Relying on technology to make decisions is bad for business. This is according to researchers at Leeds University Business School who claim that people who use computers to make good decisions are being unwittingly led by technology into making bad ones.
They also claim machines gathering the huge volumes of data produced every day assume people act logically. Yet most software has been developed without a full understanding of how we actually think and behave.
The university’s report ‘Can Computers Overcome Limitations in Human Decision Making?’ says the result is a lack of confidence in trusting a hunch or a gut feeling because we rely too heavily on what computers tell us.
Few would argue that when it comes to customer profiling to improve the targeting of offers or personalisation, and the return on investment from marketing campaigns, data analytics software is extremely valuable. However, unless the best data insights are combined with the best business brains, common sense can be ignored.
Film director Mike Binder’s story in The Wall Street Journal highlights the perils of relying on data to make marketing decisions. In an interview in the newspaper, he recalled how in 2002 he used his TiVo digital video recorder to record his 1999 movie The Sex Monster. The film is about a man whose wife becomes bisexual. Afterwards, his TiVo assumed it should recommend him a stream of gay programming.
Real people must be involved from a quality control perspective
Forrester’s more recent report ‘Use Customer Analytics To Get Personal’ outlines how using analytics for profiling has a positive impact on brands. Yet analyst Srividya Sridharan says that without human intervention consumers can become emotionally disconnected even if an organisation holds extensive information about them.
She gives this example: “A large credit card issuer ran a direct mail campaign in the US promoting luxury cruises during the time of the Italian cruise tragedy,” says Sridharan. “The company had the analytical capability to personalise direct mail but the context of the offer was misplaced. Human intervention in situations like this ensures the output of customer profiling/analytics is contextual and relevant. Real people must be involved from a quality control perspective.”
EasyJet marketing director Peter Duffy agrees that however good the technology might appear to be, consumers do not always behave in the way they are expected to.
“You need forensic data analysis,” he says. “Insight experts are needed to ensure nothing is missed or the wrong conclusions are reached. Incorrect data sets can mean someone receives offers for something they have already bought, for instance.”
Duffy says reliable profiling is crucial because there is a perception among consumers that they will receive relevant communication from brands. EasyJet’s website is translated into 16 languages and analytics mean the relevant hotel and car hire ads appear when someone books a flight.
“Organisations must bring operational data into the marketing department to improve profiling and campaign effectiveness,” he says. “If someone books a flight, a few days before take-off we send an email reminding them to complete their passport information online. This is operational data but it has to come into the marketing environment for an email to be sent.”
At Telegraph Media Group (TMG), marketing director Graham Horner obsesses about data capture. Every transaction with a Telegraph-branded business – such as its Travel Shop or Fantasy Football – is fed back into the system to generate a single customer view, he says.
He is adamant that others in his marketing team also understand the mass of data being generated and realise its value in driving the business forward. “Analytics packages can churn out profiles but it is how you present that insight and what you do with it that matters,” he says. “You need an analyst who can look at what the data is saying and query any anomalies. To do this, they must be commercially aware and understand key business drivers.”
When TMG profiled print subscribers who had activated their free iPad access, the analytics software presented what it said was a ‘typical’ subscriber. “This was not that insightful, so the marketing team had to manually interrogate the data by extracting it from the analytical package and importing it into programmes such as Excel, where it was assessed by real people without using pre-built models and tools,” says Horner.
Even giants like Amazon get profiling wrong. Its chief technology officer and vice-president Werner Vogels admits it has been guilty of errors when it comes to recommending products based on previous purchasing data. Yet using analytics to learn more about its customers remains central to Amazon’s business model.
Often, he says, problems occur because there is not really enough profiling data on a customer to accurately make recommendations. “We have offered Windows 7 to people who previously bought Japanese steak knives, for example,” he says. “However if a company is not deeply measuring the interaction it has with its customers and using that data to improve interaction, it is missing out.”
Another online shopping retailer Shop Direct, which includes the Very.co.uk and Littlewoods brands, uses machines and humans to gather data through online transactions, telephone research and face-face focus groups.
Shop Direct has annual sales of around £1.7bn and more than 5 million active customers, with 75% of its sales generated online and mobile transactions accounting for 11% of that figure. The company conducts extensive data mining around purchasing and web browsing behaviour.
“Effective personalisation is about more than just data and analytical tools,” says online targeting manager Matthew Doubleday. “Software is a means to an end, but the data is useless without skilled, experienced people to translate it into engaging content.”
He says being aware of fashion trends is an integral part of the brand strategy for Very.co.uk, whose customers value the insight they get from the site’s buyers and style experts. “We would never rely on data alone to devise recommendations and offers,” he says. “We know customers want real advice on replicating looks from the catwalk.”
Software is a means to an end, but the data is useless without skilled, experienced people to translate it into engaging content
Media company IPC also combines the use of hard data with human analysis when identifying how likely subscribers are to renew. Database manager Yemi Okunade says a limited profile may start off as just an email submitted on a newsletter sign-up but it will expand as IPC collects more personal information and insight from surveys, which people are given incentives to enter.
“Today’s analytics software packages save a lot of time and increase the level of complexity in our modelling so we get better at predicting the results of any campaign,” he says.
Nevertheless, Okunade says his team spends hours overseeing the insight the analytics software churns out. “They cast a critical eye and correlate what they find to their real-world experiences. There must be checks on segments and scores to create relevant strategies in line with our business objectives.”
He gives a recent example where an initial data model revealed that the propensity to renew a subscription on several titles was low. However, when his team reviewed the data, they noticed the propensity scores did not relate to what they were seeing in their day jobs.
“By drilling deeper into the data, we discovered that old subscriptions acquired through a specific piece of marketing activity no longer in use had caused a significant skew in the results,” says Okunade. “We removed these records from the dataset and rebuilt the model. This avoided a costly mistake that could have had a negative impact on renewal rates and customer satisfaction.”
In the business-to-business sector, it is arguably even more important that costly profiling mistakes of existing and potential clients are avoided.
Vicky Oakham, head of global marketing of financial services at international information technology services company Atos, says humans must always be involved.
Atos works with The Marketing Practice to ensure the profiling process is thorough. This includes desktop research and qualifying phone calls so messages are targeted and personal.
“We did this with our Power to Perform nurturing programme, which targets key decision makers globally,” says Oakham. “This campaign generated meetings with about 30% of the companies we targeted in multiple countries. We’d never have achieved that without the substantial amount of manual effort.”
She acknowledges that the cost of human intervention is higher in B2B marketing because the volumes of data are lower, especially when it comes to producing high-value direct marketing campaigns aimed at a relatively small segment of hot prospects. “It would damage our ROI and our budgets if we got our profiling wrong,” she says.
With tight budgets, effective profiling should mean better marketing decisions are made, but an over-reliance on technology could prove costly.
Case study: Aviva
Dr Anne Filatotchev, UK marketing director at insurer Aviva, says there must be zero tolerance towards data inaccuracies when it comes to profiling customers.
The company uses profiling to understand how people are likely to behave, what their relationship is with the brand and how their attitudes have changed over time. This helps the company identify what would be an effective, pro-active media campaign to retain them.
She cites an example from the end of 2011 when people and data worked effectively together. The marketing team had identified an increased interest in multi-car insurance so data analytics assessed whether there was actually a trend towards this – and who would most likely buy such a policy from Aviva.
“When we talk to people about our products and their insurance cover, the information we have must be 100% correct,” she says. “Data analytics can tell us what a customer is like and if they are likely to lapse but there must still be human involvement where we can talk about why they might be thinking of surrendering a policy and to discuss different options. This cannot be data-driven.”
She says the marketing team operates to a set of principles to respect the sensitive medical and financial information the company holds. For example, there must be a relevant reason to have a conversation with a customer at a particular time.
“Data identifies different groups of consumers but we must ensure it is being used in the right way,” she says. “The data quality and planning team sit within the marketing function in the UK and this is their responsibility.”
Filatotchev says the data team members must also ensure any correlation between different customers identified by the data analytics is accurate. “There could be a spurious reason behind it. The data picks up on trends but we must take this information back into the business to see if these trends match what real people working every day in this sector are actually seeing happening.”
Ethics of profiling
There is one element of profiling that can be overlooked by marketers. This is the ethical question of how and when to use what can be very personal and sensitive profiling information in a marketing campaign.
Dr Neil McBride is a reader in information technology at De Montfort University in Leicester and he says there are still many grey areas surrounding profiling and privacy.
“Organisations need a code of conduct for how they handle data and the assumptions that are made from any analysis,” he says. “Consumers do not always know what data they are putting out there and a marketing department can often know more about someone’s preferences and buying habits than the person themselves or their family. There is a duty to treat such data with care and respect, and this is where human intervention is critical.
“Individual marketers need empathy for the way a customer will be feeling,” he adds. “They might be buying something sensitive which could cause a family rift, for instance.”
Managing account direct, acxiom
The big data analytics debate must include the role of human intervention. Will a real-time decision engine supersede the role of the consultant analyst? Ultimately it comes down to the power of intelligence. Whether it is 100% automated or driven via a human value judgement, someone needs to justify the decisions.
Yet analytics cannot exist in a vacuum. The emergence of the data scientist is an example of this. This position requires a hybrid skillset/ the rigour of a statistician, the detail of a software engineer combined with the entrepreneurial flair of a business consultant. An organisation with analytics at its heart will have made a considerable investment.
But driving the value from this investment relies on the data scientist using the analytics to create a compelling narrative for executives. There are few marketing directors who will take direction from an analytics report alone.
Data scientists possess curiosity and creativity. These are not characteristics associated with automated analytics. Thoughts such as “What if I get external weather data and link it to retail transactions over time?” will identify obvious patterns but it takes the human intervention to find and load the weather data. Other patterns that can drive incremental revenue are waiting to be found by the curious.
Also, we do not exist in a purely digital environment. The most complex opinion-tracking software cannot track a conversation down the pub, and while we can track high street sales, multi-channel activity and exit survey opinion, it is often the store assistant who clinches the sale with intuitive well-placed flattery. Human intuition must be used to qualify the findings of the software.
Big data exists already within most organisations but understanding the value of the data once it is joined together in a single view is a dream rarely achieved. Organisations need to know the value of all of their assets at their disposal and individual data owners must understand the importance of sharing assets and employ the services of specialist organisations and experts.
To capitalise on the myriad big data opportunities, businesses need an organisation-wide data strategy. Clients frequently rely on Acxiom to navigate the path between the CMO and the CTO to develop the strategy, initiate the build and then galvanise the insight.