Computers and artificial intelligence have come along at an exponential rate over the past few decades, from being regarded as oversized adding machines to the point where they have played integral roles in some legitimately creative endeavours.
The explosion of interest in predictive analytics among Australian marketers has been met by similarly-explosive growth in the number of vendors and consultancies vying to meet their needs.
With such rapid growth in demand, how can marketers be assured they are getting the best result for their money? This is an especially pertinent question given the complexity of the skills required to execute a predictive analytics project and the lack of benchmarks for performance or expected returns.
Providing clarity to buyers is a topic of interest for the Institute of Analytics Professionals of Australia (IAPA). Its CEO, Jodie Sangster, is regularly quizzed by marketers on who they should turn to.
“There are very few organisations that offer the service, so there aren’t enough resources to service the need,” she says. “The other issue we have at the moment is the organisations looking for an analytics agency are not in a position to know whether that agency can actually provide them what they want.”
Analytics providers range from large traditional IT suppliers such as IBM and Oracle through to boutique consultancies, with varying approaches, tool set and capabilities. The background and responsibilities of agencies also varies widely.
Sangster says agencies that have come from a background in data analysis may lack a strong understanding of the needs of the marketers that want to engage them.
“So there is a lot of sorting out to do so that the marketing and analytics world come closer together,” she says.
Crystallising your data analytics approach
Director of commercialisation at NICTA’s Broadband and Digital Economy Business Team, Glenn Downey, has had significant experience with analytics agencies through developing the market proposition for NICTA’s own analytics spin-out business, Ambiata.
“It seems some people think that analytics in general is a kitchen sink that you can throw at any problem, and it will solve it,” he says. “I don’t think that is accurate at all. You can’t just throw the kitchen sink at the dart board and expect that it will solve the problem. It will have an impact, but probably not the one that you want.
“You need to fashion that kitchen sink into a dart, and the dartboard into a bullseye, so you can address the specific problem with the right product or service offering.”
That means defining the problem to be addressed before engaging a predictive analytics supplier. Downey says common scenarios for initial analytics engagements include for customer acquisition and retention or next deciding next-best action.
“There are so many things predictive analytics could be applied to in marketing,” he says. “Unless you narrow that down, then the conversation that you have with just about any vendor is going to be a waste of air.”
One of the key debates in selecting an analytics provider is whether a pre-packaged solution can deliver effective results, or whether the complexity of problems demands a bespoke consultation and solution. Downey believes the exploratory approach to selecting a supplier generally leads to a consultative service.
“They will have a broader scope about what predictive analytics or marketing analytics can be and how it fits with an organisation’s maturity,” he says. “If you just leave it up to a specific vendor, they are going to try to shoehorn their solution in, whether you want it or not.”
Numerous analytics consultancies have sprung up in recent years, including Melbourne-based Predictive Analytics Group. Managing director, Dr Theo Gazos, echoes Downey’s position that a client needs to think clearly about their underlying problem before engaging with an agency, as each problem and the data used to solve it will be unique to each company.
“You cannot buy these algorithms off the shelf, it will not work - the algorithms need to be specific to you,” Gazos says. “Whatever way you try to slice and dice it, you need an expert mathematician or statistician to look at your data and then observe what is going on using mathematical models.
“If you are going to go to one of these big guys who produce generic models, and you don’t undertake that further customisation required, then it is all for naught. Because when you buy the generic model, you can be certain your competitor has the same thing anyway, and that the confidence intervals associated with those forecasts are going to be so large that commercially the upper and lower bound will be huge.”
Gazos believes there is confusion in the market today about whether an analytics capability is based in technology or people. He comes down on the side of the human capability.
“Who are the people that are going to be trawling through your data and how are they going to do it?” he asks. “That analytics team that builds the model for you will become part of your team for that period of time.
“And you need that associate professor who has worked at a research institute and published papers and has an encyclopaedic knowledge that can develop that algorithm for you.”
While this method may appear more expensive, Gazos says it is generally cheaper than spending millions on an update to enterprise systems and databases.
Questions to ask
Sydney-based consultancy and technology developer, Zetaris, is another supplier that emphasises understanding client needs to ensure that the analytics activities align with overall business strategy. Chief executive, Vinay Samuel, says from a buyer’s perspective, it is important to understand an analytics capability is much more than just an understanding of Hadoop and the ability to provide a visual representation of data.
There are a number of questions that need to be asked before a client engages an agency, particularly to ensure the agency is capable of not just providing the service a client needs today, but also the services it might need a year from now.
“To what extent have they got the full predictive capabilities and the statistical capabilities?” Samuel asks. “Can they do the pattern recognition hidden in lots of structured and unstructured data? Can they join videos, pictures and tweets to transactions?”
Samuel also cautions the need for buyers to truly understand who will be delivering that capability and cost of handling different workloads, as well as the extent to which the supplier will train the client to eventually take on some of the analytics capability itself.
But according to IBM’s regional vice-president for analytics Brock Douglas, nothing beats experience. That means investigating a supplier’s case studies to understand how its analytics engine has solved real world problems for clients.
While IBM offers a number of generic models for predictive analytics around common topics such as customer churn, increasing basket size and sales conversion, the company has also built industry-specific models.
“The way you get to industry models is by working with clients around the world on specific industry use cases and specific industry problems, and then you productise those models,” Douglas says. “You would want to look for evidence that prove that these models haven’t been just made up in a lab and have been built up by peer industry players around the globe and then productised. The higher return on investment is going to come from those industry models.”
Douglas says vendor selection needs to include the ability to integrate the analytics capability with existing data sets and enterprise systems such as CRM and enterprise and operational systems. Other selection criteria include ease of use, flexibility, and the commercial model through which the service will be purchased.
“You don’t want to be calling your IT team all the time to run a campaign,” Douglas says. “You want to be agile, you want to be quick to market. How many people can actually help you implement that, and how many people can help you maintain it and do that at a commercially competitive price?”
To help get customers over the line, IBM has developed a model whereby it will take a data set from a client and provide back an insight within four to six weeks.
“And frankly if I don’t give you insight, then you shouldn’t buy our stuff,” Douglas says.
Ultimately, Sangster says the decision on which vendor or agency to select will be significantly enhanced by having someone on the inside that knows what to ask for.
“If you are working with an analyst company, you should have somebody internally who can be your barometer,” she says. “You need some degree of internal expertise to be working alongside an agency to be getting the maximum out of an agency. And that can help set the agenda to make sure everybody is on the same page and your agency is delivering to what you need.”
But the relative immaturity of data science and analytics as a profession can also make it hard for would-be employers to understand the skills they are hiring. To this end, Sangster says IAPA is developing an accreditation program for analytics professionals, with the goal of having a full program in place in early 2016.
Follow CMO on Twitter: @CMOAustralia, take part in the CMO Australia conversation on LinkedIn: CMO Australia, join us on Facebook: https://www.facebook.com/CMOAustralia, or check us out on Google+: google.com/+CmoAu