Groupon recently announced its fourth CEO transition in three years. For technology investors, it’s a story so gruesome – like a bad car accident on the side of the highway – that it’s hard to look the other way. On a long flight back from Rio, I fell down my usual rabbit hole… could Groupon be a good business?
In one of my favorite of his speeches, Warren Buffett explained putting on a 20% position in Coca Cola stock in 1989:
“At the time we bought Coca Cola just two years ago, we bought 7% of the company. We paid $1 billion, so we were essentially paying $14 billion for the whole thing. Now, if Philip Morris were to buy Coca Cola that day, they would have paid $30 billion. But Coca Cola wouldn’t have sold it for that. And you wouldn’t have sold it for that. The company was actually repurchasing stock at the time, so, in effect, they’re buying for you. They’re buying out your partners, at 50 cents on the dollar or less, which is a magnificent sort of business…” ”It’s an easy business, there’s no doubt about it – you can sit down in five minutes – I mean, everybody here understands Coca Cola. There are 660 million eight-ounce servings of Coca Cola products being served around the world today. So in effect, we’ve got a 45 million soft drink business with our 7%. We think of businesses that way. I say to myself ‘just increase the price a penny and that’s another $450 thousand per day for Berkshire.’ I mean, it’s a nice sort of thing… Now, you tell me whether you think there’s a penny, worldwide, of price flexibility per serving of Coke. Well, the answer is: you know there is.” — Warren Buffett, 1991 speech at Notre Dame”
Most people don’t realize this, but Coca Cola didn’t just invent soda – they pioneered the modern coupon. Legend goes that in 1888, Asa Candler – a pharmacist based in Atlanta – bought the recipe to a ‘tonic and headache remedy’ for a mere $2k; $70k in today’s dollars. Candler re-branded the drink to the famous “Coke” moniker and launched the first ever coupon marketing campaign, putting slips of paper into magazines and mailboxes that were redeemable for a free $0.05 Coke.
The rest is history: from 1894-1913, 1 in 9 Americans tried a free Coke. At a $0.05 sticker price, the campaign cost $425k – less than $15 million in today’s dollars – to bootstrap a brand that today generates $40 billion in annual sales. As for Buffett, his $KO shares, diamond-handed for 33 years and counting, have returned 32x (vs 23x for the S&P500), generating $10b+ of outperformance for Berkshire shareholders. M-O-A-T.
A century later, coupons are still here, evolved into the digital age. A very high-level overview of the industry in the US today:
- 10b digital coupons are minted each year, of which roughly 7% or 700m are redeemed. The industry is highly-competitive with 2.7k sites competing for consumer eyeballs and merchant partners. The top five networks – Rakuten, Groupon, Slickdeals, Honey, Ibotta – have saved users $80b+ since inception.
- 180m Americans, or 2/3rds of American adults, redeem a digital coupon every year. We assume user behavior follows a typical 80/20 Pareto curve, implying there are 35m power users (redeeming digital coupons every three weeks) and a long tail of 145m casual users (once per year).
- Given the publicly-reported financials of major platforms, we estimate annual GMV is around $30b (after discounts), implying an average ticket size of $45 . Typical commissions are 10-20% , implying a revenue pool of $3-6b.
- Print coupons are a different beast: the total number of printed coupons is an order of magnitude larger than digital, but redemption rates are more than order of magnitude lower . Net-net, digital coupons outpaced physical coupons for the first time in 2020.
While the top five web2 rewards platforms have saved consumers >$80b, their aggregate value created for shareholders rounds to zero at best. The posterchild, Groupon, has lost >65% of its users, >80% of its revenue, and >99% of its stock price since its heyday:
As a pure-play publicly-traded company Groupon is the most visible of many such cases. LivingSocial, the Lyft to Groupon’s Uber, eventually sold for (*checks notes*) $0. Rakuten paid $1b to enter the space in 2014, and despite growing GMV by 2.5x since then, generated $63m of operating profits last year (a 6% yield after eight years, at a time when treasuries yield 4%+). Ibotta raised $245m and is pivoting to power first-party rewards ecosystem for huge merchants like Walmart, i.e. aiming to take home a smaller slice of hopefully-bigger pies. Honey top-ticked the market and sold to Paypal for $4b at the height of the fintech bubble, and is now fully-integrated into Paypal’s payments ecosystem. The data points in one direction: web2 rewards networks are not viable standalone businesses.
The more interesting question is, why not?
The proximate cause is poor unit economics. The average Groupon user – and remember that simple averages obscure the Pareto nature of user behavior – redeems 1-2 coupons, worth $35-70 in GMV, before churning; of which Groupon takes home 30%, or $11-21, as gross profit.
Groupon’s cost to acquire a customer was >$25 during its growth years (+40m net user adds from ‘09-’12), meaning they lost money on the average incremental user. The company pulled back aggressively on marketing beginning in 2018, bringing CAC down to $10 and LTV/CACs into positive territory (1-2x), but at the cost of letting the majority of active users churn off the network (-30m net users churned from ‘18-’22).
|GROUPON: UNIT ECONOMICS||Metrics||Source|
|How long does the average user stick around?||5-10 months||30% retention (Q3’22)|
|(x) How often does the average user make an order?||Every 4-5 months||2.7x per year (FY’22)|
|How many orders does the average user make before churning?||1-2||(calc)|
|(x) How large is the average order? (based on total cart size or GMV)||$35||$36 average (FY’22)|
|How much does the average customer spend before churning?||$35-$70||(calc)|
|(x) What is Groupon’s gross profit take-rate? (based on % of GMV)||30%||FY’22 10-K|
|How much gross profit does Groupon earn per average user? (LTV)||$11-$21||(calc)|
|(%) How much does it cost Groupon to acquire a user? (CAC)||$10-$25||S-1, 10-Ks, TC|
|LTV / CAC||0.5x (during growth) to
2.0x (during runoff)
In any business, it’s extremely difficult to make money when growth is a matter of two steps forward, one step back. Viewed from the supply-side perspective, 3 out of 4 deals that are onboarded sell less than 10 units per month, despite 14m active users visiting the site/app.
Again, the more interesting question is, why?
We count three key factors:
- 1) Indiscriminate supply growth. Groupon’s initial model was based on deeply discounted local services. For users the value proposition was super clear: half-off. But they could only grow so fast within that niche. Groupon eventually expanded beyond the original Deals product, launching an Offers product (15% discounts on always-on inventory) and a market-rate product (cash-back rewards). Groupon also began allowing merchants to self-onboard inventory, which quickly grew to represent the majority of new inventory, and expanded into physical goods, which added a layer of logistical complexity. All of this drove indiscriminate supply growth which, while it helps hit quarterly numbers, bungles user value prop and impairs the network in the long-term:
“Our current Deals have too many restrictions… Not every merchant wants to run a deeply-discounted Deal and even for those who do, a deeply-discounted Deal may not make sense at all times… By providing a full catalog offering to our customers, we believe we can unlock customer purchase frequency and drive billings growth.” — Q1’21 investor deck “With self-service, North America continues to leverage our improved platform to help our merchant partners to launch and help their deals and also to bring new merchant partners to our marketplace. In fact, in September, over 75% of our new inventory onboarded through self-service.” — Q3’22 earnings call
- A related point is the quality of inventory listed on the marketplace. Groupon spent a decade in the public markets, with a peak sales force of nearly 6k, without implementing outcome-based sales incentives. This drove artificial inflation of their supply-side, with inventory that provided little value to users and effectively served to transfer wealth from shareholders to the sales team. Per the company’s own admission from (now-former…) CEO Kedar Deshpande:
“Sales reps now have very specific target merchants. They are responsible for acquiring by vertical and geographic location… This is departure from our previous structure when the sales reps had the freedom to bring on any merchant with any deal structure and still make their sales quotas. Second, we are aligning the sales compensation to the performance of the deals and merchants that bring on the marketplace. In other words, they will be compensated based on the output of their deals, not just the number of deals launched or merchants acquired.” — Q4’22 earnings call
- 2) Weak local network effects. Despite local merchants driving >90% of Groupon’s revenues and gross profits, historically the company operated with a top-down management structure that hinders local network effects. Local managers lack the autonomy and therefore the credibility to form strong commercial relationships with local merchants. Groupon finally acknowledged the power of local network effects, announcing a pivot to a “hyper-local” strategy on their last earnings call:
“Our new merchant acquisition strategy will be centered on fulfilling consumer intent by acquiring higher quality inventory that we know consumers want, instead of acquiring more for the sake of more. To do this, we are implementing a hyper local merchant acquisition approach, which is focused on acquiring the right merchant in the right category, the right location and with the right deal structure. In the first step of our hyper local approach, we are prioritizing merchant acquisition in our top 5 U.S. markets… Once we have proof of the concept, we expect to expand this hyper local approach to other North America and European capitals…. To do this, each of our top markets will be led by a city CEO, who will be responsible for leading inventory acquisition and management and the performance of overall market.” — Q4’22 earnings call
- 3) No user ownership ⇒ no user loyalty. The biggest roadblock to Groupon’s unit economics is their 70% annual user churn. The company has tried many things over the years (except proper incentives for its sales team…), most recently issuing G-Bucks – site credits that are non-redeemable for cash, expire in 180 days, and whose terms can be arbitrarily changed by Groupon. In a world where Americans are forced by economic circumstance to question the solvency of their banks – and even the solvency of their nation – the likelihood that corporate Shrute-bucks (watch the link—you’re welcome) manages to successfully entice users asymptotically approaches zero:
“We are also testing ways to leverage Groupon incentives as a means to get new customers to browse and engage with our full catalog of inventories. In our initial test, we are giving Groupon Bucks to customers when they purchase our full-price market-rate inventory. Our hypothesis is that by providing this additional value via G Bucks, it will incentivize customers to purchase our full-price inventory and get them back to Groupon later to use their G Bucks and make another purchase.” — Q3’22 earnings call
How does a decentralized protocol stand up to the above criteria? Pretty damn well, at least for certain types of inventory:
- Programmatic incentives can reward participants proportional to the gross profit generated for the network, or with some other reward scheme if appropriate (analogous to proof-of-coverage), and supply constraints can be managed with staking/slashing mechanisms, protocol rules for acceptable inventory, or by decentralized governance.
- Local network effects are strengthened, since there is no need to pay a large take-rate to shareholders, a larger share of value accrues to local networks; analogous to DeWi unlocking value for ISPs/landowners.
- User ownership is a core feature of tokenized models – instead of “G-Bucks”, users get inalienable utility / economic / governance rights to an open-source protocol, to avoid situations like these:
We’re not the first to the idea of web3 loyalty/rewards networks – we count at least a dozen related projects that have collectively raised $85m in seed capital over the past year – but note that most attention has gone to brand loyalty/membership models vs Groupon’s core use case of helping local service businesses drive traffic to their storefront:
- NFT-based loyalty platforms, including companies focused on enterprise-scale brands – e.g., Hang (raised $16m & partnered with Budweiser, Asics, BleacherReport), Forum3 (raised $10m & partnered with Starbucks), and Glow Labs (raised $4m & partnered with Neopets, Barbie, Forever21) – and others focused on medium-sized brands – e.g., TryYourBest (raised $10m), Kalder (raised $3m), and Flaunt & Taco.
- Other web3 teams building adjacent ideas include Cub3 (raised $8.5m) & Narratic building loyalty programs for web2 customers on top fungible tokens rather than NFTs; Spindl (raised $7m) and Sesame Labs (raised $4.5m) building attribution & marketing platforms for web3-native customers; and Spatial Labs (raised $10m) & Blackbird (raised $11m) building verticalized protocols for fashion/media and restaurants.
- Finally, a handful of networks have gotten to scale in their respective markets around the world and could have a path towards bootstrapping a decentralized rewards network from their web2 business. Kard in the US (raised $23m) connects 10m+ fintech app users to its merchant rewards; Drop in Canada (raised $74m) serves 5m+ users; Twid in India (raised $12m) has 40m+ registered users and CashKaro (raised $36m) serves 2m users; and ShopBack in Singapore ($390m raised) serves 35m+ users and does $2b+ of annual GMV (more than Groupon).
Designing the Groupon Protocol
“Groupon Protocol” means a protocol for local services business to monetize spare capacity. We understand Groupon Inc is/was involved in many other business over the years, which perhaps makes the protocol’s namesake confusing. But we like it so we’re keeping it. The protocol is explicitly not designed for physical goods (which require a different moat—logistics—which crypto is not yet well suited for). It’s not meant a cash-back, loyalty/rewards, or loss-leader business. It’s meant to do one thing only: enable local service businesses to fill up their space capacity, quickly, at a small-but-still-meaningful marginal profit.
The protocol needs to do five things: 1) onboard merchants, 2) onboard deals, 3) acquire users, 4) process payment, and 5) resolve disputes.
- Onboarding merchants. Web3 Groupon’s major advantage over its web2 cousin is that it can offer merchants direct ownership/control over the protocol. This comes with the major disadvantage that it’s non-trivial to filter out the “bad” merchants (aka Byzantine attackers) when all the protocol sees is a public key. The paths for mitigating this are: 1) trusting off-chain info (KYC), or 2) requiring a financial contingency (stake).
- Onboarding deals. Groupon’s moat is based on its proprietary access to a scarce resource: deeply-discounted deals at high-quality local merchants. Any inventory that doesn’t fit the criteria, including low-quality merchants, small discounts, or just plain fraud, must be filtered out to protect integrity/value of the network for users. Deeply-discounted deals are particularly interesting because they are ephemeral in four dimensions – that is, users must be at the right place at the right time to get value from it (similar to cellular coverage).
- Acquiring users. This section should really be called “retaining users”, since that’s the much more difficult half of the growth equation (e.g., Groupon pays <$10 to acquire a new customer, but 70% churn within a year). Luckily, tokens massively open up the design space in ways that can drive retention and therefore growth. Protocol developers have already proven they can cut churn in half with a well-designed airdrop, and there’s so much whitespace left to be explored.
- Processing payments. Payments are trivial for protocols built on open programmable rails… once users have a wallet. The hard part is getting merchants/users a wallet for the first time, but thankfully wallet infrastructure is leagues ahead of where it was just a few years ago and will make another step-function leap with EIP-4337.
- Dispute resolution. Resolving disputes is closely related with managing fraud. The typical dispute happens when a user buys a coupon (paid in advance), yet the merchant fails to provide the service, so the user demands a refund. Since revenue is split between Groupon and merchants, both parties must agree to fully refund the customer, or one party has to take assume liability for the losses. Web2 Groupon, for reference, has 3k+ complaints and is rated 1 out of 5 stars by the Better Business Bureau.
At it’s core, the protocol is pretty simple:
- Merchants ‘list’ deals permissionlessly by depositing % of face value into a vault.
- Users ‘purchase’ deals by sending funds into the vault.
- Merchants and users co-sign a transaction at the time of service (e.g., QR code scan). This begins a stream of funds from the vault to the merchant’s wallet over  hours.
- Users can pause the stream anytime during the  hours by sending a “complaint” transaction. In the case of a complaint, the protocol assigns a “case worker” to make a determination off-chain.
- Either the merchant or user, whoever is determined to be correct by the case worker, gets paid/refunded in full. Case workers are funded by protocol fees and are subject to appointment/oversight from protocol governance.
In practice, a bunch of non-core mechanisms would need to be added to help bootstrap and secure the network:
- A huge amount of tokens should be allocated towards rewarding users. The highest value users are those that come back to the protocol repeatedly and try out new merchants each time. This behavior should be most handsomely rewarded in all sorts of creative ways. On the other hand, repeat users who use the platform to always buy from the same merchant are more likely to be fraud-related (since authentic users would purchase from the merchant directly), and should be de-emphasized in the rewards scheme since it adds less discovery value to the network. Users who drive virality through referrals are also highly valuable and can be extremely effective when properly incentivized.
- Deeply-discounted inventory has an interesting dynamic: merchants only want to list at deep discounts near expiration, but once they reach that point, they are willing to price nearly as low as marginal costs. This creates logistical challenges for protocol design, but also the opportunity to lock in proprietary supply via local network effects: the protocol should find ways to incentivize cohorts of people that have structurally advantaged knowledge about local businesses’ spare capacity – like concierges or club promoters – to list inventory on behalf of businesses.
- There needs to be an incentive to curate high-quality deals on the marketplace. In practice, this means deciding “how we decide which deals make it to the top of the page?” One option is to rank by merchant reputation score: this would provide an even stronger incentive for merchants to join the platform quickly (since the reputation score drives both capital efficiency and top-of-funnel visibility), but would eventually prevents new merchants from competing. Another option is to rank by % discount: this would theoretically encourage merchants to list the steepest discounts, but unfortunately is impossible to enforce since the denominator – ie the MSRP for any given item – is unknowable. One potential workaround is to have merchants self-report MSRP to calculate both the collateralization ratio (how much capital they must put up) and the % discount (how much revenue they’ll make). This way, setting a higher MSRP moves deals higher on the results page but requires more capital at risk, and vice-versa setting a lower MSRP requires putting less capital up but lowers visibility.
- A large portion of merchants will reject any protocol that requires putting capital at stake. One solution is to allow merchants who undergo KYC to vastly reduce or eliminate their staking requirement. This approach adds centralization – since effectively the protocol is taking on credit risk to merchants based on centralized KYC – but we think it is the right tradeoff to drive growth in the early days.
- The real-world costs of hiring fraud dispute specialists puts a lower bound on the size of deals that the protocol can support without operating at a loss. The protocol will likely need to enforce this minimum deal size in order to disincentivize small disputes. For example a $20 minimum deal size, 10% protocol fee, and 5% dispute ratio implies a fraud budget of only $40 per dispute.
Finally – ten pages later – since no investment memo is complete without a ChatGPT mockup in 2023:
 Groupon reported $1.8b in gross billings, $0.6b in revenues, $0.5b in gross profits, and -$172m of free cash flows on a consolidated basis in 2022; their business is ~70% US by gross profits. Rakuten Rewards reported $10b+ of gross billings, $1.0b in revenue, and $63m of operating income in 2022; Honey reported $0.1b in revenue at the time of the Paypal acquisition (2018), growing 100% YoY. Ibotta claims $15b+ sales per year through their app.
 Groupon charges 15-25%; Slickdeals 1.0-8.5%, Honey 0.5-10%, Ibotta 3-10%, Rakuten <a “https://www.datafeedwatch.com/blog/selling-on-rakuten#:~:text=Rakuten%20seller%20fees&text=a%20listing%20fee%20of%20%240.99,on%20the%20items%20you%20sell.”>$0.99 plus 8-20%.
 Per Vivvix 2022 Trends & Insights report, print coupon distributions were 125b (vs 10b digital) and redemption rates were 0.25% (vs 7% digital) in 2021.