‘… let’s seek to understand how the new generation of technology companies are doing what they do, what the broader consequences are for businesses and the economy …’ —Marc Andreessen.54
It’s exciting for me to sit with business leaders to explore ways in which data and analytics can solve their challenges and open new possibilities. From my experience, there are different paths that lead a company to the point where they are ready to take a significant step forward in their use of data and analytics.
Companies that have always operated with a minimal use of data may have been suddenly blindsided by a crisis or may be growing exasperated by:
They end up forced to run damage control in these areas, but are ultimately seeking to improve operations at a fundamental level and lay the groundwork for future growth.
Companies that have been operating with a data-driven mindset may be exploring innovative ways to grow their use of data and analytics. They are looking for new data sources and technologies that will give competitive advantages or are exploring ways to quickly scale up and optimize a proven product by applying advances in parallel computing, artificial intelligence and machine learning.
Regardless of which description best fits your company, the first step you’ll want to take when re-evaluating your use of data and analytics is to form a strong programme team.
Your data initiative programme team should include individuals representing four key areas of expertise:
You’ll need to include someone on the team who has a deep understanding of your corporate strategy. The strategic vision will form the basis for all plans and discussions within the company, including the data strategy. This vision will have been specified by the shareholders, refined by the board of directors, and shaped by the culture within your organization. It will dictate the purpose and principles that underpin how you use data. If your new data strategy is not aligned with the company’s overall strategy, your subsequent efforts will not support parallel initiatives within your organization and will fail from lack of internal support.
There are many excellent frameworks for developing corporate strategies, and I won’t explore those but rather emphasize the importance of synchronizing data and corporate strategies. I’ll illustrate with the framework of Tracey and Wiersema,55 in which the strategic focus of a corporation may be to develop an advantage in exactly one of the following sectors: customer intimacy, product leadership or operational excellence. Within this strategic framework, if your company has chosen to differentiate itself through enjoyable shopping and excellent customer service (e.g. customer intimacy) you’ll choose a data strategy leading to improved customer experience (e.g. an intelligent personalization engine) rather than one focusing on cutting operational costs.
According to a recent global study, business teams lead the adoption of big data more frequently than IT teams.56 Big data is a team sport, and your efforts will likely fail without strong input and support from your non-technical colleagues. If they aren’t convinced of the benefits of your data initiatives, you are probably doing something wrong.
Harvest the intuition of colleagues with the deepest understanding of the customer, the product and the market. Get them involved at the start of your analytics programme and keep them closely involved at each stage of development.
You’ll see many advantages from involving business stakeholders in the initial stages.
Incorporating business intuition is a tricky balancing game, and I’ve seen it go both ways. I’ve seen analysts sitting in isolation come up with models that completely missed key aspects of the business, and I’ve seen business directors make strong statements later proven by the data to be completely wrong.
Intuition sometimes is right and sometimes not, so it’s important to take seriously any non-measurable input from the business, while subsequently seeking data to verify that intuition.
Involve the business experts in the following ways:
Don’t start an analytic initiative without including someone with a strong background in developing and deploying analytic models. I know this sounds obvious, but I’ve seen large companies try to do just that, and it’s painful to watch. You need to include an expert who understands what is possible with data and analytics and can estimate what is needed to attain what is possible. This person should have a strong technical background, several years of analytic leadership experience within a comparable industry setting and a broad understanding of the models, tools and applications available for bringing business value from data and analytics. You can substitute academic experience for industry experience, at a risk.
Your company needs to identify high-potential analytic applications, and it needs to choose from a broad selection of analytic methods and models. In addition to the methods of artificial intelligence and deep learning, which are currently trending, many traditional methods rooted in the fields of statistics, network/graph algorithms, simulations, constrained optimization and data mining have proven their worth over recent decades. Each algorithm has strengths and weaknesses, and you must take care to choose algorithms suited to your data and applications. Considering only a subset of these tools when approaching a business problem will limit your success and can lead to a tremendous waste of effort.
You’ll need to understand what techniques will continue to work well as your problem size grows. Each algorithm will differ in how it scales, so your analytics expert should understand whether a technique that performs well with a small data set is likely to scale well to a much larger data set. In summary, it is important to bring a broad understanding of analytic models to the project team.
You’ll need a broad understanding of programming languages and analytic tools, not only the most recently trending technologies, but also a strong understanding of traditional methods.
The tools you’ll consider range from basic to advanced functionality. At the most foundational level, there are many programming languages that can be used for building solutions from the ground up. These include Java, C++, Python, SAS, R, S-plus and a host of others, some of which are free and some of which are proprietary. Each programming language will have its own strengths and weaknesses, some of the most important considerations are:
You should also be familiar with the development frameworks available for optimizing development and deployment of your base code, and how you can deploy custom code within larger software packages you may use, such as R or Python code within an SAS application.
Overall, you’ll need to make a nuanced version of the classic ‘build vs buy’ decision, deciding how to mix and match pre-existing analytic tools that have differing levels of complexity and inter-operability. Some of these tools will be open-source and some will be proprietary. For some well-established analytic models that have had time to be highly optimized by specialist vendors, such as linear programming, there are strong advantages to investing in an off-the-shelf solution. For specialized AI methods with limited but important application within your business, consider a pay-per-use model from vendors such as Google or Salesforce (Einstein) rather than expending internal effort.
The analytics specialist should bring a broad understanding of potential applications for data and analytics within your organization, along the lines of the applications I’ve listed in Chapter 4.
As you think about how to utilize data science within your organization, consider:
For each of the above points, your analytics specialist should be able to estimate the data, technology and personnel requirements necessary both for prototyping as well as for deployment.
As part of this process, they should be forming a holistic picture of the company’s existing and potential data assets, including standard operational, customer and financial data; the raw data collected in data lakes; third-party data that could be purchased or collected; and possibly even the dark data within your systems.
The project team will need technical expertise to ensure operational success. This will be someone with an understanding of the technology for data collection and transfer, general infrastructure and corporate databases. Data is a critical component of analytics initiatives, so your technical expert should bring an understanding of the extent and accuracy of the various data stores within the organization.
Your company may also have one or more operational data stores, data marts and/or data warehouses, which will provide the data for analytics projects. In addition, analysts will need to create tables, sometimes large tables, in one or more of these databases and may even need additional types of databases, such as graph or document databases (discussed further in Chapter 8).
The technical expert will provide direction and assistance related to your computing infrastructure, whether it be in-house server capacity or possibilities to provision in the cloud.
It’s important to build solutions that you can maintain long term. This will help you maximize the chance your analytics projects will bring long-term value. Your technical expert will help ensure these projects can integrate with existing technology.
To this end, ask the input of your technology expert regarding:
Your analytics efforts depend on IT support. Many projects fail because they do not get the buy-in of IT. Involving IT from the start serves four purposes.
After years of being awoken at 3 am or phoned during a holiday to fix malfunctioning production code, many IT staff will be extremely sensitive to analytics initiatives with even the slightest risk of breaking existing code. They may also be averse to any vagueness or uncertainty in project planning, wanting every project step to be laid out at the start. As we’ll see later, analytics projects don’t typically lend themselves well to extensive pre-planning, so some tension can easily develop in this regard.
Other IT staff can be very eager to try new technologies and new analytic methods. These are typically the younger ones with fewer 3 am experiences under their belts, but sometimes also the senior staff. Quite often, the developers in IT are extremely enthusiastic about analytic projects. These people will typically be some of the strongest and most valuable supporters of your analytics initiatives.
Your IT professionals may have two goals: stability and/or innovation. Many measure the success of IT in terms of reliability. Others measure its success in terms of creativity and innovation, even if the new features aren’t perfect.
Once you’ve selected your programme team, plan a programme kick-off meeting to lay the strategic foundation for the analytics initiative, sketching the framework for business applications, brainstorming ideas, and assigning follow-on steps, which will themselves lead to initial scoping efforts. The four skill sets represented in the programme team should all be present if possible, although the business expert may cover the strategic input and it is also possible (but not ideal) to postpone technology input until the scoping stage.
Also helpful at this stage is to have detailed financial statements at hand. These figures will help focus the discussion on areas with the most influence on your financials. Bring your standard reports and dashboards, particularly those that include your key performance indicators (KPIs).
Start the kick-off meeting by reviewing the purpose and principles that govern your efforts. Continue by reviewing the strategic goals of the company, distinguishing between the long- and short-term strategic goals. Since some analytics projects will take significant time to develop and deploy, it’s important to distinguish the time-lines of the strategic goals. If there is no executive or strategic stakeholder involved in the process, the team members present should have access to documentation detailing corporate strategy. If there is no such strategic documentation (as is, sadly, sometimes the case), continue the brainstorming using an assumed strategy of plucking low-hanging fruit with low initial investment, low likelihood of internal resistance and relatively high ROI.
After reviewing the principles and strategy, review the KPIs used within the organization. In addition to the standard financial KPIs, a company may track any number of metrics. Marketing will track click-through rate, customer lifetime value, conversion rates, organic traffic, etc. Human resources may track attrition rates, acceptance rates, absenteeism, tenure, regretted attrition, etc. Finance will typically track financial lead indicators, often related to traffic (visits, visitors, searches) as well as third-party data.
At this stage, probe more deeply into why certain KPIs are important and highlight the KPIs that tie in most closely with your strategic and financial goals. Identify which KPIs you should most focus on improving.
The business experts should then describe known pain points within the organization. These could come from within any department and could be strategic, such as limited insight into competition or customer segments; tactical, such as difficulty setting optimal product prices, integrating data from recent acquisitions or allocating marketing spend; or operational, such as high fraud rates or slow delivery times.
Ask the business experts to describe where they would like to be in three years. They may be able to describe this in terms of data and analytics, or they may simply describe this in terms of envisioned product offerings and business results. A part of this vision should be features and capabilities of competitors that they would like to see incorporated into their offerings.
By now your business objectives, principles, and strategic goals should be completely laid out (and ideally written up in common view for discussion). At this point, your analytics expert should work through the list and identify which of those business objectives can be matched to standard analytic tools or models that may bring business value in relieving a pain point, raising a KPI, or providing an innovative improvement. It’s beneficial to have cross-industry insight into how companies in other industries have benefited from similar analytic projects.
To illustrate this process, a statistical model may be proposed to solve forecasting inaccuracy, a graph-based recommendation engine may be proposed to increase conversion rates or shorten purchase-path length, a natural language processing tool may provide near-real-time social media analysis to measure sentiment following a major advertising campaign, or a streaming analytics framework combined with a statistical or machine learning tool may be used for real-time customer analytics related to fraud prevention, mitigation of cart abandonment, etc.
If IT is represented in your kick-off meeting, they will be contributing throughout the discussion, highlighting technical limitations and opportunities. They should be particularly involved during the analytics phase, providing the initial data input and taking responsibility for eventual deployment of analytics solutions. If your technical experts are not present during the initial project kick-off, you’ll need a second meeting to verify feasibility and get their buy-in.
The first output of your programme kick-off should be a document that I refer to as Impact Areas for Analytics, consisting of the table illustrated in Figure 7.1. The first column in this table should be business goals written in terminology understandable to everyone. The next column is the corresponding analytic project, along the lines of the applications listed in Chapter 4. The next three columns contain the data, technology and staffing needed to execute the project. If possible, divide the table into the strategic focus areas most relevant to your company.
By the end of your kick-off meeting, you should have filled out the first two columns of this matrix.
Figure 7.1 Template for Impact Areas for Analytics document.
The second document you’ll create in the kick-off will be an Analytics Effort document. For each analytics project listed in the first document, this second document will describe:
1. | The development effort required. This should be given in very broad terms (small, medium, large, XL or XXL, with those terms defined however you’d like). | |
2. | An estimate of the priority and/or ROI. | |
3. | The individuals in the company who: | |
a. | can authorize the project; and | |
b. | can provide the detailed subject-matter expertise needed for implementation. We are looking here for individuals to speak with, not to carry out the project. | |
These are the ‘A’ and the ‘C’ in the RASCI model used in some organizations. |
Distribute the meeting notes to the programme team members, soliciting and incorporating their feedback. When this is done, return to the programme sponsor to discuss the Impact Areas for Analytics document. Work with the programme sponsor to prioritize the projects, referencing the Analytics Effort document and taking into consideration the company’s strategic priorities, financial landscape, room for capital expenditure and head-count growth, risk appetite and the various dynamics that may operate on personal or departmental levels.
Once the projects have been discussed and prioritized with the programme sponsor, you should communicate with the corresponding authorizers (from the Analytics Effort document) to set up short (30–60 min) scoping meetings between the analytics expert and the subject matter expert(s). The exact methods and lines of communication and authorization will differ by company and by culture.
During the scoping meetings, speak with the individuals who best understand the data and the business challenge. Your goal at this stage is to develop a detailed understanding of the background and current challenges of the business as well as the relevant data and systems currently in use.
The subject experts and the analytics expert then discuss:
After each scoping meeting, the analytics expert should update the corresponding project entry on the Analytics Effort document and add a proposed minimum viable product (MVP) to the project description.
The MVP is the smallest functional deliverable that can demonstrate the feasibility and usefulness of the analytics project. It should initially have very limited functionality and generally will use only a small portion of the available data. Collecting and cleaning your full data set can be a major undertaking, so focus in your MVP on a set of data that is readily available and reasonably reliable, such as data over a limited period for one geography or product.
The description should briefly describe the inputs, methodology and outputs of the MVP, the criteria for evaluating the MVP, and the resources required to complete the MVP (typically this is only the staff time required, but it might entail additional computing costs and/or third-party resources). Utilizing cloud resources should eliminate the need for hardware purchases for an MVP, and trial software licenses should substitute for licensing costs at this stage.
Feed this MVP into whichever project management framework you use in your company (e.g. scrum or Kanban). Evaluate the results of the MVP to determine the next steps for that analytics project. You may move the project through several phases before you finally deploy it. These phases might include:
with each successive stage requiring incremental budgeting of time, resources and technology.
It’s very important to keep in mind that analytic applications are often a form of Research & Development (R&D). Not all good ideas will work. Sometimes this is due to insufficient or poor-quality data, sometimes there is simply too much noise in the data, or the process that we are examining does not lend itself to standard models. This is why it’s so important to start with MVPs, to fail fast, to keep in close contact with business experts and to find projects that produce quick wins. We’ll talk more about this in the next chapter when we talk about agile analytics.
Not all good ideas will work. Start small, get continual feedback, and focus on projects with quick wins.
The Otto group, a German retail conglomerate, employs over 54,000 employees operating in more than 20 countries.58 Since its establishment in 1949, it has grown to become one of world’s largest online retailers. The Otto group has developed several internal applications of AI, one of which nicely illustrates the benefits of close cooperation between business and analytics teams.
At a business level, Otto realized they were losing millions of euros annually because of the costs associated with product returns. The business and analytics teams worked together to address this problem in two phases.
The first phase was to analyse product return data to see what insights emerged. The analysis revealed that a significant portion of returns were products that took more than two days to arrive. Customers left waiting would either purchase the item elsewhere (perhaps at a local shop at discount) or lose enthusiasm for the product. The result was a lost sale and sunk shipping costs. Otto did not itself stock many of the products that it offered, hence the shipping delays.
This data insight led to the second phase of the analytics solution. If Otto could accurately forecast the product orders, it could itself order the inventory even before the customer placed the order. This would allow them to deliver within a shorter time window, resulting in fewer returns. For the analysis, Otto used several billion past transactions, combined with several hundred potentially influential factors (including past sales, online customer journey and weather data).
At this point, the analytics team had a choice of several analytic tools and models. They could have used a classic rule-based approach or a statistical model, selecting and refining features to construct forecasts for product groups. They also considered feeding big data into a deep-learning algorithm.
In the end, they utilized deep-learning technology and what they eventually produced was an analytic tool that could forecast 30-day sales with 90 per cent accuracy. This system now automatically purchases several hundred thousand items per month from third-party brands with no human intervention. Thanks to this analytic project, Otto’s surplus stock holding declined by one fifth and their product returns decreased by more than two million items per year.59
We see in this example how the Otto group used data and analytics in two key ways in addressing the problem of item returns. The first way was to diagnose the source of the problem, the second was to create a tool they could deploy operationally. These are two of the four primary uses of analytics within organizations. We’ll discuss all four in the next chapter.
18.117.98.250