Quantifying Loyalty
A look at the metrics and models used to measure loyalty
Welcome to another edition of Pacenotes, where I dive into the literature and research from a diverse array of fields and try to apply them to social tokens.
We’re continuing our close read of Scoring Points, a fascinating insider’s account of how Tesco’s ClubCard loyalty points system was created. It’s written by a founder of Dunn Humby, the analytics and marketing agency that does this type of work for Tesco.
The Tesco tale is particularly interesting because supermarkets are such an emblem of Britain’s class structure. As discussed previously, where you get your groceries situates you within a whole host of social, economic and political phenomena.
British supermarkets are also locked in particularly fierce competition, so Tesco’s ClubCard system is battle-tested. Anyway, on to the next theme from Scoring Points: measuring loyalty.
This topic is interesting because the ClubCard people came up with several interesting heuristics or mental models to think about customer loyalty. The book also contains some examples of where things worked—and didn’t. Finally, the idea of measuring loyalty also touches on a major theme of the crypto world and Web3: the notion that users own their data, and they’re not trading it for 5% off yoghurt from a supermarket. I’ll summarise some of the stuff from Scoring Points and then try to tie it back to our social tokens discussion.
RFV = Recency, Frequency, Value
The first concept retailers use to measure loyalty is “RFV”, which stands for recency, frequency and value. Here’s what each of those things means:
Recency This is a log of the last time someone bought something from you. As the book notes, this is a deceptively simple concept with far-reaching ramifications. If someone hasn’t shopped with you for a while, then it’s a sign they’ve “abandoned” the brand. If a group of customers haven’t shopped with you for awhile, then an entire segment or demographic might be abandoning the brand.
But it gets worse: Since the supermarket sector is so fiercely contested, the defections from your brand probably mean gains for your competitors. This can create a spiral of defections and gains to your competition that ends at a point of no return: when former customers outnumber your current loyal advocates. At this stage, “no amount of advertising, public relations or ingenious marketing will prop up pricing, new-customer acquisitions, or the company’s reputation.”
Frequency This is a measure of how many times a customer shops with you. Again, while a simple measure, interpreting it to understand customer loyalty makes all the difference. As the book points out, a nuclear family unit might be equally loyal to a supermarket even though they all have different frequency profile, ranging from shopping once a week as a family, to one parent visiting three times a week, to fortnightly online buys.
Value The book defines value here as something beyond the financial transaction performed by a customer (or “basket size” in supermarket jargon). It also includes intangibles that a supermarket might offer a customer as value. The idea is to create a cycle of value between the customer and supermarket: Value begets loyalty, which creates growth and profit, which turns into more value, and so on.
An interesting aside here is the authors assertion about the nature of retail loyalty. They say:
As we have already described, loyalty is not about monogamy in a retail context. Customers may like a supermarket, but very few are exclusive in their affections.
—Humby, Clive; Hunt, Terry; Phillips, Tim. Scoring Points (p. 93). Kogan Page. Kindle Edition.
Loyalty Cube
The concepts behind RFV are refined into a more sophisticated model called the Loyalty Cube. This three-dimensional model has three axes on which to place a given customer, or group of customers. Let’s go through them in turn.
Contribution This is similar to ‘value’ in the RFV model. It’s a measure of a customer’s profitablity. For instance, a loyal customer who mainly buys cooking ingredients may not score highly on profitability, since these are low-margin items. But a customer who comes in infrequently but buys ready-made meals and a bottle of wine would score high on this measure. The key thing here is to distinguish between contribution and loyalty.
Commitment This is a measure of a customer’s projected future value to a retailer. This comprises several components: how likely someone will defect; whether a defection can be prevented by a loyalty programme; and the “headroom” or the future spending potential of a customer.
The notion of headroom is particularly interesting. One way to measure headroom is to calculate the calories represented by a customer’s weekly shop, and correlating it to the average calories consumed in the area. This indicates whether a customer can possibly be buying more from a supermarket in the future or not.
Championing This is an extension of the headroom idea. It’s possible that a customer has reached max headroom (ha!), but what does this mean? It suggests this is an extremely loyal customer. Such a customer could become a “champion” for the brand, converting others and encouraging defections from rivals. Loyalty programmes can encourage this behaviour by rewarding max headroom customers with incentives.
Segmenting and airdrops
The book goes through some very detailed description of the various internal corporate battles that had to be fought to design data collection mechanisms, sift through the data and analyse it, and take actions based on interpretations of the data. Tesco went from analysing individual customers to “buckets” of people. This gave way to “lifestyles” which in turn became “segments.”
It’s probably sufficient to present one highlight from the result of all this data analysis:
The potential rewards of getting the analysis right are spectacular. For example, the analysis showed that one segment of regular, loyal shoppers regularly shopped in 12 out of the 16 Tesco store departments. If each of its members could be encouraged to shop in the other four just once in every three-month period, then Tesco calculated that the additional revenue would be worth £1.8 billion.
— Humby, Clive; Hunt, Terry; Phillips, Tim. Scoring Points (p. 136). Kogan Page. Kindle Edition.
Humby and co. conclude that all the data collection and number crunching works, but not in the way that people might think. It’s not about engineering conditions to coerce a customer into performing some new actions. Instead, it’s about “fractional changes” in the way people shop, nudging them slightly this way or that. The aggregate difference is reflected in billions in new revenue.
Making decisions based on segments makes me think of airdrops in crypto or NFT-land. For instance, and maybe most famously, Uniswap traders and liquidity providers were given free UNI tokens for being early adopters of the system. This is getting into our now-familiar RFV territory: how many swaps you did, how much liquidity you provided, and so on, factored into how much UNI you were given.
But analysing user loyalty using only on-chain analytics is still fraught. For instance, we heard about people with lots of Uniswap accounts (for instance, people who conducted workshops and tutorials to teach people how to use Uniswap) being able to claim large amounts of UNI tokens as a result of the the airdrop mechanics.
Consider the criteria the authors employ for effective segmentation:
- Identifiable Every customer must be part of only one segment
- Viable Segments must be large enough to move the needle on revenue (or other metric)
- Distinctive Each segment has to be different enough from another segment
With this in mind, how many airdrop strategies meet the criteria above? Further, is on-chain data capable of telling us enough about users to meet the criteria above?
Some of these issues are structural: crypto addresses are pseudonymous and no one has cracked the problem of unique identities yet. Therefore, someone with lots of accounts trading on Uniswap is bound to be rewarded for their loyalty disproportionately over someone who traded just as much but with only one account.
But finding the right airdrop strategy could unlock the virtuous circle of first creating value, then rewarding loyal users through the airdrop, thus generating growth, which is re-invested in more value to users, and so on. Just as Tesco tries to divine your loyalty from your preferences in oat milk, so protocols and social tokens should devote their energies to developing airdrop strategies that reward their most loyal users, across a multitude of dimensions.
Privacy and the promise of Web3
The discussion on segments and customer data also leads the authors to a brief discussion on privacy, titled “Big Brother”.
The authors claim that loyalty programmes that try to “enslave” customers won’t work in the long run. It says Tesco does not try to do this because it has several measures in place. One of them is public transparency with customers about exactly what data it collects and how it processes that data.
The authors also say that Tesco uses its data analytics for good. For instance, the data shows that the supermarket should continue to invest in its ‘Free From’ for customers with allergies and intolerances, despite that segment being a small one. It has prompted Tesco to bring in thousands of local products, because the data showed customers preferred a local alternative.
Despite the arguable benefits of all this number crunching, the very notion of billions of data points owned by a single, centralised, entity is anathema to those of us working on Web3. That’s one structural reason why tokenised loyalty programmes are structurally better than the corporate systems of yore. Everything that’s useful is already written to a public blockchain for anyone to see, analyse, or build on.
The argument in favour of Web3 loyalty programmes—or community building on social tokens—is the idea that setting all the customer data free creates a net benefit for all communities. For instance, it’s possible for a new artist to target every single wealthy owner of a CryptoPunk by airdropping incentives to them. This can be done without any data collection performed by the artist.
The team at Unlock Protocol is also working on interesting models around online identity and segmenting. Unlock lets publishers put up paywalls around any piece of content. That content is revealed, or unlocked, through the possession of an NFT in a reader’s wallet. This flips the dynamic of traditional paywalls: Instead of readers being locked in to a lengthy subscription contract, they now have tiny keys that act as bearer assets, opening up different bits of content whenever they need it.
The notion of composable NFT paywalls means that a reader can now have a “portfolio” of NFT keys, paying only for, say, the Euro 2020 live blog but not the recipes section. When you zoom out from this, it’s possible to imagine readers with portfolios of NFT keys representing both segments of content, and the brands they subscribe to. Someone might be identified as a foodie because they have keys to five cooking sites; another person might be identified as a financial professional because have keys from the Financial Times, the Wall Street Journal and Bloomberg.
It would then be possible to start segmenting users based on their interests, expressed through their brand preferences. A personal wealth advisor might be able to airdrop tokens to all the readers of the Financial Times and the Wall Street Journals’ personal finance sections, for instance. Or a crypto exchange might be able to target all the people who carry NFT keys to Bloomberg’s crypto coverage. It would also be possible to construct personas who might be desirable to some companies: For instance, who wouldn’t want to reach the average reader of the FT’s “How To Spend It” luxury section?
What’s cool about all this is that it happens in an opt-in way. Readers choose what keys they want, and what content they wish to consume. This signals something about them to the marketplace that others are free to act on. True, readers can’t reject an airdrop—but that’s not too different from the problem of spam on email. It will be possible to filter out the offers you are most interested in. More importantly, it leaves readers free to opt in to systems, while keeping their data to themselves.
