How to go ‘all in’, with a Data Science or Analytics bet for your business
How big a bet are you making on Data Science, are you going ‘all in’?
With any event as rich in content at Data Insight Leader Summit, its useful to also hear others ‘takeaways’.
So, to supplement my own insights from this conference, I’m delighted to welcome new guest blogger, Hanne Sorteberg.
Hanne is BI Manager for SpareBank1 Forsikring, an alliance of savings banks in Norway (together Norway’s 2nd biggest bank).
I’m grateful to her for recording so many useful reflections on lessons highlighted by this event.They should also be a useful guide to anyone new to Data Science or Analytics to improve their business. Over to Hanne to provide a useful overview, of using analytics, so you don’t have to go ‘all in’ without knowing the risks…
How to make Data Science (really) work in your business
The last 5 years have seen an increasing investment in advanced analytics/data science. The last year has had a substantial increase in analyst roles and hires. Data Science has an increasingly important role in new and existing business models. Many companies have moved from “pilot mode” to “production mode”. In doing so, they have encountered some real problems and issues, and uncovered some real opportunities.
This blog post is inspired by my participation in Data Insight Leaders Summit 2017. The article is a collection of messages from speakers, discussions with other participants, many research sources & my personal views. I have added some named citations from memory. It would be difficult if not impossible to provide all the proper citations and references, so ask for forgiveness for not providing all of them.
Here is a collection of success factors and pitfalls, companies and data insight leaders have faced, going “all in”:
Personalize – Finally realizing the segment of one
In the early days, it was meaningful to define an “online user” or “digital” segment. Not so anymore, since practically everyone is in that segment. Digital has evolved from shouting a message to a stadium filled with customers with a megaphone (posting on your website). You now need to listen to each individual, to discuss their needs and preferences (personalized messages and products).
“The battle for personalized loyalty go to the one with the most data, and the ability to gather and use it.“ Martin Squires, Walgreens Boots Alliance
Likewise, segmentation criteria have moved on. From obvious measures (like age, gender and location), to actionable needs, preferences, context, status and timing.
There’s an increasing amount of data, deeper customer behavioral insight and new data sources (inc. open and public data). This makes it possible to create very fine-grained segments. Advanced analytical platforms allow the utilisation of these segments, to be more relevant to each customer.
If you have succeeded, in dividing your customer base into microsegments, can you handle it? Do you have the product mix and breadth of content to match? Imagine every customer walking into a store being offered a unique product mix. Should you offer customers, buying the same product, a different price based on their data, behavior and/or profitability?
A note of warning – If you assume you know your customer and act upon that, you had better get it right. Annoying your customer with too many or non-relevant messages is still not a good idea.
The value of data, Me™
Data is an increasingly valuable asset. Using data, to lower costs and increase revenue in your existing business model, is old news. Packaging data to create new products and services can represent new revenue streams. For example, selling customer data your business has collected to other businesses, for their analytical use. Providing services and advice back to your customers, based on their data, can create higher loyalty and even monetary gains.
“We are looking at the opportunity to sell information back to our restaurant owners, to enable them to open new restaurants and refine their menus. Selling data requires packaging and the productification of data.” Rufus Weston, Head of Insight, Just Eat
With increased focus on data protection and regulating data portability, an individual can be seen as a data bank. Aware of their personal information worth, they use it to negotiate with their suppliers and business partners. Every person is a potential data product, Me™.
Value proposition – from cost centre to profit centre
Traditionally, BI and reporting have been regarded as cost centres, typically organised within IT or Finance. Analytics is now used to enhance business processes, and creates real value. There is a shift from thinking about data as a costly necessity, to data as a driver for growth and increased profitability.
Data Science centres are increasingly set up as value-creation units. Having profit & loss statements, and rigorous KPI reporting aligned with business strategy.
“Data Science is no longer a Science fair project” Olivier Van Parys, Head of Advanced Analytics, Bank of Ireland
A recurring issue, for many Data Science leaders, is getting buy-in. This needs to be anchored not only at the C-level, but also with a range of stakeholders throughout a business.
Product owners, Distribution Channel owners, Marketing, IT, Compliance.. they all need to understand, commit and contribute to make Data Science happen.
Here are some tips to get there:
Fix basics first. It is hard to obtain trust, in moving on to Data Science, if your BI dashboards are not performing. Gain trust by providing robust descriptive analytics, before moving onto predictive and prescriptive.
Educate. Besides training your technical staff, run a “Machine Learning for Dummies” or “Management analytics” program, to get everyone on board. Sometimes the models used are a “black box”. Both business stakeholders and customers need a better explanation. To understand why they were recommended a specific product, or given a certain offer. Not, “because the robot said so”. The power of machine learning methods, finding new and unexpected results with your data, must sometimes be balanced with the need to provide accountable and documented calculations.
Prove your point with pilots. Running pilots, with low-cost and low risk, is a way to show what analytical models can do. The pilot should solve a real business problem, and be implemented in a real business process.
Balance quick wins with long-term direction. Pilots are useful to communicate and get buy-in, from business stakeholders, and to prove whether a concept is promising or not. However, the promising pilots should evolve to long-term, scalable analytics solutions, with substantial business impact.
Establishing an agile, continuous model development cycle
A model development cycle typically involves the following phases:
- Establishing business need. What business problem is Data Science trying to solve? Summon the Business owner, Data Scientist, other main stakeholders, and preferably, the Data Protection Officer (DPO). Together, formulate a feasible business requirement/problem statement. Get business to sign-off and prioritize work on this opportunity.
- Proof of concept. Do I have the data I need to develop a successful model? Are there technical issues or other showstoppers? Unite Model Developers, Data Preppers and Platform Specialists. Develop together, a first shot at the model, and visualize the preliminary results.
- Establishing business value. Present a working model to your business. Define use cases for the model, and a Business Case for the model development effort.
- Develop and test. Develop, train and evaluate the model in iterative steps. Most models should be A/B tested. Have business verify the results.
- Distribution and roll out. Educate and anchor with stakeholders. Incorporate the model results in your business processes.
- Learn. Create a feed-back loop, to capture results and developments, from your deployed models.
- Model Portfolio Management. The performance of models will change over time. For some models, it is necessary to say – “You’re fired!”. Monitor model performance, towards predefined KPI’s, fire and hire models as needed.
Any of those phases may uncover show-stopping issues or fail to prove business value. If that happens, development should be halted, in favor of the next analytics opportunity.
Filling the data lake – and collecting the gold
Most Data Science models require much data to be meaningful and successful. A LOT of data. New data sources like video, images, sensors and social media generate loads of data. However, often you require data from sources that are scarcer. For example, transactional data, process data and documents.
How can you obtain a critical mass of data, when the starting point is too meagre? Here are several options to try:
- Develop a strategy for data collection. Changing your user/customer interactions to collect more data. Collaborating with third parties, or other actors in your field, to obtain a broader set of relevant data.
- Create more data by sampling existing data, or data generation. Automate data preparation and processing.
- Create model satellites. Define some small outlier segments, where you test out model parameters, to evaluate results, that are fed back into your main model.
- Free your data. Make data accessible to a wide audience. You may have to challenge and co-operate with IT security and your Data Protection Officer. But, seek to get the access-control and masking mechanisms in place, to offer all the insight you can. Challenge obstacles and formalize data provision workflows and approval processes.
- Make the data understandable. Do you have a clear definition of your KPIs, such as “active customer”? Maybe, you have 20 definitions for “active customer”? Making the business agree upon data definitions, and establishing data governance, is a valuable investment. Describe the data in a way business understands, and make the documentation easily accessible.
- Visualise data. Support your organization to be increasingly self-serviced, for information and data insights.
Privacy – GDPR Confusion
Collecting and exploiting data are often in conflict with protecting our customer and user’s privacy rights. The new regulations, mainly GDPR, can be unclear. Businesses are awaiting clarification and concretization. When is user/customer consent required, and when can you argue that the data collection and use, is acting in your customer interest?
It seems that the best we can do, when in doubt, is to make a qualified and thorough evaluation of what data you collect. Who has access to it and why? Document that. Regulators are happier with a decision that is erroneous, than one that has not been subject to any evaluation.
“This video, explaining why we collect viewer’s data, was even more successful internally. Making our stakeholders, and the organization as a whole, conscious of privacy issues.” Sanjeevan Bala, Head of Data Planning and Analytics, Channel 4
Have you considered not buying an iPhone, or downloading an app, because you disagree to something in the mile-long privacy agreement? I thought not. Some argue that GDPR is too late. The big, non-European players have been collecting data for a long time, and have all they need to pinpoint our behaviour and preferences. Is it even possible to regulate and control user data, from being collected and used for good or not-so-good purposes, on a global scale? What happens to our competitiveness if the regulatory landscape is skewed?
Talent – The hunt for unicorns
Everybody is desperately looking to hire Data Scientists. Those unicorns, who can translate Business needs into machine learning models, and roll them out to your enhance your business processes.
One person seldom has all the personal abilities, time and dedication to cover business and technical, operational and strategic tasks. Data Science is best covered by a team that complement each other. There is an onset of new roles and names to cover the different functions in this team, but you may need the following (one person can fill several roles):
- Evangelist – creates enthusiasm in your business for Data Science, communicates success stories and results.
- Strategist – sees the whole picture, and has long term plans for Data Science, in your business.
- Interpreter – translates business objectives into defined questions. Which can be answered, by the use of advanced analytics/machine learning.
- Data provider – collects, cleans and manages the data needed to feed the models.
- Model developer – models, test and evaluates the analyses and models.
- Integrator – creates APIs to access the models. Integrates results into business processes and collects resulting data.
- Distributor – rolls out the use of models, into the organisation.
- Janitor – maintains the models, and ensures their performance over time.
Look for talent in unexpected places – from other industries or professions. Hire for attitude – train for skills. Moreover, most importantly, retain the talent you have secured. Provide professional growth, the right autonomy and a strong, exciting purpose.
Organisation for Analytics and Data Science – centralised or de-centralised?
Once you have your Data Scientist, analysts, strategists, evangelists, engineers, dev-ops and programmers all lined up – where do they go? Do they form a core team within IT, Finance, Marketing, Operations? Or, do you distribute them across your business?
The centralisation vs. de-centralisation discussion is still ongoing. Going back and forth like a pendulum, also within organizations, over time.
Centralising your team ensures standards and one way of doing things, and a strong professional environment. However, the need to be close to business and value creation, is regarded as increasingly important. A spoke and hub model is proving popular. With data science teams in each business area, reporting to a home-base for standards, lessons learned, re-use and professional input. This may cater to both needs.
Centralised or not, the team must have a clear purpose and KPI’s defined by the business. Add some ‘Hackathons’ and create some free time slots to explore the unknown, to boost innovation.
Data in your DNA
The bottom line is to build a data culture for your business. Top down, and bottom up, everybody should, over time, regard data as a strategic asset.
So, will you stick, fold or go ‘all in’?
Many thanks for Hanne, for lots of sensible advice there & a great summary of several key themes from #DILS2017 event.
What about you? Have you bet big on Analytics to transform your business? How is that working out for you?
Having recently heard from real practitioners leaders like Martin & Hanne, it would be great to hear your experience too. Feel free to share below, in Comments box, or on social media. Let me know if you’d like to offer your own guest blog post, for other Customer Insight Leaders.