- Tags:: #📚Books, #✒️SummarizedBooks, [[Data team vision and mission]] - Author:: [[Jason Foster]], and [[Barry Green]] - Liked:: #3/5 - Link:: [Data Means Business (cynozure.com)](https://info.cynozure.com/data-means-business) - Source date:: [[2021-03-06]] - Read date:: [[2022-03-01]] - Cover:: ![[cover_data_means_business.png|100]] As I've already expressed many times (e.g., in [[📖 Fundamentals of Software Architecture. An Engineering Approach]]), I love survey-like books. They help me cope with [[uncertainty]] by having a kind of "map" for a long travel, or as Patrick Collison says more bluntly, by [[✍️ Refusing to stand on the shoulders of giants#Refusing to stand on the shoulders of giants|cheating]]. These books come with some risks, however. As [[Eric Ries]] says in [[📖 The Startup Way]] (a book that is, ironically, referenced here): >Our world is awash with gurus and experts telling us all to move faster, be more innovative, and think outside the box. But we are short on specific details: How, exactly, do we attain these results? (p. 11) On my first read of this "Data Means Business", I was disappointed: that was precisely my thinking. My tl;dr was "spinning up Data in an org is going to be hard, but just focus on gaining momentum through easy opportunities". After writing this summary, I no longer think this is true. While I still think that some book sections are vague (e.g., I really missed more detail on the process of continuous discovery of Data Science opportunities), and most of it won't blow your mind, it does provide sound practical advice on all the key things you need to sort out, and more importantly, "templates" to frame your thinking and thus, your answers. I see loads of meetings, thinking, writing, and ROI bounds calculation on my horizon... ## 1. Thinking differently ### The business ecosystem Businesses are complex and thus, managing data is complex: that cannot be abstracted away. This is similar in plain software engineering. >[[James F Moore]] coined the term *business ecosystem* (...) a setting in which *companies co-evolve capabilities around a new innovation: they work cooperatively and competitively to support new products, satisfy customer needs and eventually incorporate the next round of innovations* (...) if we are to implement the change needed to become a data-guided organisation **we need to move away from thinking about data as special and separate** (p.21). ### Data means business. Business means data >In his book [[Management: Tasks, responsibilities, practices]], [[Peter Drucker]] argued that strategic planning and innovation should be carried out on the basis that the future is uncertain and unpredictable (...) Data is the enabler that will help us (...) **Data is a business skill. A big, critical, value-adding, differentiating skill.** (p.23). >The industry tends to use the phrase [[data literacy]] (...) an organisation that is able to apply insights to a business situation in order to make better decisions. **Getting value out of data is therefore everyone's responsibility. Data means business**. (p. 24) > Everyone in an organisation uses data to make a decision. They may not refer to it as such, you'll need to help them with that realisation. But they do. What they don't usually know is what data is available that may help them to a faster, better decision, or how to ask a different question that might yield a better answer (p. 25). ### Creating the data-guided business #### A start-up philosophy The author suggests that you, as a data leader, should be in such mindset and accept, specifically, that: > - Success is not guaranteed and requires entrepreneurial thinking and leadership > - They do not have long-term funding and need to prove value to get funding, survive and thrive (p. 32). #### Culture and the science of people > No matter what the existing culture is in an organisation, if you need to improve your use of data you will need to change culture (p. 36). And the typical data self-help, as in [[📖 How to lead in Data Science]]: > You will need to be **resilient**. You can't start by trying to change the entire business culture, but you can influence how the data team behaves and interacts with the broader business. You are trying to change from within by starting small and then finding new ways to 'go viral' (p. 37). >When you are told AI is the next 'silver bullet' for your organisational problems, you need to **take a step back. There is no silver bullet. Algorithms need data to learn.** Creating algorithms that are not operationally effective, no matter how clever, will not bring true value. Data science is one of the many components needed to maximise your understanding of what and how you operate (...) **they need to exist in processes that touch real people** and be managed and move with change in both the internal organisation and the external environment (p. 39). #### Data as a product > There are proven benefits of treating data as a product in your organisation in a similar way to physical and digital products (p. 46) >Some of these data products are there to serve as ingredients for other products and some are finished goods in their own right (p. 47) > Like the consumer products mentioned, each data product: > - Exists to solve a customer need. > - Has features and KPIs. > - Has a product manager who owns the features and is measured on the KPIs (p. 48) #### Data as a service > Service is taking action to create value for someone else. That action is the experience, expertise and understanding of providing a set of data services. **This is the role of the data team (...) helping to federate data capability into the organisation through the defined set of services created to solve a business need** (p. 50) So you have both. I particularly like how these two views are combined in [[🗞️ Data as a Product vs. Data as a Service]]. ### The Level Up Framework ^89bfee ![[Pasted image 20220414070904.png]] This is the core proposal of the book. It has five stages: 1. Establish: sets the agenda and gets buy-in. Optimum duration: 3-6 months 2. Prove value: MVPs. Optimum duration: 3-6 months. 3. Scale. Optimum duration: 6-12 months. 4. Accelerate. Optimum duration: value-driven. 5. Optimize. Optimum duration: ad infinitum >The data profession tends to attract certain stereotypes (...) one being the introverted, conscientious, highly ordered, detailed, 100% right data analyst (...). The problem is when someone gives you some data, it's never up to the standards you would set yourself if the boot was on the other foot (...). **It took me a while to realise that it will kill you trying to aim for a destination for perfection in one go.** (p. 56) ## 2. Establish ### It's a bit of a mess Everybody starts on the wrong foot: >... even if you haven't consciously or strategically tried to sort it out before, it's still likely that huge investment has already been made in it (p. 63) ^b8a1d5 > No one knows who owns the data. Multiple definitions of sales and 30% of every management meeting spent discussing which sales figure is correct (p. 64) And you inherit such a mess: > This makes the starting point particularly challenging and, unlike a start-up, you have this legacy to deal with and history to untangle (p. 64) ^3c03c5 ### Educate and tell stories > You need to show the art of the possible with data. Have you taken your leadership team and key stakeholders to the market-leading players or data and digital native organizations to see how they use data to potent effect? (p. 65) ### Build out your strategy A data strategy (in the sense of [[📖 Good Strategy, Bad Strategy]]), in which data means business: >Your strategy is about identifying the parts of your business data can be used to improve. It's a business strategy and should first and foremost talk about business outcomes (p. 67) > A commercially focused data strategy forms an effeective shelter to help survive the bad times but take advantage of the good when they come (p. 69) >You should be looking to identify business problems or opportunities to focus on, along with the size of the prize (p. 67) ### Understand stakeholder buy-in How can't we have [[Kitchen sink]] dynamics? >Business value from data will almost certainly not be limited to one team, one business problem or one area of your organisation. Data (and its value) is pervasive and prevalent in every customer interaction, every system key stroke, every employee connection (...) **Nearly all people in your organization are your stakeholders. This is a unique challenge for data strategies**. Possibly the only other area in the business like that is human resources. (p. 71) ^a513f5 Although, their bias will be different, and you will need different stuff to get buy-in: ![[stakeholder_matrix.jpeg]] ^618bf8 ### Build the case and get investment Look for gold: >Hidden within your organisation are **pots of gold** that can help improve your key business metrics. By applying data-guided thinking, there will be opportunities to increase your revenues, reduce the cost of your operations, build stronger relationships with partners and suppliers, understand and engage your employees and customers, build better products and enter new markets (p. 75) >**Knowing the size of the prize is critical.** Once we know, we need to be able to articulate this alongside the investment required. Many data initiatives are focused on developing data capabilities within the organisation (like data governance, building technology platforms, recruiting data scientists) and not on delivering incremental value. The size of the prize is often a forgotten piece when asking for investment (p. 75) And make a pitch, which includes: - Vision - Business opportunity: with the total size of it, specifying which areas of the business you can address with data. - The problem: preventing you to get the business opportunity now. - The solution - Roadmap - Investment needs - Industry benchmark ### Communication as an agent for change No one-size-fits-all: >Think about the lens you put on what you are communicating. The chief finance officer will need a different lens on the strategy to the chief technology officer. They will need a different lens to an analytics team or the line of department leads. One size doesn't fit all and tailoring the approach, personalising the message and targeting your communication will help to ensure you establish a strong agenda (p. 79) >**Giving people options for how they consume content helps maximise engagement and increases the chances of your message landing and being understood**. Get creative. Use videos, blogs, demonstrations, newsletters, articles, animations, show and tells, Q&A and 'ask me anything' sessions. This is about communicating your message, building credibility, building a culture of sharing and collaboration, getting buy-in and ensuring your strategy is clear and understood (p. 79) ### Establish the Agenda: the breakthrough criteria 1. "Emotional" buy-in from the organization. 2. Strategic backing: success with data needs to be of strategic importance, aligned to the business goals and of significance to the success of the organization. 3. Pots of golds identified. 4. A minimal team ready to "prove value". ## 3. Prove value ### Gaining credibility Start small: > Before pushing hard on further deep investments, it's vital that you gain credibility by demonstrating that applying data to known and necessary business challenges can add value to the business. If your aim is to create a business guided by data, the journey you go on should apply that concept. Use evidence that proves as an organisation you can make this work before you decide to press on. **While you want to aim for the best possible return, it's likely that you'll make small gains and small returns compared to what was previously defined as the size of the prize. This is fine and expected It's what you want, really - small investment, prove you can get a return and create some data points of success**. (p. 83) ### Rapid provision of insights At this stage, the authors warn about some issues: > 3. Unless you have actively **ensured that metric definitions are agreed, you can end up creating more problems than you solve.** > 4. It can be hard to prioritise where to invest time and which metrics to build. > 5. It can be challenging to deliver clean, accurate metrics as the underlying data work required has yet to be carried out. (p. 89) In order not to boil the ocean, focus on the business outcome, even for data engineering work: >All of these 'watch outs' can lead to **big investments in 'sorting out core reporting' with limited value or overall benefit to the agenda**. What's needed is the rapid provision of insights that prioritise the most impactful metrics (...) Even though the objective is to deliver insights, you are still looking to focus on the things that could have the biggest impact on the outcomes of the business. **Focus on the outcomes and actions you want to drive rather than the report / dashboard itself. The destination is the change and not the data product (dashboard) itself.** (p. 90) ### Building your minimum efficient organisation Skills that the team will need: - Business discovery - Product management - Business analysis - Business enablement: "works to ensure that the organisation is ready, educated and able to understand data in a way that helps them make decisions, implement change and create action". - Delivery (similar to the [[🗣️ Down with Data Science]] roles) - Architecture and design - Engineering - Data Science - Analytics and business intelligence - Data management - Assurance - Protection - Adherence: "the work required to ensure the process of assurance and protection is adhered to" And an interesting way of organizing the team from [[Ryan der Rooijen]], CDO at Chalhoub Group: >When I started in my role I was keen to hit the ground running. Therefore, **instead of building a lot of siloed teams centered around capabilities**, eg data engineering, data science, etc. I instead established three pillars focused on outcomes: Assets, Products and Impact. >- Data Assets are tasked with ensuring data quality, architecture and management. >- Data Products identifies opportunities for products and then delivers them end to end. >- Analytics Impact focuses on delivering operational transformation and profitability. (p. 95) ### Communication: the "rule of seven" *Ser más pesao que una vaca en brazos:* > The [[Rule of seven]] suggests that consumers need to hear or see a message seven times before they are likely to take action (p. 106) ^59d09a >The strategy and roadmap for data should be handled like marketing. Don't leave your strategy papers, slides, videos and infographics gathering dust. Get out to the wider organisation and continue the discussions, remind people of the plan, get champions on board with the journey, share progress updates, communicate the successes and what you've learned. Consider starting an internal newsletter. Add relevant news from your industry on how data and digital are being maximised, share stories about what's happening within the company, raise the profile of the team driving the agenda. How about running show and tell sessions in other people's team meetings or grabbing a slot at the next board meeting? What about lunch and learn sessions to demonstrate early wins, data products in development or new insights that have been found? The power of easily shareable anecdotes at this stage is strong - use them brazenly (p. 107) ### Prove Value: the breakthrough criteria - Value delivered - Organisational buy-in - Foundational team - Baseline technology platform ## 4. Scale ### To self-service or not to self-service Again, no one-size-fits-all: >You can make that decision based on the capability and maturity of each individual or team in isolation (p. 124) ### Agility x collaboration = adaptability >Improving data products should be an **always on** activity, not a one-off development project (...) This product-centric approach to data relies heavily on a huge increase in cooperation and collaboration through cross-functional teams (p. 126) ### Set up team for scaling As usual, wait with Data Science! >Unless you have robust reusable data sets, have ownership in place and are progressively using a consistent approach to data management, the hiring of a data scientist is probably not your highest priority. You may hire one or two if you have budget to understand the issues you need to resolve if you want to scale, but it will be more important to ensure you have a data engineer to support their work or make it clear they need to be a generalist (p. 128) And start proselytizing: >...we usually begin with running workshops to engage with key business stakeholders where you can share the message about change, data standards and the rest of your strategy. Your stakeholders should come out of the first session understanding that this is not just a 'data' thing but something that will fundamentally change the way the business needs to adapt (p. 130). Flexibility in the data team people is important at this stage: >Having people with existing skills is great but **if they are not adaptable and willing to learn and change then you need to look at your bench** (p. 131). Have an idea of your "keep the lights on" team: >Undertaking regular assessments on capabilities or people, tools and processes allows you to scale effectively and declutter if cost reduction becomes a priority (p.133). Don't become a cancerous cell: >Scaling is not about you and your team (...) As a leader you need to be the conduit for change and be the pragmatist in the room when others are acting in a selfish and unproductive manner (p. 133). > ### From rigid to cross-functional organisational design On [[Data team topologies]]: >Whichever model you choose [centralized vs distributed], we urge you to use the squads approach to ensure cross-functional, outcome-focused teams working to solve real business problems (p. 141). > ### Value-drive prioritization Very similar to the Priority Matrix (for delegation) of [[📖 How to lead in Data Science#4 Capabilities for leading people]]: ![[A2DAC652-A797-43F5-BAAB-239B55E28FE7.jpeg|300]] >We won't always have all the information required to know if an activity is going to generate value. How can we? (...) That's why we recommended starting with a proof of concept, test or minimum viable version of a solution with a small investment, relative to the overall potential investment (p. 145). ### Sorting out who owns the data The owner of the data? The owner of the process. >When we began to look at ownership for data, it became apparent that we couldn't just assign ownership to data only. We added process, as this gave the context of boundary conditions (p. 163) The authors propose to assign ownership as part of the risk framework of the org. >If your organisation doesn't have a risk framework, you can develop a simple one - creating a list of key risks and mitigating them as needed. When you map a process, you also understand the risks involved (p. 165). ![[C01D2179-2AE6-444B-8CE6-5D7EBC255D24.jpeg|300]] From here, the recommendations of the authors get a bit "fuzzy". They talk about a shared ownership model, with the following (non necessarily exclusive) roles: - Critical data owner. - Discrete process owner: for non-critical data. - End-to-end process owner: manages the related discrete processes owners (so they share ownership with the discrete owners). - Data strategist: accountable of setting policies, strategies, standards...u And, also makes data consumers responsable of understanding their consumption and producers accountable for what they produce. Of course, the authors also ask for everything to be documented (processes, critical data...) And, as in the [[Data Mesh]] principles, Data (team) can't do it alone: >A key part of creating a data-guided organisation is to **federate data services** into the business process. A great place to start is data quality (p. 180). >... as you mature and become a data-guided organisation, you should be developing people capability in various parts of the business through federation of the services and products. If we take the data quality example, **over time you should have poeple in the business who can develop their own data quality rules** (p. 182) ### Scale: the breakthrough criteria > - Working on the top business priorities: the efforts of those working on data products should be focused on the top three to five business priorities > - High demand for data products: you should have the problem of too much demand rather than trying to create demand. The main business units, functions and teams should be represented in your backlog of use cases. > - Embedded collaborative and efficient ways of working. > - Data ownership and management process in place: questions over data ownership and how your core data is managed should be mostly ironed out (p. 184) ## 5. Accelerate >At this stage it's about accelerated decision-making (p. 187) TBH, this chapter seemed a bit vague to me. ### Federated data knowledge > A key success factor for understanding if your organisation has made the transition culturally is the push for more data change (p. 197) There is also a section asking you to consider to sell your data to third parties... ### Accelerate: the breakthrough criteria > ...there is no longer a data strategy; the business strategy and the plan to implement it are amalgamated. Process work is synonymous with data flows and digital enablement. The strategy is connected and focused (p. 200). ## 6. Optimise A short section with some company stories. It blew my mind how Timo Boldt, founder of Gousto (a meal subscription box service) talks about his company: > Timo Boldt markets his company on LinkedIn as *a data company that loves food*. This is a top-down message that *we are data* (p. 204) And certainly data is everywhere in his company: data-driven menu algorithm to optimise the recipes to display, routing of orders, placement of ingredient in their factories, churn prediction models to identify subscribers that will likely stop their subscription... ## 7. Plotting and Tracking your Journey A very interesting (because it is highly practical) chapter. However, it is disappointing that there are no specific hints on how to measure value ([[Measuring a data team impact]]). ### Six pillars of a data strategy #### Vision and value >Key outputs: vision statement, use case prioritized based on the potential business return #### People and culture >Key outputs: skills required, skill gaps identified, target organisational structure, culture change activities identified and articulated #### Operating model >Key outputs: prioritisation framework, chosen delivery and management methodology... #### Technology and architecture >Key outputs: techonology strategy, technology capability gaps, architecture approach. #### Data management A whole book on the topic: [[📖 Data Management at Scale]]. >Key outputs: decision on policies, standards, procedures, regulatory controls, data ownership plus creation of key metric and data catalogue. #### Roadmap ![[4FEF4849-F452-440B-9541-23A5B58F7B27.jpeg]] ### Relationship between data, information, and business strategy It requires a different approach depending on the domain: ![[7CACE9CA-80D5-4D0E-99BF-559AD69E353D.jpeg]] (I guess that the data management of Advanced analytics and Analytics innovation is *expected* to be low, not that it should be low). ### Putting your data strategy together We need quick gains, but also to look quantify bigger opportunities > It's also important to lay out the short-term opportunities that you can crack on with (...) as they pose and opportunity to start making a mark > Try to make conscious and ocmmunicated decisions rather than **sleep-walking into new challenges** that you'll need to clear up later. We have seen organisations invest vast sums of money and months and months on an exercise like this. It doesn't need to be that way. For small and medium sized organisations this should take between one and two months. For large organisations, possibly two and three months (p. 220) ### Measure and monitor progress > In the case of data strategy we really should be looking to measure the following key metrics of success: >- Business outcomes of the data strategy. >- Business outcomes of each data product put into action. >- Costs incurred to deliver those outcomes. >- Time taken to deliver those outcomes. >- Throughput of data products and data product changes. (p. 235) ### The modern chief data officer >When is a CDO not a CDO? When they are a head of data science, head of data and analytics (...) We see a CDO as a business role, accountable on a 100% committed basis for the data strategy (...). If you think about borad level positions, they all have strategic responsibility and ensure that operational efficiency is executed correctly (p. 238) > But does eveyone need a CDO from the start? Have you ever tried pushing something twice your body weight up a steep hill? Without someone to help it is almost impossible (p. 239)