Why Strategic Plans Aren’t the Best Plan For Data Science

Author(s): Jonathan Burley

Originally published on Towards AI.

Tech and data science winners treat strategy differently

Data Science as an exercise in building new products and companies retains the same pitfalls as any other business process before it: clinging to old, formerly successful business practices and understandings past when they are fit for purpose.

In 2015 Chris Outram (founder of OC&C Strategy Consultants and my former boss) wrote “Digital Stractics: how strategy met tactics and killed the strategic plan” arguing that the “digital” online channel and its so-called Big Data allowed smart B2C business to empower their organizations with faster tech-enabled workflows and that for these smart businesses, the traditional strategic plan was being replaced with something new: stractics.

Stractics can be roughly understood as the result of the actionable cycle of feedback-conclusion-action shrinking so much that it fundamentally renders the traditional strategic plan ineffective. Instead of a plan with concrete steps for many months, stractics is a strategic vision underpinned by excellent tactical execution rather than preplanned steps. Understanding stractics in more detail, and how it interacts with data science, is the focus of this article.

2015 was a time when Data Science was a nascent term (we thought more of big data than the jobs and processes to make use of it), but having moved from strategy consulting into data science and business leadership my experience has been that the same fundamental forces that made “Stractics” pre-eminent for B2C are also ensuring that stractics follows tech and data science into all businesses.

Therefore, when data science is involved, the classic strategic plan has become one of these old business practices that hinders far more often than it helps. And depending on where you are building AI products you may encounter more or less resistance to changing business practices to make the best use of data science.

The purpose of this article is to lay out the background for where businesses are coming from, why that can destroy data science value, where businesses need to get to, and advice for doing so.

Former best practice: statistics and strategy

The best practice in the business world used to be the strategic plan. The major stakeholders collated information from across the business, had a statistician calculate the “so-whats” from the data that were some combination of statistically significant and meaningfully impactful on the business¹, and then the C-suite decided what endgame was optimal for the business. Reaching that endgame was achieved by strategic plans and each was commonly a multi-year effort to achieve an important benefit.

After a strategic plan was in place the next challenge was implementation and observation. Simplifying for the sake of summary, this is setting up the correct environment for the tactics that action the strategy. The typical process for good implementation is leadership setting consistent messaging such that everyone down the chain feels empowered and incentivized to act in aid of the strategic plan (this is important: details of how to actually achieve the plan are controlled by staff and middle-management). Good observation is defining metrics of success for the plan, assessing them on a monthly/quarterly cadence and changing implementation details as required.

This practice worked well: based on some Bayesian reasoning, a plan with high expected return was selected, accountable goals laid out, and updates scheduled based on the speed at that the organization could handle information and train staff.

The problem with traditional strategic plans

However, for any current B2C business these processes are too slow. Company-scale tactical decisions are no longer limited by the speed at which staff can be trained but by how fast a computer can be updated. For most businesses, there is a large chunk of value that is addressable by technology (e.g., B2C website product recommendations & UX), and a laggard business will lose that value to faster, more responsive competitors.

Traditional strategic plans were disrupted in the majority of public B2C companies in the 2010s due to the online channel and the consequent rapid iteration on customer preferences. The winners of this disruption innovated more than just their websites. They also changed other processes to bring their whole operation closer to the speed of online: for example, fast-fashion retailer Zara had sophisticated development and logistics to bring products to market faster than competitors. Online grocery was enabled by websites, but the winners innovated with new staffing, product replacement guarantees, dark stores, automation, and narrow delivery windows.

(Expect the same today with data science and related technologies, many winners’ competitive advantage will be how they update their industry to leverage data science’s faster feedback-conclusion-action cycles rather than just mastery of the data science itself.)

Those examples of successful disruption may sound like classic strategic plans, but they were not. Any attempt to plan the path from start to finish of the transformation would have been wrong. What worked was a clear strategic vision of the end goal and teams empowered with the ability to make the best tactical decisions in response to the changing competitive landscape.

To put it another way, a traditional strategic plan couldn’t survive first contact with reality in the 2010’s B2C landscape with its faster feedback-conclusion-action cycles. With faster actionable updates, fewer strategic plans remain useful for more than a fraction of their intended lifespan.

This is not to say that business strategy is a dead art. If anything, the rapid changes to operating conditions make the strategy toolkit more important than ever, both in terms of picking an endgame that is resilient to disruption and in terms of evaluating whether a sudden market change should alter the endgame vision or strategic plan.

For example, Uber and Convoy are classic platform plays for vehicle transport (of people and goods, respectively) with the hypothesis that this platform ownership will be highly valuable when autonomous vehicles proliferate and an intermediate game plan to grow the platform by being a more efficient use of human driving capacity.

But the nature of how to achieve those intermediate plans is nothing like traditional implementation.

Current best practice: data science and stractics

As automated systems proliferate, they enable decisions to be implemented across the business with perfect compliance. You can have stakeholders review data and adapt actions on a weekly basis or even decide upon the measurement criteria and have an automated system continually alter its own actions.

Returning to our examples, Uber leadership could review data and decide that, say, underperforming airport pickups cause churn and tweak software (potentially within hours) to create a surge of available cars at peak airport times. They can also monitor if exposure to improved airport pickups better retains customers on the Uber platform.

For any business, having automated processes updated at this speed necessitates changing the rest of the implementation around them. More than ever leadership must provide consistent messaging and a clear vision for staff to follow, but training will focus on flexibly following the process of rapid updates rather than focusing on the specific steps of a strategic plan. The space between tactics and N-year strategic plans has thus collapsed. The most value will be realized from a strategy that continually updates in response to tactical data.

It is this merging of strategy and tactics that defines “Stractics”.

Data-science-driven stractics often means more autonomy being given to computers. This is both a blessing and a curse, as the less human “common sense” is involved in decisions the faster runaway disasters can occur (the problem with computers is that they do exactly what you tell them to…). This sensible fear of disaster has caused businesses to be slow in adopting tech- and AI-enhanced decision-making, but ultimately avoiding technology adoption is a plan whose best result is to “lose slower” to your competitors rather than a plan to win.

In 2022 we began seeing investors come around to this view, with private equity firms (eg. Silver Lake) spinning up internal data science capabilities to both assess the sophistication of data initiatives in their targets and to build up the data science hierarchy to create value during the investment period.

A smart business now operates with a strategic endgame in mind and a vision of how they will be differentiated and defensible in their market, but the strategic plans to get to that endgame are now the more fluid stractics.

Conclusion & Recommendations

AI technologies and their underlying data systems enable decisions and actions to update across a business on a weekly (or faster) cadence.

The dynamics behind AI & data science changing business are similar to how in the early 2010s “big data” and online ordering transformed B2C companies. B2B companies should follow the examples set by the B2C winners of that time.

Therefore, strategic plans based on reviewing statistical summaries every few months are no longer fit for purpose. Data comes in quickly enough that we must often act weekly. And that difference in scale is enough to be a difference in kind; the fundamentals of how we plan and execute strategy must change if we want to maximise value.

The updated fundamentals are to have a clear vision of where you need to get to, why that is important, and having a team with the capacity to get there and adapt their tactical choices to changes along the way. Multi-year strategic plans predicated on quarterly tactical updates will be outperformed by operating with those faster tactical changes and less concrete implementation plans.

Based on my experience, here are actionable recommendations for how to succeed:

Recommendations for Businesses:

• Understand your exposure to data science disruption: For the digital disruption in the 2010s it was trivial to understand which companies were disruptable. For data science, it is not. Conduct a review of your business (and competitors) to understand what can be disrupted with data science / AI and the approximate revenue at stake. Then decide if you will invest in data science.
• Review company vision: If data science is going to be vital over the next few years, does your company vision align with that? For larger companies, the implementation will be done by staff several steps removed from senior leadership, and you need the guiding principles to guide those staff to the right decisions.
• Build or buy your data science hierarchy?: Data science is not a magic model that overhauls your business in a fortnight. Good data science relies on foundational datasets and capabilities that you need to either build internally or buy externally. Decide which is best for you, and it will determine the personnel you need to succeed.
• Personnel is policy: Most companies will not have the internal skills to create or manage data science. Executive sponsorship of these fledgling skills is vital. Organization, culture, and talent acquisition may need to pivot to find and retain the right talent. Dedicate an appropriate leader to spearhead this change and, if needed, hire the staff who will perform the day-to-day work. Those staff will be both data scientists and people who transform other parts of the business to take advantage of what data science offers. Your policy cannot succeed without the right personnel.
• Walk before you run: The AI initiatives that make headline news are built on extensive foundations, and without those foundations, a (probably very expensive) AI initiative is unlikely to succeed. Build your data science hierarchy iteratively and start with descriptive analytics that allows you to make smarter, positive ROI decisions today². After that, diagnostic analytics, then predictive models (etc.…) to make a full product range for internal and external users. The staff who stick with you through this process will be your superstars of tomorrow.
• Commit to your vision: Fledgling data science capacity is often pulled in a dozen different directions (or asked for “quick data requests”) that create lots of activity but little long-term value. Pick a small number of projects that match your long-term vision and commit to them. You will create more value and reduce the churn of frustrated data scientists. A great executive sponsor should enable the data science team to prioritize work on these valuable projects.

Recommendations for Data Scientists:

• Your best action depends entirely on the business structure you are embedded in. Probable actions are upwards management to give leadership the right context for decisions, advocating for tightly scoped projects with provable returns, advocating for a long-term view of the product range, and defending your time against low-value requests. The key transformation will be when a senior sponsor views you as a provable money-making asset for the business.
• If you are lucky enough to be in a data-sophisticated business, then your boss and mentors will have better recommendations than I can provide remotely.

Footnotes

1. Readers who have been in these meetings will be aware that the staff responsible may not literally have been statisticians, nor is it a guarantee that statistical tests were literally applied. Regardless, the important feature is that data was gathered and reviewed with some assessment of its reliability and beliefs updated in light of that information — leadership of smart companies has always performed something akin to Bayesian reasoning to make their strategic plans.
2. Iterative, profitable improvements conflict with the “moonshot” policy popularised by FAANG, particularly Google, that going big and ambitious as soon as possible generates the best ROI because the successes are huge, and even the failures are valuable learnings that form the foundations of the future products.
   I say big tech did that because of a competitive environment that does not apply to most businesses: FAANG could take risks and absorb losses in hopes of returns on a massive addressable market, and knowledge that any incidental advance in architecture and algorithms offered them a competitive edge.
   The majority of companies experience those benefits to a much lesser extent and are better served through the iterative development of sequential, profitable data science initiatives (exceptions being companies of huge revenue like Walmart, or with technology as a key product like Snowflake).

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI