Personalization — where to start?
How to Start a Personalization Program
Personalization programs are complex, multi-year transformations. For this reason, they require substantial leadership mindshare to break the inertia of the status quo, but also a deliberate sequencing of actions and investments. This paper is a practical, experience-driven guide on how to start, and how to prioritize the most critical decisions.
If I were to force-rank in terms of criticality, the two most pressing questions are invariably around value and process:
· What’s the size of the prize?
· How do we get started?
To address these queries we decided to draw from parallels, by analyzing successful, large-scale personalization programs, and collating both the themes that are specific to a specific organization, and lessons learned that apply more broadly. Concretely, how did they size the opportunity? How did they get started?
We will first describe the diverse approaches taken by organizations such as Starbucks, Allergan, a retailer, and a global airliner when they launched their successful personalization programs, subsequently provide a framework to inform the “how to get started” questions.
Learning from exemplars
Starbucks: “Scale and repeat” leads to quick impact
By 2015, Starbucks was already offering personalization in both physical and digital channels. Back then, like now, you can fully customize your drink. The barista puts your name on the cup, and calls you when your order is ready at the counter. Starbucks Rewards, the star-filled reward and loyalty program, offers an extra incentive through an earn & burn mechanism, seamlessly executed within a digital ecosystem that includes app-based store-locator as well as ordering and payment functionalities.
This success begged the question: how can they pursue even deeper personalization?
The first target was the application of advanced analytics to the already wildly popular weekly emails that Starbucks sends out to nudge consumers towards incremental purchases. That success derived from challenges and games aimed to reward a behavior with bonus stars, a virtual currency convertible into free drinks or food. For example, a customer could be nudged to “buy a Latte, a breakfast sandwich, and a hot tea and win 50 Bonus Stars.”
The challenge was to take this incentive-based personalization to the meaningful individual level. The upside of this hyper-personalization is that you can truly tailor the message for an individual customer, i.e. using the three best products for her. The downside is that traditional marketing processes and channels are inadequate to manage this level of complexity and versatility. A combination of few hundred SKU’s and multiple bonus levels means that even the “three best products” approach has millions of permutations and requires hundreds of thousands of different email copies rather than a few standardized templates. An additional challenge is the onus to prove that this deeper personalization will augment and improve an already well-oiled marketing steamroller in a cost-effective manner, without disruptions or PR blunders.
To pull this off, Starbucks embraced a phased approach — which we refer to as “scale and repeat” — built on two guiding principles: limit the tech burden, and prove the value before making significant investments. The first foray into deeper personalization in 2015 took barely four months from ideation to execution, i.e. from launch to performance measurement). It consisted of three hyper-personalized email campaigns sent to a smaller but statistically significant subset of the population. Once the pilots proved the “incrementality” of a hyper-personalized version of the existing segmented tactics, the mantra has been “scale and repeat.”
· Scale is conceptually simple. You extend hyper-personalization to every campaign, not just three; you address the full population, not a subset; and you use all digital channels, not only email.
· Repeat means to replicate the pilot à scale journey to other areas. A good candidate for Starbucks was the in-app ordering feature. Starbucks personalized the digital experience by developing a carousel of relevant products, similar to Amazon, to complete your mobile order. It expedited the selection, because your most coupled items are visually surfaced, and encourages menu exploration because the most relevant new products matching your tastes are rank-ordered and presented once you swipe the carousel.
The recommender carousel is a strong example of non-incentive based personalization, following a tested development cycle: start small, prove the value, then invest in scaling and hardening the product.
Starbucks has invested over two years into its personalization effort, with an unrelenting focus on measurable lift. The initiative is credited with generating over 8% of top-line sales within the loyalty program. Living this journey end-to-end allowed us to observe the decisions that drove the success. The first material decision revolved around the development of a complex AI platform that, to date, embraces the latest paradigms of algorithmic-driven marketing. Starbucks needed a custom platform tailored to its unique reward program, but acknowledged it was not in the business of building tech. Instead, it spun off the IP into a multi-tenant NewCo in which they retain equity participation.
The second critical decision was to build an organization dedicated to personalization within its loyalty group. Its pilot à scale à repeat roadmap relied on six-month milestones. Every 6-month pod was responsible for bringing well-codified experiences to market, and the subsequent pod would build on the previous one, focusing on new experiences and behaviors. Teams were co-located into a single space (aptly called “Personalization Lab”). They were cross-functional in nature, fully dedicated, and led by a marketing VP with a strong analytical background. And well-caffeinated, of course.
Allergan: The efficiency of building bigger first
Allergan is primarily recognized as the company that developed and distributes the wrinkle-defying injection Botox, one of the most recognized brands in medical aesthetics. Less well known is the fact that consumers (i.e. the end users that visit a provider for an injection) can earn and burn tangible discounts through Allergan’s reward program. Brilliant Distinctions (BD), the reward program, is focused on the end consumer but mostly administered via providers. The next step — similar to Starbucks — was hyper-personalization of incentives towards such consumers. Both organizations fully embraced personalization and built the tech around it, but Allergan pursued different choices, and built a program with greater scale and complexity during early stages than Starbucks did.
In January 2018, Allergan’s long-term ambition was clear: focus on data, build the infrastructure and the technology to harness the data, and then leverage the new capabilities to disrupt and re-invent marketing, starting with the medical aesthetics business. Over the course of the year, Allergan launched a number of digital ventures aimed at boosting acquisition (Spotlyte), discovery of and access to services (Regi) and incentive-based personalization (Allergan Data Labs, ADL in short).
In contrast to Starbucks’ journey (pilot - scale - repeat), Allergan’s blueprint called for a tech and analytics tour de force from day one. This required a longer upfront build. While Starbucks would launch skinny pilots and then scale the tech, Allergan’s tech was developed at scale from inception. While Starbucks went from ideation to execution in four months. Allergan did not even make go-to-market a priority until month 6, placing significant emphasis on “production-ready code.” It dedicated the first three months to building data pipes and to setting up the cloud-based production environment and the core analytics engine. The engine has three common elements: a co-located data layer, an insight layer, a predictive layer around consumers’ behaviors, engagement and preferences. Not your run of the mill pilot.
Allergan launched two waves of hyper-personalized bundle offers and re-activation campaigns, measured against a control group. Early results proved that personalized incentives significantly outperform mass or segmented marketing, with tangible benefits in terms of efficiency (dollars generated per each dollar invested in marketing) and bottom line (return, net of discounts). Early campaigns didn’t require further data engineering, but rather led to re-thinking the MarTech ecosystem, because the analytics engine could power faster variations of marketing campaigns than the execution channels could support.
Allergan’s philosophy also differed from Starbucks in terms of investment, organization, and ownership of IP. Allergan’s investment in tech and analytics was more front-loaded, which is arguably a more efficient allocation when leadership alignment is strong (i.e. more comfortable with longer lead-times to in-market execution). At peak, Starbucks had over 70 FTEs dedicated to personalization while Allergan had around 25 FTEs. The organization in the first months was mostly designed around data engineering and analytics, sitting next to the original marketing organizations. The more substantial organizational disruption coincided with the first wave of marketing execution (around month 6), and led to the formation of Allergan Data Labs as a stand-alone unit. Allergan went outside to bring in a talented VP with acquisition and personalization experience in the MedTech space to lead the charge, build a stronger digital team, and fully leverage the analytics workbench. ADL is now the center of excellence that powers the analytics behind new waves of tailored, personalized experiences.
Finally, all the intellectual property (IP) is internal. While ADL feels like an agile-friendly startup, ownership and development of algorithms for personalization remains in-house, according to a two-phased roadmap. The first step is for ADL to centralize the insights from other digital ventures (Spotlyte and Regi); the second step is to expand analytics-driven personalization beyond medical aesthetics.
Global retailer: Pursuing personalization within a larger program
In other large-scale programs, personalization was not the catalyst for large digital transformation, but rather the second or third wave of a broader innovation agenda. In our experience, such initiatives usually reflect scale, maturity, or simply readiness. The most recurring argument is that the organization needs to build the foundations first before it undertakes a more complex endeavor such as personalization. The risk, however, is that the “build phase” might never end completely, leading to costly delays and stretched timelines.
A global retailer (Retailer X) has re-written the playbook on how to embrace an ambitious multi-year transformation, with personalization being a pillar, but not the primary driver. In its journey, it has identified four areas of disruption around analytics, and built both the organization and the technology around these areas . Instead of pursuing everything at once, the organization has staggered the execution along an engineering roadmap: every use case needs to build and fund the technology and processes for the next one. By the time it launched personalization, Retailer X had already launched cash-positive initiatives, and developed a number of relevant assets: a centralized cloud-based environment, a common data schema, strong data automation, and a support engineering team to enable new analytical applications.
The first personalization use case mirrored Starbucks’ nimble approach. It aimed to launch a small pilot of personalized email incentives, improve known tactics, and then scale only when uplift is statistically significant. Since the IT team owns the infrastructure at scale, scaling a pilot has more to do with processes and marketing execution than with core engineering work. This has allowed Retailer X to launch several pilots in parallel with a relatively small personalization team of around 10 FTEs. In our experience, this lean set-up is more the exception than the rule, and mostly hides the fact that the engineering and infrastructure work took place under a centralized organization that serves a broader analytics agenda (e.g. pricing, supply chain, etc.). It is also a rare instance of a complex personalization program that is not the sole occupation of a dedicated VP in role.
From an IP development perspective, this retailer has embraced the path of owning only the critical assets, but buying the rest. This is common practice when there is a strong online component, and the incentive is to integrate off-the-shelf solutions. For example: install third-party customer data platforms (CDP) and basic product recommenders, then integrate them with existing workflow engines with proprietary business rules. The core in-house development is around experimentation frameworks (test and learn, reinforcement learning) and more advanced recommender engines and decision models. For instance, Retailer X, has developed a remarkable suite of predictive algorithms to serve the most complex forms of hyper-personalization: surfacing a product that completes a prior purchase, the best new product that could be of interest, a product more amenable to be purchased full-price, and so on. The company continuously updates and enriches the suite as new open-source algorithms are brought to market, either by testing a stronger algorithm (e.g. a more accurate ensemble model) or by embracing a new technique altogether (deep learning instead of machine learning approaches.)
A Global Airliner: An additional layer of complexity
Starbucks, Allergan and Retailer X have each followed a tried and trusted approach to embracing analytics for personalization: first optimize within the tactic (e.g. right price, right product), then optimize the choice of tactics based on context (e.g. send a discount or a reminder offer today?), and then finally optimize a whole sequence of tactics. I have written extensively on the topic here.
The global airliner (we’ll call it Airline X), has taken a different path, with a successful 12-month implementation of an advanced AI system that optimizes the correct sequence of marketing touchpoints. The intelligence is focused not just on the individual tactic (what is the best route and discount combination for each customer?), but rather on optimizing the rank-order of subsequent tactics: is sequence A-B-C-D-E better than sequence D-C-A-B-E?
The decision to tackle the most daunting analytical problem head-on is sensible under a number of conditions. The first condition is to have enough content to cycle through. Airline X meets the condition because it has a vast range of offerings — including own-brand and third-party credit cards, various insurance products, miscellaneous merchandising and leisure-related products — that goes far beyond air transportation. The second condition is that the fruition channels (Email, Web, App) are mature enough to accommodate the level of personalization — individually tailored timing, channel, and content — produced by the AI platform. The third condition is that there is enough historical data around customer-level response to the different marketing messages — alternatively, that the organization carves out enough time for experimentation to overcome the cold start problem. In other words, the AI platform needs prior data to understand what “good” looks like (possible but usually unlikely), or a process to create that experimental data (often cleaner but more time-consuming).
Meeting all three conditions is a testament to an analytically mature organization. Creating no-regret marketing content and augmenting the MarTech eco-system can be accomplished in months, but creating longitudinal experimental data to optimize entire sequences or journeys has no finite timeline — models converge successfully in months (and they did) but the learning process never ends, and most organizations are not comfortable with that burden. This consideration is not an argument for complexity at any cost, nor a statement that personalization is a distant accomplishment years in the making — it’s not, and it wasn’t for the Airline X. The key difference is the extent to which the sequencing matters.
Starbuck’s gamification does not necessarily require sequence optimization. Customers expect to play a game every week, and thus a simple rotation suffices as long as the weekly game feels relevant and personalized. Differently than Airline X, Starbucks therefore did not require years of experimental data to develop and reap upsize financial benefits from a cutting-edge AI system. In Allergan’s case, the sequence matters, but to a lesser extent, because most aesthetic treatments are regulated in terms of frequency (90 days need to lapse after a filler). In the airline business, the context matters (think pre- and post-flight incentives), but relevance of communication during a purchase hiatus matters even more. Outside out frequent business travelers, capturing the attention in spite of infrequent flight purchases is like honing a dialogue, which is a sequence of words.
Airline Xs execution strategy has been less idiosyncratic than its analytical roadmap. It launched a pilot in eight months to prove value, and only then invested in scaling and hardening the solution into a “continuous improvement” steady state. The IP rests in the proprietary engine that optimizes the sequence, whereas all the other components of the MarTech stack (tagging, workflow manager, etc.) are integrated from mainstream solutions. Again, we find the notion of cross-functional, co-located teams, with around 90 FTEs coming together into a mini organization within the organization, split into different squads: “marketing technology”, “marketing and content”, “engine build”, “engine operations”, “measurement” etc.; this has been indeed one of their largest programs, blessed by broader alignment and support: jointly sponsored by Exco members representing airline and loyalty, with senior stewardship from analytics.
Distilling a common recipe — what we can learn from others
Personalization programs are ambitious and rewarding at the same time. As with many digital transformations, there are commonalities but also idiosyncrasies driven by industry, geography, technical maturity, and competitive environment. As I stated at the outset, our conversation with executives on the topic of personalization eventually lead to ROI (what’s the payoff) and execution (how do we begin?),
Yes, personalization programs are worth it
Data on this first question is somewhat sparse, but fortunately, it is favorably consistent across different dimensions. From an engagement standpoint, personalized touchpoints are more relevant and timely, manifested in increases of 20 to 30% in open-rates or click-rates.
From a financial standpoint, we consistently record top-line improvement ranging from 3% to 8%, driven by a significant uptick in marketing efficiency. This is a combination of 30% to 100% improvement in net incremental revenue (return from marketing, net of discounts) and improvements of 20% to 100% in marketing efficiency (dollars generated per each dollar of marketing spent). Ambitious programs can span 24 months, representing an investment of $25-$50 million depending on the initial level of sophistication and the level of parallelization. We have led smaller and still successful programs running for 12 months, with an investment of $12-$15 million. A typical break-even window is between 12 and 18 months, which means the roadmap is self-funding, and generates a generous and ongoing upside after year two.
Short-term financial impact is a very compelling lens, but not the only one for viewing the benefits of a personalization program. Personalization builds long-term competitive advantage in the form of increased consumer engagement, stronger stickiness (i.e. reduced churn) and a more rewarding culture of experimentation.
In short, personalization is worth it.
Execution: Common themes and approaches
Personalization programs are marathons, not sprints. The level of investment required to unlock the potential puts pressure on the organization and can potentially result in delays, half-hearted attempts, or inertia altogether. In our experience, some execution elements are common to practically all programs and organizations, even adjusted for industry, technical maturity, and competitive environment. At the same time, when it comes to designing the very first steps of a personalization journey, we find that some decisions remain company-specific.
We propose a framework that subsumes the common themes and key decision points. The list below is a high-level picture of consistent themes or patterns, derived from direct observation of successful and mature programs. In our experience, these themes should also inform new programs, and become the basis for the desired blueprint, regardless of the stage of maturity.
· A dedicated organization: The personalization programs develop a mini-organization within the larger organization, with a defined perimeter and reporting structure (usually within the broader organization, but sometimes as a separate entity altogether). This structure exists mostly to guarantee a level of independence in terms of decision-making, funding, issue escalation and performance measurement. But it also reflects the fact that personalization requires niche and scarce talent. Also, separate reporting allows for flexibility on titles, compensation, and career progression.
· Strong leadership at the top: This means a VP of Personalization fully dedicated to the program and with clear performance targets in terms of financials and/or engagement metrics. Successful programs enjoy very senior sponsorship, often emanating directly from the C-suite. A typical reporting structure leads to the Head of Strategy or the CMO. Tech and Analytics are enablers, but not the right home for personalization leadership.
· 100%-dedicated resources, cross-disciplinary in nature and co-located whenever possible: The complexity of personalization demands that resources with very diverse expertise maintain “one job.” In other words, they do not split time across other commitments outside of personalization. Full dedication does not translate into commitment ad libitum. Niche resources could be 100% dedicated just for a couple of months and then return to their original organization when the need subsides. Typical examples are API developers, UI experts, graphic designers, who are all contributors with a very clear activity and timeline. In our experience, co-location tangibly enhances collaboration across the diverse talent pool dedicated to personalization.
· Agile rituals and ways of working: Personalization is about ongoing improvement, experimentation, and focus on measurable results. The nod to the agile rituals is just one part of the equation. The organization and its leadership also need to embrace an agile approach for the go-to-market by developing a level of comfort with uncertainty (e.g. directional plan for what will be built in two months, but not a full spec sheet) and experimentation (e.g. the notion that some approaches will fail, while others will succeed.). Furthermore, it means comfort with “giving analytics time to learn”, and not just in the machine learning sense of things, but also allowing cross-functional teams time to hone the solution and learn from early-stage approximations.
· A cadence of development in “waves”, with each wave funding the next: The imperative is to manage complexity, and to compartmentalize the effort in time-boxed “waves”. Each wave drives end-to-end execution of select use case(s), and defines start and end-dates for resources dedicated to personalization. The breakdown into waves allows the company to prioritize work with two criteria in mind. First, each wave should continue to build and/or expend core capabilities (tech, data, processes), and second, each wave should deliver measurable financial lift to fund the subsequent wave. It is not uncommon for most efforts to start with a “Wave 0” dedicated to set the rules of engagement and plan the subsequent journey.
Decisions or approaches that are more idiosyncratic in nature are mostly influenced by factors such as internal culture, heritage (or legacy systems), the level of technical prowess, and the investment philosophy. We have clustered these “decision nodes” into three core pillars
Sequencing
The notion of a “wave” approach entails a coordinated sequence of use cases or experiences to bring to market. What constitutes, in practice, a good use case for Wave 1? We suggest a definition of use case as a desired behavioral change expressed in terms of tactic, context, and channel.
A good example of a use case could be something like: encourage catalogue exploration (the behavioral change) via complementary product recommendations (tactic) displayed when on the check-out page (context) of the e-comm platform (channel).
When it comes to influencing behaviors, some organizations see personalization as an opportunity to innovate and start fresh. They introduce novelty and freshness in one of the three elements of a use case (e.g. new tactic, a new triggered response, a brand-new app, etc.). For Starbucks, personalizing existing segmented tactics meant sticking to their guns: start with improvements of established tactics in wave 1, expand the scope only in wave 2 (new channels: the app) and 3 (new context: recommendations at order check-out). For Airline X, an algorithmically-optimized sequence of touchpoints was a net-new experience and approach, which also prompted to focus on the “what”, in terms of what content is ultimately relevant for the consumer.
Another key decision is around the number of experiences to develop concurrently. In our experience, less is more; one or two distinct use cases is the sweet spot. It is not a hard requirement, and mature organizations with strong technical foundations can successfully parallel-path three or four use cases in market. Discussions about embracing 10 or 20 use cases imply that the definitions are misaligned and/or unrealistic.
How to build the foundations
The ideal personalization use case delivers a fresh experience while building new assets for the organization. For example, developing personalized retention initiatives also builds a single view of the customer (also called Customer DNA or Customer 360), a measurement framework, and a flexible delivery mechanism. These are all foundational assets.
One decision node is about how much to invest in building foundational assets right from Wave 1. In several instances, the organization elects to “start lean,” prove the value first, scale only what works. The first use case pilot can be in market after as few as two or three months. While this approach tends to create some throwaway work (e.g. scrappy code base that needs refactoring in Wave 2), in general it is preferable for creating internal momentum and justifying a larger investment down the road; Wave 1 funds the journey for Wave 2, Wave 2 funds Wave 3 and so on. This approach represents roughly 80% of our experiences, with Starbucks a notable example.
The remaining 20% elect to build more lasting assets right from Wave 1, leading to a more fleshed-out pilot around month six or seven. This is typically a sign of strong leadership alignment that considers the personalization stack a no-regret investment from inception. Allergan exemplifies this. It built out a solid data hub before launching personalized campaigns.
Another decision node is about “ownership” of those foundations, the proverbial build vs. buy quandary. Especially with technology, some of these decisions are influenced by legacy systems, internal capability, CAPEX vs. OPEX preferences, and so on. When it comes to personalization, the priority is more about defining one’s core IP, the unique “secret sauce” that differentiates one’s flavor of personalized experiences from that offered by the competition. Each organization should strive to define, develop, and enhance that core differentiator, leaving the rest of the personalization stack to an integration of off-the-shelf solutions. Starbucks ended up custom-building its AI platform, because the gamification aspect of its tactics was so unique and proprietary that no other solution in market could power it. Retailer X has smartly integrated a suite of recommender engines in a proprietary ensemble that makes it unique and differentiated because tailored to the fast-fashion industry.
Organizational themes
Building multi-disciplinary teams under strong leadership has been a consistent success factor across industries and geographies. But there are many ways to go about it.
The first key decision is around the organizational placement: where should the personalization team sit? Who should the head of personalization report to? In many instances, personalization sits under marketing and its leader reports to the CMO or the Head of Strategy. That is the case for Starbucks, Airline X, and many others. In rare instances, personalization is very much an analytical powerhouse, with technology driving innovation. Retaler X is such a case, with analytics overseeing much more than personalization. There are other outlier scenarios where personalization becomes a different entity altogether, operating as a mini-venture, with a higher degree of independence. Allergan Data Labs (ADL) is an example. We consider Retailer X and Allergan to be (successful) exceptions.
The second key decision is around talent acquisition/retention. The sheer diversity of skill sets required to develop personalized experiences often means that the organization needs to tap the market to fill gaps. Data engineers, machine learning engineers, and digital marketing experts are just a few examples of scarce profiles at the core of a personalization program. The quandary for the organization is whom to hire vs. and whom to contract externally. We suggest two defining criteria:
· Don’t hire for the build (a transient effort). Instead, hire to run the program (an ongoing effort) — see article “Building an Analytics Organization for your Personalization program”
· Don’t hire talent you cannot retain. Instead, preferably partner with small local boutiques or develop new business models to attract niche profiles
Tying back to the previous theme, we emphasize one important exception: talent focused on core IP (the secret sauce) should be developed in-house as much as possible. That makes intuitive sense, yet we witness several organizations that have entirely outsourced marketing execution to external agencies. The sweet spot is likely in the middle.
The chart below recaps the three pillars, highlighting no-regret approaches common to successful personalization programs, and idiosyncratic decision nodes that are unique to each organization.
Finally, organizations that do business in several countries/ continents face the decision of where to start first. This is really a sequencing decision but strongly tied to organizational and foundational capabilities. We see an equal split between two extremes: one extreme is to prioritize by expected impact, which usually suggests launching in the biggest market (e.g. US, Germany, China), especially when leadership wants to prove value before scaling. The other extreme is to prioritize based on capabilities, i.e. start where the best setup is available (stronger data hub, richer experimentation, more mature marketing, an existing loyalty program, etc.). The option to focus on the least developed geography in order to build its core assets is a clear outlier. Somewhat obvious remark: all else being equal, sample size dictates the minimum requirement for execution: if it is too small to measure impact with statistical significance, don’t start there.
***
Personalization programs are fun and exciting journeys — they require a high degree of competence, planning, resources and coordination. They are also, by definition, highly idiosyncratic. Thinking ahead in terms of correct sequencing, foundation-building and people/org themes is a first step towards unlocking the full potential. Ultimately, they revolve around doing things differently — whatever the starting point, embrace continuous, progressive and incremental change. Managing that change is indeed the most salient ingredient in any digital transformation — hence the 10–20–70 rule.
In BCG Gamma we are passionate about algorithms and technology at the bleeding edge of innovation, yet even the most complex technical deployments can only be successful when supported by people and business processes.
Good luck on your personalization journey, and keep asking yourself the question: what will we do differently this time?
Silvio Palumbo
BCG — Managing Director and Partner
Analytics lead for Personalization
