned out to be a mobile browser that peculiarly specific to Nigeria.

“Fundamentally, if we’re looking at expanding into other territories, that kind of information is really, really useful to us,” Moran said.

That kind of data exploration helps the team understand the relationship between all its different channels too, including social and search.

“Unknown [traffic] really interests me because there’s a lot of rubbish about it on Twitter. People, really, really intelligent, well-qualified, respected people talking only about dark social – because it’s a wacky term,” Moran said. “The simple fact is: social is clearly important in it and Facebook definitely, a bit… but search is a huge part of this, and anyone who’s not talking about search when they’re talking about unknown doesn’t know what they’re talking about.”

An assumption too far

While based around data and analysis, Ophan is essentially designed to be an insight machine, in the right hands. Perhaps controversially though, this means that Moran thinks that it should be used to inform editorial judgement and what gets produced. Obviously, some people worry that this could lead to impaired judgement.

The interesting thing about that is that it assumes a couple of things. The first is that operating in total ignorance is good and will always lead to the right decision; I know personally just from watching the data versus that process that this is very much not the case.

Secondly, it assumes that the data is in control of you, you’re not in control of the data. The point of this is, and the reason it’s not automating into all of our various processes, is that we want human beings between the data and editorial. We want to be data-informed, not data-led. There are times when the data will tell us something, or confirm something we might already know and we might very well ignore it.

He continues, arguing that being informed by the data can actually lead to higher quality journalism. Writing about Justin Bieber isn’t the recipe for success you might think it to be.

One of the reasons behind that, if you look at the data, is that you’re probably talking about a lot of Google traffic, if you’re thinking about scale, and Justin Bieber is one of the most competitive search terms in the world, with a lot of places like TMZ who have much, much bigger domain authority around it. So it’s quite useful just to burst myths like that.

“We’ve consistently shown on Facebook that our core journalism around social ills, politics and everything else works really nicely on Facebook, on a mass platform,” Ophan’s architect Tackley, added. “That’s real journalism, real articles, not listicles. If we just trusted those lazy myths, we’d never try pushing that kind of content out.”

Part of the battle seems to be to understand what the data is actually telling you; there are a whole lot of different metrics you could look at, but if you don’t fully understand what you’re seeing, it’s easy to miss the real implications.

For example, page views per visit are often measured to give an indication of loyalty, but if you improve your article page so that you link better, your page views per visit should rightly go down; people find things quicker because they don’t have to navigate to a different section to find something else to read.

The future of news?

Loyalty is something pretty high on the agenda for Tackley and Moran. Right now, Ophan’s not really very good at measuring it.

“There’s some vague approximation to it in there, but we need to do a better job of understanding whether this is someone who’s coming back to this article or series every week, or is it someone who’s just coming in for the first time etc. Having some indication of that in an understandable way that you can action [would have value]. I don’t know how to do that, but we’ll figure it out,” Tackley said.

One of the other big things that the pair are excited about is tracking video, both on-site and off-site.

“The offsite is really critical for the future,” Moran said. “We know that natively we can embed a video in Facebook and the reach will be huge compared to, say, on-site. Everybody knows this, but how do you compare that success to something like a piece that gets 10 times fewer plays on-site but that carries a pre-run, for example? That’s really difficult.”

With humble beginnings at a hack day, and now 950 monthly users within the organization and the power to inform the international news operation, Ophan has clearly demonstrated that the future of The Guardian is a data-driven one.

Data-informed. I mean, data-informed.

With that comes an editorial responsibility to resist the temptation to cover topics in ever decreasing circles. Indeed, if Tackley and Moran are correct, Ophan’s value comes in allowing exploratory promotion of content, as well as the potential to experiment with the editorial agenda.

Originally posted via “How The Guardian’s Ophan analytics engine helps editors make better decisions”

Source

There’s no denying it: the OEM analytics market is huge. A trend now recognized by Gartner in their 2018 Critical Capabilities for Analytics and Business Intelligence Platforms report, more and more organizations see the value in providing their customers with analytics.

But what’s the best way to architect your OEM offering? This is where the fun begins.

But First! White Labeling vs. Embedding

Here at Sisense, there are two relevant use cases for OEM platforms: white labeling or embedding. Often mistaken for one another, there are some differences between the two.

In the white-labeled use-case, a Sisense server is completely rebranded so that the Sisense logo and name is replaced, the color palette is changed to the match the organization’s application, system emails are customized by the system to match the brand and more. The tenants (our customer’s customers) access the Sisense server directly, and experience Sisense analytics as provided by the customer. They are not necessarily aware that the analytics server they are accessing is actually a Sisense server.

In the embedded use-case, our customers have their own web application and embed Sisense within it. We support embedding the full Sisense application, including the data modeling, analytics and administration areas, or embedding specific dashboards and widgets using IFrames. Widgets can also be embedded within web pages using the SisenseJS infrastructure.

Three Flexible OEM Architectures

Option 1: Shared ElastiCubes with row-based data security

The first type of architecture for OEM deployments utilizes shared Sisense servers for multiple tenants and shared ElastiCubes and dashboards. Segregation between tenants is achieved by using row-based data security within ElastiCubes. All of the customers’ data resides in a shared ElastiCube, but each of the tenants gets access only to their own data.

• Advantages: Lower hardware costs, high resource utilization, simple asset change management
• Disadvantages: Tenant resource usage may affect other tenants
• Typically Best for: Tenants with identical data models and dashboard requirements

Option 2: Dedicated ElastiCube per tenant

The second architecture for OEM deployments utilizes shared Sisense servers for multiple tenants, together with providing a dedicated cube and dashboards for each tenant. In this deployment, multiple tenants use the same server.

Looking for an even deeper understanding of OEM architectures? Read an in-depth look at each of the three options described above in our whitepaper.

Typically, the OEM has default ElastiCubes and dashboards and creates a dedicated copy of them for each of the tenants. The ElastiCubes and dashboards can be identical copies for each of the tenants or customized per tenant. The OEM uses access control for ElastiCubes and dashboards to ensure each of the tenants only has access to their own data. Typically a user group is created for each of the tenants. All of the tenant’s users are assigned to the same group. The relevant ElastiCubes and Dashboards are shared with the tenant’s group. In this way, the asset access control layer ensures that users of each tenant only gets access to their own data.

With this solution, you need to consider how the system scales to support your future needs and support additional tenants. While initially, you’ll enjoy shared server resources, as you add more tenants you may have to provision additional servers, increasing the hardware costs of this solution.

• Advantages: Low hardware costs for a small number of tenants, high resource utilization
• Disadvantages: High hardware costs for a large number of tenants, more complicated asset change management, tenant resource usage may affect other tenants
• Typically Best for: Tenants who require customized data models

Option 3: Dedicated Sisense server per tenant

The third architecture for OEM deployment is to provide a dedicated Sisense server for each tenant. Typically an OEM customer will have a server image including default ElastiCubes and dashboards. Each of the customers receives their own instance of the server. The ElastiCubes and dashboards can be identical between the servers or customized for each of the tenants. The data for each tenant is completely separate as each server has its own assets, including configuration, users, ElastiCubes, and dashboards.

• Advantages: Highest level of security, dedicated resources per tenant
• Disadvantages: Low resource utilization, higher hardware costs, complicated asset change management
• Typically Best for: Tenants with strict security regulations, such as financial or healthcare institutes and tenants that need a high level of schema and dashboard customizations

What’s The Right Choice?

The type of architecture suitable for a specific customer depends on the use-case, the needs of the customer, the resources that can be dedicated to the deployment (both allocated people, and allocated hardware), and preferences. Of course, in order to make the right decision for you and your business, it’s always best to review all options in relation to your current and ongoing needs.

Looking for an even deeper understanding of OEM architectures? Read an in-depth look at each of the three options described above in our whitepaper.

Source: OEM Architecture: What’s Right For Your Business? by analyticsweek

Like every other feature in your application, the world of embedded analytics is not static. The outer bounds of capabilities customers want are constantly evolving. At the same time, older capabilities like data visualizations—which customers once considered modern and innovative—are now table stakes.

Your end users will always have new requests (and complaints) about your application’s embedded analytics. It’s inevitable. And as long as everything’s working as it should— you’re keeping bugs in check, your app is reliable—most complaints are likely new feature requests in disguise.

>> Related: 5 Early Indicators Your Analytics Will Fail <<

Unfortunately, translating those requests into actual analytics features can be difficult. What do your users really want from their dashboards and reports? Decoding these complaints means adding valuable new features to your roadmap, and avoiding a panicked scramble to add them before it’s too late and your customers start churning.

Use this chart to translate common end user requests (on the left) into the analytics features your users really want (on the right):

To learn more, get our Blueprint for Modern Analytics >

Source: Which New Features Do Users Want? Decoding Customer Requests

Insurers have a lot of customer data across different systems peppered throughout their enterprise. Customer data would typically live in multiple systems like CRM, Billing, Policy Administration, and so on. This approach however suffers from multiple challenges:

 

1. Duplicate data across multiple systems
2. Multiple Versions of the same data point
3. No Single source of truth
4. No correlation between cause and action
5. Completely under utilized customer interaction data

 

Imagine a structure so flexible and scalable that, that it could bring all your data sources together, irrespective of the data formats, tie in with the customer key result areas (KRAs) and at the same time deliver predictable insights at the point of decision. In real time.

 

Enter single customer view. Or as we call it – Customer OneView.

 

OneView in Action

 

Customer OneView is part of Aureus data analytics platform called CRUX. OneView is not anything like a CRM. While a CRM would show only static information, OneView delivers intelligent, usable and real time insights that can be put to use immediately.

OneView can integrate with (atleast) four broad event data streams:

1. Customer
2. Relationship
3. Transactions
4. Interaction

 

Nitin had written about stream based data integration on his insightful post titled “Cheers to Stream Based Integration“

 

These data streams could originate across multiple data systems – Policy Admin, CRM, Billing, etc.. Between them, these four cover some of the most critical customer data, that often lies under utilized. OneView not only brings these data streams together, but it also helps build a comprehensive customer life journey showing important milestones, critical customer interactions, sentiment at each interaction or transaction level as well as at a relationship level.  While OneView is a powerful insights delivery framework, it also helps to deliver the output of predictive analytics models in a form that is usable by the business users. OneView can help translate the output of the analytical models into usable insights. Imagine a customer sales representative talking to a customer, or a field sales agent going to meet a customer. OneView will give them unambiguous insight into the customers history, sentiment and even potential  action to take, without burdening them with the Hows and Whys.

 

Imagine a typical customer cross sell scenario. Most organizations tend to throw (figuratively speaking) the entire product catalog at the customer without any consideration for their lifestage needs, portfolio, demographics etc… Not only is this a highly ineffective cross sell approach, but it is a terrible customer experience approach. With OneView the customer service representative or the field service agents knows exactly what the customers latest and overall sentiment is, what her product portfolio looks like and which product the customer is most likely to buy.

 

The end goal of any activity is to make the end customers experience epic. By knowing how a customer is likely to behave, modeled on her previous behavior, insurance companies can ensure that the customer experience is always moving to the right.

 

Source: A Single Customer View : The Secret Weapon Everyone Must Use by analyticsweek