Successful analytical outcomes that are well regarded by business stakeholders are primarily the result of empowered people applying processes to deliver analysis and are not simply the result of deploying technology. Although technology and tools are critical and necessary for enabling analytics, it is the people on the analytics team who are the primary users of analytics tools. People apply their analytical knowledge and analysis experience to answer business questions. Technology and tools by themselves do nothing to create value—and can be, in fact, overhead in many businesses. Most ecommerce analysts have run into the baggage of legacy and minimally supported systems containing data for analysis. To work with legacy and new systems, analytical processes need to be established to access and use the data. People who aren’t provided guidance on the process for doing analysis within an established company or who don’t develop process in a new company will not be able to maximize analytical effectiveness. A lack of analytical process impedes success. It’s cliché to say that the triumvirate of “people, process, technology” is important for success with analytics, but that’s actually a very accurate meta-concept. However, it’s not just having people, process, and technology alone that makes for successful analytics. Companies that succeed with ecommerce analytics organize, manage, align, work, communicate, and socialize differently. What successful companies do is build analytics organizations. They create analytics teams, appoint leadership, and resource the group. Successful analytics teams do not concentrate entirely on technology but instead focus on the output and outcomes of analytical work. It’s not the tools that are the largest factor in success in these teams; it’s the way people team up, organize, define analytical roles and responsibilities, operate, interact, and communicate analysis, which makes for success.
This chapter is about how you build an ecommerce organization, though the principles suggested in this chapter are almost universal for analytics in general. We will review the set of activities necessary for building analytics teams across ecommerce companies. The use-cases, goals, and activities for doing ecommerce analytics are similar, if not identical, across companies; therefore, it is possible to identify universal steps for building an ecommerce analytics organization. In fact, I wrote a book titled Building a Digital Analytics Organization, in which you can read hundreds of pages on the subject. What follows in this chapter is a perspective on a universalized approach for building analytics teams, which aligns with my previous work and has also been updated with a few more years of my professional experience.
The universal approach for building successful analytics begins outside of analytics—in the minds and ideations of business leaders. These top-level business executives will have recognized and understood the value and importance of analytics, and they desire to use accurate data and expert analysis to guide and inform their business activities, decisions, and goals These executive leaders will make a decision to hire an analytics leader to run the analytics function. In 2016, the role of Chief Analytics Officer (CAO) is increasingly common as is the Chief Data Officer (CDO). The CAO is a business-focused role that ensures analytics are applied at their highest and best use in the business, while the CDO is a more technically focused role responsible for ensuring data is governed, available, and accurate. The responsibilities of these two roles may be collapsed into one role in charge of using data and analytics as an asset, including all related analytics technology and tools, data governance, analytics engineering, analysis and data science, team management, and strategic leadership.
After a leader is hired, the work to build an analytics organization becomes focused less on why the analytics team is needed and more on internal work to identify what the team will actually do. This leader will be tasked with a lot of work that crosses business, technology, management in order to build an analytics team. The leader may have the luxury of creating the team and the analytics function from scratch, or they may have to work with existing baggage from previous attempts at analytics. In that context, a universal process for building analytics organizations is presented here:
1. Determine and justify the need for analytics. Typically, senior leadership will make the decision based on business need.
2. Gain support for hiring or appointing a leader for analytics. Whoever determines the need for analytics will also “drive the bus” and ensure that they have the right support and authorization to command the function to be created.
3. Hire the analytics leader. In an industry with a dearth of leadership expertise, this process can take a long time and require substantial salaries.
4. Gather business requirements. The analytics leader will run point to gather business requirements and understand the business catalyst driving the need for analytics.
5. Create the mission and vision for the team. The analytics leader will work to create these statements, which define what the team does and what success looks like.
6. Create an organizational model. Aligned with requirements, the analytics leader will determine the staff needs along with the roles and responsibilities.
7. Hire staff. Analysts command and deserve higher salaries than other workers with the same amount of experience and can be hard to find.
8. Assess the current state capabilities and determine the future state capabilities. The team will map requirements to what is feasible, determine the gap, and identify what needs to be done to deliver them.
9. Assess the current state technology architecture and determine the future state architecture. The team will determine if the technology can support requirements and, if not, figure out what should be done.
10. Begin building an analytics road map. The analytics road map sequences over the time the work that will be delivered.
11. Train staff.By training staff, you can ensure they know what to do in the current state—on top of the experience they bring to work.
12. Map current processes, interactions, and workflows. The team will need to document current processes and fix, evolve, and create new processes when needed.
13. Build templates and artifacts to support the analytics process. The team will create the format for work products that support the analytical processes.
14. Create a supply-and-demand management model. This model will help to match analytical demand from stakeholders to available supply of human and technical resources.
15. Create an operating model for working with stakeholders. This model documents how the team will work with stakeholders to initiate, execute, communicate, escalate, and close project work.
16. Use, deploy, or upgrade existing or new technology. As need arises, the team will work with technical partners to ensure the technology needed is available and operational.
17. Collect or acquire new data. The team will work to collect new data as needed to support analytics requirements.
18. Implement a data catalog, master data management, and data governance. The team will act as helpful data stewards of their data and apply data management and data governance practices.
19. Meet with stakeholders and participate in business processes. The team will work with peers and other groups to do their jobs.
20. Do analysis and data science and deliver it. At the heart of the function is the production and delivery of analysis that is helpful to people.
21. Socialize analysis. The analysts must go to stakeholders and communicate to them what the data says and answer their business questions.
22. Lead or assist with new work resulting from analytical processes. The team will initiate new work to answer questions that occur from their excellent analysis.
23. Document the financial impact and business outcomes resulting from analysis. It is important to quantify the financial impact of analytical work to justify the team’s existence.
24. Socialize the business outcomes and highlight the financial impact. Once the positive financial impact is documented, the members of the analytics team should use data to advocate for themselves and show how they have positively impacted the business.
25. Continue to do analysis, socialize, and manage technology to emphasize the business impact ad infinitum. Analytics work is ongoing. There is always more to do.
26. Manage change and support stakeholders. Business goals change and the nature of the business questions will evolve. The analytics team needs to stay on top of the business to maintain alignment.
Analytics teams don’t just appear in corporations out of thin air. They are often conceived in boardrooms and in meetings between senior leaders and executive and senior vice presidents. These teams can develop organically when business leaders realize that they have resources performing the analytics function and decide to align them under one team or across a line of business. Other times, an analytically savvy manager takes a solo initiative to embed analytics into the business function and enlarges someone’s job or jobs to support. There are several signs that indicate a need for analytics in an ecommerce company:
• Senior management does not have any data to make decisions. A lack of visibility to the key data that is perceived to be accurate or helpful is a primary driver for analytics team formation.
• Senior management has too much data to make decisions and needs help working through it. Analysis paralysis can result from too much data that is deemed to be relevant but isn’t because of the lack of interpretation and analysis.
• Senior management has conflicting data that can’t easily be identified. When data is not controlled, it can seep through many different pathways in an organization. This can result in ungoverned data proliferation without data definitions or standards. The end result is conflicting data that confuses people.
• Business goals demand careful attention to tracking data that informs business goals. In goal- and performance-driven and incentivized cultures, measurement and analytics is a necessity. The old adage applies: “You can’t manage what you don’t measure.”
• Competitors are determined to have analytics teams and are benefiting from the work. The competitive intelligence team may have identified that a competitor leads with data, has advanced analytical capabilities, uses data sources, or applies data and analytics in innovative or new ways, thus creating competitive advantage that the company wants to counter.
• Line-of-business leaders need help setting strategy or managing programs that can be tracked quantitatively. Teams in ecommerce companies, such as merchandising, customer service, and marketing, benefit from enhanced data analysis, whether tracking the impact on conversion of particular brands or marketing programs.
In all of these cases, it is important to justify the investment in analytics, document and socialize your rationale for investment into staffing the analytics function with all the key stakeholders, and then present it to senior management. This approach is bottom-up (i.e, it comes from middle management to senior management). The alternative is the top-down approach, where senior leadership commands the function be created and resources it. In either case, it can help to justify the investment and develop a business case specific to ecommerce:
1. Specify the ecommerce business problem or challenge or initiative. The business issues, concerns, needs, requirements, or, ideally, business questions must be established, documented, and agreed on. This detail provides the foundation for which to map needed future state capabilities and helps to frame the type of resources—from hardware to software to human—that you will need.
2. Indicate the financial impact that would result from the problem not being solved. Missed opportunities or ways to generate new or incremental revenue or decrease costs can be quantified. If you frame the business problem monetarily, the impact is clear.
3. Identify how analytics can help solve the problem or augment the company’s ability to execute. The analytical capabilities you map to the business area and its goals must guide decisioning or provide input to solve the business challenge. In some cases the work may be solely led by the analytics team, and other times analytics will assist, guide, or advise other teams.
4. List the investment needed in terms of the fixed and variable costs, including head count, software, and hardware. Be specific about the investment required. Put it in clear financial terms and extend the analysis to incorporate all cost factors. Don’t try to hide or diminish the investment.
5. Create a financial model, using NPV or IRR, that identifies the financial impact of the investment. After you have all the inputs outlined previously, it will be possible to calculate the impact of the investment in today’s dollars (using a net present value method) or to evaluate if the investment would exceed the company’s cost of capital (the internal rate of return).
If you are at this point in building an ecommerce analytics team, congratulations, your investment decision was improved. You may need to hire a Chief Analytics Officer or other appropriately entitled expert, such as a VP or Director of Analytics. What’s important at this stage is where you will source candidates from three areas. Let’s explore each:
• Have the line-of-business ecommerce leader or senior executive who justified the investment lead the analytics team. In some cases, the executive who got the budget to hire for analytics will run the team. This person might be in marketing, finance, or merchandising and will simply add “analytics” or some derivative (consumer analytics, marketing analytics, data science, and so on) to their title and then lead the analytics function. You can appoint someone internally who already works at the company to lead the function, or you can hire someone externally to run the team. The decision point here is whether the existing employee can handle taking on a larger role from an expertise, capability, and efficiency perspective and whether he has the skill set to deliver. In some cases, an executive may be empire building: When commanding analytics falls under his jurisdiction, he takes on the role as the analytics leader, on top of his other work, and then hires a subordinate who does the real analytical work while he manages his empire.
• Promote from within the organization. In companies whose staff have analytical ability and may be fulfilling analytical roles either part-time or full-time, it may be logical to promote them. This path to hiring an analytical leader is helpful because existing staff can have a strong understanding of the business and its drivers, the data, and the technology in-house. On the other hand, employees can become entrenched in ways of working and may suffer from learned helplessness that diminishes their ability to excel as analytical leaders. Politics may exist and carry into the newly established analytical function.
• Hire from outside. For analytical roles, this is easier said than done, because as of 2016 there is still an acute insufficiency of analytical leadership, especially at the more senior levels. Although the rank-and-file analysts and data scientists who work in ecommerce are found more frequently than in years past, it is still a difficult role for which to find talent. The benefit of hiring from the outside is that someone external can bring perspectives on what has worked in the past, may have business accelerators and ways of working that advance the analytics function, and is removed from politics.
Ultimately, the best path for your company is choosing from one of these alternatives. If you exist in a corporation heavy with politics, where divisions don’t collaborate, or where there is little analytical management expertise, you want to hire from outside. If you have an analytical leader in your midst to appoint, in many cases, the company already knows who that is, so enlarge, broaden, or create a new role for them.
Hiring an analytics leader requires the prerequisite of writing and posting a job description, The description can be created from the requirements that justified the investment. The job description should express the job title, the role and responsibilities, the analytical expertise, including the concepts, methods, and technology the leader should be skilled in. The job description should also advocate for the company and give helpful information about the culture, benefits, and salary. After the job description is created, there are several choices for filling the role:
• An internal human resources team hires from outside via referrals or by surveying the available market. Most mid-size and large companies have one or more human resources staff who work via established processes for hiring human capital.
• The hiring manager sources candidates from their network or social media. In this approach the hiring manager gets no or minimal support and is just told to hire the person. They look for staff on LinkedIn and on employment sites, consult their social networks, and attend industry conferences and events.
• Recruiters and employment search specialists are used. For-pay recruiters can be contracted on an individual job-hire basis or for a period of time to find analytics candidates.
My experience has been that recruiters in analytics are generally more successful in finding candidates, but internal human resources teams can be better at getting people to take the offer. Recruiters often work across geographies and have a deep network of experts and a large database of potential candidates. Human resources teams use traditional channels, such as job boards and employment-based social networks. It’s a lot to ask hiring managers to find a suitable candidate who is available from their social networks without other support from HR or recruiters. After all, they have their own job to do.
Of course, the right path to hiring an analytics leader depends on the company, the role, the recruiter, the HR team, and the hiring manager. In start-ups, it not uncommon to see hiring managers bring their team from other jobs. Or for some companies to choose not to work with recruiters. Other large brands that are common household names cast a wide net and have HR teams that are better suited than external recruiters to land candidates. Your mileage will vary from mine—and I prefer to use all the channels: myself, my network, the HR team, and recruiters if I can get the budget.
At this point, the executive who created and gained approval for the business justification to staff for analytics has hired the leader for the analytics team. The primary function of the leader initially will be to understand the business and the needs of their peers and stakeholders and to gather business requirements that will define the scope of work for the analytics team. The business requirements must capture specific business questions, data sources, data sets, and the in-scope deliverables that will prove success when provided. Gathering business requirements at the leadership level requires meeting with people in-person or virtually. To gather business requirements, you can interview people one on one, do large workshops, or small focus groups (Liles 2012). Whatever your method of choice, you must create the materials for eliciting and documenting business requirements, and then you must do the work to capture the requirements, socialize them, and gain approval and sign-off.
One of the first actions of the analytics leader is to define the team’s mission and vision statements. A mission statement is an unwavering declaration made to the key stakeholders that defines the ultimate goal and core purpose of the analytics team over time. It defines what type of work the team will do (and by proxy, what work the team will not do) and sets the frame for what the team will accomplish. The mission statement is often elaborately articulate but could be as simple as “the analytics team mission is to increase gross revenues via conversion analysis and optimization.” On the other hand, the vision statement declares the aspirations of the team to accomplish over time. A vision statement could be “the analytics team vision is to embed data-informed decision making in all aspects of the company.” The mission statement and vision statement are symbiotic and accompany one another to define the purpose of the team and the outcomes of it (Liles 2012).
An organizational model is the hierarchical organization chart that defines the management structure and employee roles and responsibilities. It will reflect standard operating procedures and, by virtue of the stated roles and responsibilities, indicate who manages the team, who makes the decisions, and who does the work. There are many ways to create an organizational model for analytics (Davenport et al 2010):
• Centralized. This model is essentially what I am laying out in this chapter. One leader owns the function for the entire ecommerce company. This person may be a senior leader, in the case of the CAO or CDO, or may report to one. All aspects, both technical and business, flow through this team; however, the technical resources to support a centralized team will likely, for the most part, live in IT. The centralized team will own the vendor relationships and the budget for business analytics tools.
• Decentralized. Decentralized models involve a lack of corporate structure and alignment. An analyst or a team of analysts is under some management structure in each line of business: merchandising, buying, planning, shared services, fulfillment, warehousing, marketing. products, finance, legal, privacy, customer service, IT, and so on. These analysts have self-sustaining capabilities to generate analysis specific to their line of business. Across the company the analysts rarely talk to each other or collaborate. There may be little logic to the size and function of each team.
• Center of excellence. This model requires a team of people who act as advisors to other business units in the form of methodologies, tools, technologies, capabilities, models, and expertise that can help the company do analytics and embed it into their daily operations. The center of excellence model works well for large companies, and parallels the centralized model, but the centralized team has a larger responsibility for analytical delivery and outcomes. The center of excellence team, on the other hand, provides reusable components, training, best practices, and other guidance to help lines of business use data and analytics.
• Functional. Functional models exist when one or more lines of business hire analysts and analytics managers to serve their function. It’s like a decentralized model except the lines of business have the same organizational footprint or size with minor differences. Collaboration occurs across functional areas and shared systems may exist.
• Consultative. As in a centralized model, in a consultative model the analytics team is “hired out” by other business units and deployed to address a particular program or project.
• Hub and spoke. These models are centralized and consultative at the same time. There is an analytics team that is recognized as a formal corporate structure where all analysts work, but each analyst is assigned to navigate a primary line of business and help the team to which they are assigned use the data.
After the analytics leader has determined and prioritized business requirements, it is time to start writing job descriptions that identify the key needs for ecommerce analysis. A formal approval process will occur, and after approval the leader will work with an internal human resources team, use a recruiter, or find the hire through her own professional networks. Hiring staff means having a plan for what they will do when they start working. The business requirements and the analytics team mission and vision will guide this work. But there is some other core and foundational work the first hires onto your ecommerce analytics team will do. You want them to have bought into or soon buy into your strategy. Thus, you want these resources to evaluate the current state, assess gaps, and plan for the future state, as discussed in the next few sections.
A current state assessment is a core deliverable from the nascent, newly formed analytics team of more than one person. The current state assessment should rank the company on some maturity curve and identify the difference between the company’s current capabilities and what the future capabilities could be. Assess the current state maturity against these criteria:
• Scope can be assessed in terms of the areas of analytical coverage ranging from individual projects to across the enterprise.
• Sponsorship may be from a single person as the sponsor or ranging up to C-level executive sponsorship.
• Funding for analytics may not exist or may be self-funded in the case of analytics teams who can justify their business impact.
• Value comes from increasing revenue or reducing cost, so analytics teams vary from being immature with no financial contribution to being very mature such that the analytics team monetizes the data in some way and creates new revenue streams—for example, selling the data about the sizes bought for popular clothing back to manufacturers or using the analytics data to power personalization.
• Architecture manifests itself in mature companies as self-service analytics based on a data lake or an enterprise data warehouse. The most immature architectures are built-in tools like Excel or loosely cobbled-together data structures and sources that are ungoverned.
• Governance like architecture may not exist or it may be cross-enterprise and steered by committees.
• Data ranges from untrustworthy to highly trustworthy across the entire company.
• Communication in terms of whether any consistent standards are practiced, such as regular meetings, newsletters, office hours, and so on. The most immature analytics teams have no standard cadence for producing or communicating analysis. The most mature companies do.
• Delivery can range from people sending data in Excel to elaborate automated architectures that use artificial intelligence and machine learning to produce analysis.
• Analytical outcomes may not exist, in the sense that people just release reports, or they may range all the way to predictive and prescriptive analytics that provide insights and make recommendations that guide decision making and positively impact financial performance.
A current state technology assessment begins by identifying the technical architecture used for analytics. This work will be done in coordination with IT. You want to find out what software has been bought, by whom, and under what license and terms. The hardware, if in a data center, or the cloud infrastructure must be understood. The technology for collecting and acquiring, governing and mastering, transforming and preparing, reporting and analyzing, optimizing and predicting, and prescribing and automating must be mapped to an architectural diagram of considerable detail. It is common for an analytics stack to contain the following components, which will be evaluated and examined in a current state architectural assessment:
• Storage: The systems like databases and big data processing platforms that house and provide access to raw data.
• Data management and governance: The systems and processes that move data from a raw state into a state usable by the business such that all data is consistently defined and accurate, such as master data management (MDM) tools.
• Analytics platform(s): The tools and technology used primarily by the analytics team. These tools may be part of one vendor’s offerings and may include the functionality for connecting to data, extracting and transforming data, preparing and cleaning data, defining custom data models, unifying and joining data, building reports, visualization, and KPI dashboards, doing data science, and feeding good data to other systems for automation and artificial intelligence.
• Models: The scientifically rigorous way of analyzing the data such that statistics and math are applied. In other words, data science! A model may also include data collection, ingestion, processing, and transformation for use in data mining, machine learning, and artificial intelligence.
• Visualization: Visual representations of key data elements that are presented in an intuitive way that exposes and illustrates relationships, patterns, outliers, and relationships in the data, including the ability to explore the visualization by filtering, using dimensions, drilling up and down, and applying custom metrics.
• Self-service: The tools that stakeholders, outside of the analytics team, use to work with data and analytics. The analytics team, of course, will also use these self-service tools.
The analytics road map is a detailed, multiyear plan that shows conceptually how the analytics team will deliver against business requirements. A road map is often done by fiscal quarter or within periods as short as a week to as long as a year. The technology, data, people, and projects are mapped over a timeline and sequenced in priority order. In this way, a person who looks at the road map will know, for the most part, what work is being done and will be done in the future, when, and for how long until completion; who is doing the work, using what data; and what the projected financial impact could be.
The road map must be constructed from multiple data points. Use the business requirements that were gathered and contextualize them against the known current state capabilities and technical architecture. By triangulating the type of work needed from the predecessor work outlined in this chapter, you can plan the projects to execute to deliver business requirements and know what technology and data you will need, and when to deliver.
A road map is still supported by Agile analytics approaches. The road map, in Agile, is at a very macro level and is used to guide the sequencing of scrums and sprints as work is integrated. Programs become epics. Projects become stories. Road maps, of course, are central to waterfall methodologies.
Although it is expected that analysts who join an ecommerce company will understand the basics of ecommerce analytics, that is not always the case. “Entry level” employees need training, as do people who may have worked in different industries before ecommerce. All employees need training about the company, what it does, how the analytics team fits in, who the stakeholders are, what the supporting processes and teams are that impact analytics, the types of analytics deliverables, and even training on how to use specific technology and tools. Dedicating time to training people how to do their jobs effectively and use the tools provided can only be helpful. Training may be costly or time-intensive, but it is most often a net positive investment that generates team collaboration and leads to better analytics outcomes more quickly.
As your analytics team begins to do analytical work, analysts must pay attention to the processes they use, the interactions they have with people, and the workflows in which they are asked to participate alongside other teams. These processes must be documented and mapped out using process diagramming techniques. The ways to interact with the team, specifically the handoffs of work to and from other teams, must be defined. Workflows for interacting with supporting teams and stakeholders must be created. Processes are a series of steps that must be executed to accomplish a goal. You may want to map the following processes:
• Request for data collection process. How is data requested and what inputs are needed to begin data collection?
• Prioritization process. How are different projects prioritized so they can be delivered in a sequence that is timely and meets requirements?
• Analytics escalation process. How are the concerns of analytics escalated to leadership when necessary?
• Data science model creation process. How is data science accomplished?
The analytics team should map out the following interactions they will have with other teams and people:
• Requirements gathering and project elaboration with stakeholders. How are requirements gathered? What is the process? What are the artifacts?
• Communicating and prioritizing projects with IT. How are analytical needs and projects communicated, prioritized, and delivered by technology partners?
• Delivering analytical work. How is analysis delivered to end users and communicated to stakeholders?
Workflows are sequences you will want to document, including these:
• Data sourcing from raw to curated. Who does this work; what stakeholders are involved; what technology is used by whom?
• Requesting work and responding to work requests. What is the process for requesting work from the analytics team? What should the team do when a work request is submitted? How should the team respond? What inputs should people requesting work provide to the analytics team?
Templates are standard formats for analytical deliverables; artifacts are procedural documents that support processes. Templates can be used to communicate analysis and can include items such as reports, dashboards, visualizations, PowerPoints, and related presentation tools. Artifacts can include items such as the analytical plan, report wireframes, visualization mockups, and the documents that support processes, such as the analytics specification, the data specification, and epic and story templates that support Agile analytics.
This activity is crucial to newly formed and existing analytics teams. The team needs a model for taking in requests from the business and delivering accepted results back to the business. In practice, such a simple-to-understand set of activities can be very hard to execute for a number of reasons: resources, politics, complexities, technology, priorities, and so on. The activities for managing supply and demand must be mapped out into a model that does the following:
1. Intake requests for demand for work.
2. Capture sufficient details of the demand.
3. Guide the team in how to respond to the work.
4. Prioritize accepted requests and backlog rejected ones.
5. Understand available supply to match against project demand.
6. Align resources to execute and manage them.
7. Manage change, urgencies, and escalations.
8. Communicate status to management, the work team, and requesting stakeholders.
9. Deliver analysis.
10. Gain stakeholder acceptance.
11. Close the work requests after delivery of analysis that has been accepted by the person or team initially requesting the work.
An operating model is an abstraction of how the analytics team works to accomplish its functions. You build the operating model based on the organizational chart, process, interactions, and workflows you have previously identified and documented. An operating model stitches together all of these items in a way that can be understood by people outside of analytics. The operating model is usually expressed as an authored document or as a presentation that encapsulates a business view of how the analytics team operates as a functional entity and how it interoperates with other business and technology teams. An operating model explains the following about the analytics function:
• Who is the leadership and how do they manage and govern analytics?
• What is the organizational chart and the names, titles, roles, and responsibilities of people on the analytics team?
• What are the processes, workflows, and interactions for analytics and who or what teams participate in them?
• How is demand for analytical work from the business mapped and managed to the available supply of analytical time, resources, and technology?
• What are the analytical services, deliverables, and outputs provided by the team; what are they and why are they important; what is timeline and frequency for delivery?
• What are the success measurements and benchmarks that indicate the analytics team is or has been successful in creating business value?
Analysts use tools. All sorts of tools—from operational business intelligence tools, to conversion testing tools, to digital analytics tools, to data preparation tools, to data visualization tools, to data science tools. The list of tools for different analytical work is extensive. New tools are released or updated almost every day. Because you are an analytics leader who has mapped your current toolkit to business requirements and trained people on how to use the tools (assuming you are following the suggested work in this chapter), now your team will be using the tools. The tools will need to be maintained; thus, upgrades will be required. To accomplish these tasks, you will need strong alignment with IT, which you should have had because you created processes, mapped workflows, and socialized your well-conceived operating model for the analytics team. On top of this, you should consider establishing service-level agreements (SLAs) with your company’s internal technology teams—and verify the existence of acceptable SLAs for your cloud infrastructure and tools.
One of the ongoing, never-ending 24/7 activities in analytics is the process of data collection. Sites and mobile apps are instrumented with data collection tags and other methods to track behaviors, events, interactions, transactions, goals, and so on. All this data is being captured in one or more databases. Data is being created from purchase orders, fulfillments, inventories, warehouse, and shipping. All this data is part of the overall ecommerce data pipeline, and the analytics team will likely need to expand that pipeline to include new data from existing sources or from new sources entirely. When building a team that is responsible for collecting data, you must hire expertise that can identify the target data to collect, determine how to collect it, implement the data collection, store the data, load and transform the data, and make it available to the system used to do analytics.
A data catalog is a software system that can contain definitions of data, queries, metadata, and information about values of the data in the database, including database information such as tables, views, indexes, and users. This information is helpful to ecommerce analysts who want to discover what data exists, what values the data has, what other team members created or use the data, the query for the data, and other information to use to navigate the data. Data catalogs also have a management layer where data or queries can be deprecated, the usage understood, and lineage visualized.
Master data management is the name for a set of methods for creating a master set of data residing in one database or file. This master data or master data file is considered to have validity, integrity, and accuracy. It is used as a reference when creating derivative data or to validate new data created. Master data management provides the single source of truth that many companies aspire to create.
Data governance is the name for a set of activities related to the management of all data within a company. Usually governed by committees that are made up of data owners and data stewards, the practice of data governance requires the creation of defined procedures for managing data that is aligned with a plan for continually governing data. The goal of data governance is closely related to the goals of master data management such that a data governor would use master data management principles to ensure the accuracy, usability, integrity, lineage, security, and consistency of data within the organization.
For more information, see Chapter 14, “Governing Data and Ensuring Privacy and Security.”
The positive perception by stakeholders of the value and benefit of the analytics team helping them drive their business goals is crucial to success. Stakeholders get value from analysis by being in the loop. The only way to keep people in the loop is for the analytics team to communicate with them on a regularly scheduled cadence. It sounds easy enough; you just need to talk to people, right? But it can be difficult based on work schedules, project demands, and other distractions. It may sound obvious, but there are several ways to communicate analysis, such as in-person, virtually, on the phone, via e-mail, SMS and text, business social networks, the intranet, and so on. The cadence for analytical communication by the analytics leader could be daily with team members, weekly with technology partners, biweekly with your manager, and monthly with business stakeholders. Of course, the right cadence for your team and your company could be totally different. Cadence is less important than delivering excellent work that people like and value. This perception is what will make or break the analytics team. Perception is reality—and you want all stakeholders across the company to perceive the analytics team as experts who do valuable work that helps people improve the business.
Your job is to do analysis and data science and deliver it to people in a timely manner when they want and need it. The primary outcome of successful analytics is the creation of business value. The value is created by people taking action from recommendations based on information found or knowledge learned through analysis. Many times it is not even the analysts themselves who take action on the data. As you can read in this book, there are many types of ecommerce analysis that can be done for different lines of business and stakeholders. At a macro level, Jeffrey Leek, Professor at Johns Hopkins University, identifies several analytical archetypes, which can be applied to many types of ecommerce analysis:
• Descriptive analytics does just that. It describes a set of data so that basic statistics are known about it. Usually applied to a larger data set, descriptive analysis involves interpreting data and then describing it. Think of the U.S. Census.
• Exploratory analytics attempts to find hidden, unseen, or previously unknown relationships. You explore data to find new linkages and connections between data points just by looking at the data. Correlation may be used to uncover relationships, which may not be casual but associated or dependent. Data visualization is often used to guide exploratory data analysis.
• Inferential analytics uses sampling to tell the analyst something about the larger population; thus, statistical error can have a dramatic impact on the meaningfulness and utility of this type of analysis. Inferential modeling is routed in statistics.
• Predictive analytics is a set of mathematical and statistical methods that attempt to predict an outcome from a set of data. The outcome is not caused by the variables in the data set; it is dependent on them. In that sense, the proper construction of the model, including dimension reduction and variable weighting, can have a material impact on the predictive power of the model.
• Prescriptive analytics uses predictive techniques and other methods to automatically suggest the best possible decision to take based on all available options. It is probably the most complex and challenging type of data science to do.
• Causal analytics attempts to understand the influence or impact of one variable or set on another variable. Causality is a difficult concept to prove, and thus great lengths are taken to use random data to determine causality.
• Algorithmic or mechanistic analytics seeks to explain the relationship, influence, and interdependence of a set of variables such that changes to one variable are understood in the way they impact other variables. Algorithmic approaches to analysis can involve machine learning that uses deterministic approaches with no randomness. Any randomness is considered to be error. As you might imagine, algorithmic analysis is complex (Smith 2013).
For more information, see Chapter 3, “Methods and Techniques for Ecommerce Analysis.”
Analytics teams don’t just want to “live on an island.” There will certainly be times when the team must cloister itself away and have long uninterrupted periods to do analytical work. And there will also be other times when the analytics team must lead or assist in leading with new work resulting from analytical outcomes. To ensure that analytics is focused on the stakeholders and not just on doing analytics, it is suggested that you do the following:
• Set up regular checkpoints. In these checkpoints, collaborate with team members and talk to stakeholders to find out what analytical work will be needed in the future.
• Get involved with recurring business planning meetings outside of analytics. Go to meetings hosted by other teams. The analytics team has a right to be there if that team wants the analytics team to do work for them or with them.
• Create a queue in the ticketing system you use. If your company uses a ticketing system, like Jira, then you should create a way for people to “ticket” analytics work. While work tickets can be cumbersome at first, they make work transparent. Ticketing systems also have reports that can show what’s been done by whom and for whom and what still needs to be done.
• Host office hours. Open up the doors of your office or book a conference room in which the analytics team and the experts on it can meet with stakeholders. In these office hours, analysis can be discussed, data can be reviewed, and future work determined. Office hours are also helpful when stakeholders need more guidance or benefit more from “higher touch” analytics services.
To prove the impact of data, ironically, you need data. In this case the data you need is financial data that demonstrates that analytical work has been used to improve the business and create net positive outcomes. In the optimal world the financial estimates you create to justify your company’s investment in analytics must be tied back to actual outcomes. You must quantify the financial outcomes of business programs and projects in which analytics contributed in some way. To influence people with analytics, you need to socialize your analysis. Analytical outputs and deliverables must be made available to the wider corporate audience; project status and results must be communicated; stakeholders must be updated on progress toward their expectations. Some of the ways to document and socialize analysis include publishing successful work on the intranet, holding meetings to highlight business impact of past work, creating an efficacy dashboard that shows what the analytics team has done and the impact, creating infographics to show key data, doing periodic business reviews with stakeholders, and so on.
Analytics is an ongoing process that continues day by day, week by week, month by month. The analytics team must continue to analyze data, socialize it, and manage the technology that supports it. Because you have business requirements and an organizational structure including an operating model, as well as the artifacts to support the development, creation, and rollout of analytics, you simply need to continue to do what you said you would do—and make sure you are answering the key questions that can help stakeholders do their jobs more effectively. Here are some tips that work for guiding how you continue to do analysis, socialize, and prove the effectiveness of the analytics team:
• Double down on what’s working. If people like the work and it is providing business benefit, consider doing more of it more frequently. Or model and apply the successful approaches to analytical work in one line of business to another.
• Pay attention to industry and technology trends. Keeping up to date with what thought leaders, experts, and practitioners are saying about the industry keeps the team current in their methods.
• Publish a newsletter. Sending out a companywide e-mail or an e-mail to a smaller set of stakeholders that contains a newsletter is helpful. The newsletter can contain data, analysis, visualizations, a list of available analyses, the team’s business impact, and so on.
• Do periodic education sessions. Meeting with employees and explaining to people what the analytics team does and how to work with the team is important.
• Demonstrate best practices for self-service tools. The team may want to train people or guide them on how to best use analytical tools to get the data they need without requesting work from the analytics team.
In the best of times, there will be change to manage. In the worst of times, there will be escalations and critical urgencies and even crises to manage. Change management refers to a methodology or set of techniques that enable a team to transition to a new state. The new state may result from the introduction of new products, new promotions, new customer types, and marketing externalities that impact ecommerce (such as various holidays and seasonality). When managing change, the leader of the analytics team needs to take a top-down approach that focuses on helping people adapt to the change. Change is handled differently based on your company’s approach to delivering analytics. For example, Agile analytics processes are more flexible to adapt to change than are annual, quarterly, or waterfall planning processes. When considering how to manage change and support stakeholders, consider how the following concepts can help you:
• Change management process. Document the change management process for analytics.
• Quarterly analytics reviews. Review what the data is saying about the business quarterly.
• Project postmortems. Explore why projects and analytical work were successful (or not) so you can reuse what worked and fix what didn’t work.
• Agile sprint and scrum masters. Appoint analytics team members to work Agilely and even to learn how to be scrum masters.
• Planning committees. Create planning committees to align the work of the analytics team with other teams and business plans.
While I have discussed a universal process for building an analytics team and organization, it is even harder to build analytics culture. My friend, Gary Angel, in his book and blog Measuring the Digital World, put some thought into how to build analytic culture. His take is that there are work and work products that the analytics team creates and manages that help create analytics culture. Gary believes, and I tend to agree, that things like analytics reporting and dashboards, a cadence of communication, having an advisor work with the C-level, and “walking the walk” are all helpful. In addition, he cites having tagging standards, metadata, “rapid” VoC data, testing plans, doing segmentation, annotating data with narratives, and a focus on continuous improvement are useful. Other activities like having an independent expert audit the analytics function yearly and having frameworks that define the data to be used to evaluate success can be beneficial.
The process for building an analytics team involves each of the steps I’ve outlined previously. The order in which I’ve presented these activities, though linear, is a good sequence to follow but you may do it differently. And that is okay. Of course, all companies are different, and some of the activities I’ve identified may already be occurring or, if not, may be occurring in other business areas. In this case, the analytics team should leverage past work and not do something different. For example, there may be standard business justification templates and models for justifying investment or proving the financial impact. Overall, however, the universal approach I’ve outlined captures how to build an analytics organization in ecommerce and even in other industries. For more information, please read my book Building a Digital Analytics Organization.