We’ve announced the shutdown of Cream to our community today. After 24 months of work on the app, this was a difficult decision to make but the right one.
Cream was an incredible ride with countless great memories and friendships made. Unfortunately, it didn’t pan out the way we would have liked.
This article summarises the journey from start to finish: the problem we were trying to solve, the bets we were making, what happened, why I believe it failed, and what I’m doing next. Most importantly, there’s a thank you to those who helped. So many people.
I’ve also written a summary of some of the lessons Cream taught me. Read that here.
If there’s anything here you’d like to talk about, I really welcome it. Reach me on Twitter or my site.
Cream
Cream was a restaurant discovery mobile app that showed diners restaurant recommendations from top chefs & owners. It was live in Melbourne & Sydney.
The problem Cream was solving
Cream started by solving a customer experience problem; searching for and choosing a restaurant takes too long and isn’t reliable.
We dine out 50M times per week (pre-COVID-19) in Australia but where we get our dining information from — User-Generated-Content (UGC) platforms — isn’t as trustworthy anymore. Trip Advisor had a fake restaurant reach #1, bloggers & reviewers get paid for favourable reviews, Instagram influencers are paid to post, Yelp admitted 25% of their reviews are fake, and whether or not it’s true — it’s difficult to find someone who doesn’t believe many food publications are cash for comment.
This low consumer confidence has created long search times and decision-making and is reflected in the generally low Net-Promoter-Score (NPS) for most food search platforms.
Because customers now visit numerous platforms in a bid to de-risk their decisions, venues must invest time and money keeping each platform up to date with little to no measurement criteria to analyse success. New images, copywriting, menus, and the reviews, both real and fake, they need to monitor/dispute/despair over takes a lot of time, money, and depletes them of the physical and mental energy needed to work and run a demanding business.
The job we were doing
Employing Clayton Christensen’s ‘Jobs To Be Done Theory’, product managers like myself will begin by asking, “What’s the job to be done here?”. Just how buying coffee does a different job for different people (ie. a wake-up, 15-mins away from the desk, weight loss, flirting with the barista, getting a daily step count up), what job would they be hiring us to do?
People evolve when they exists in systems that perpetually propel them forward.
(I enjoy this thought. For each system in your life, like work or a relationship, you’ll notice you’ve stayed in those where you’re perpetually evolving for a long time, and you become frustrated and leave those where you're not. Keep evolving, or help others to do so.)
For diners, accessing insider information about the best food connected them to better experiences and to some, a more attractive social standing — knowing the best and latest hot spots is valuable. For venues, Cream’s job was to ultimately get them more customers with less work, but it also provided professional and social credibility — look what top chefs & owners recommend me.
So our mission was clear
Create a more trustworthy experience to discover the best food that provides venues a more effective & efficient way to attract & retain new customers.
We believed giving people restaurant recommendations from top chefs & restaurateurs on a beautiful platform that automated transacting, analytics, and audience building would do that.
Leveraging and concentrating readily-available valued information to generate exponential value for both sides of a market was the name of the game.
Our bets?
- Because chefs & restaurateurs dedicate their lives to great food & drink and their professional reputation is on the line, diners would value their recommendations more than the alternatives and be attracted to Cream;
- More trustworthy content within a great, personalised search experience would yield a more confident consumer who could and would transact more quickly and frequently;
- Venues on Cream would experience better ROI from much less work — the Pareto principle (80% of consequences come from 20% of the causes) would occur — and have more successful businesses.
- Recommendations are sharable from a diner aspect, and the product could even grow internationally as chefs can recommend venues anywhere, thus bringing those venues and their chefs onto Cream to create a scalable growth model.
The second-order effects? A better industry for everyone.
Every action has a consequence and each consequence has a subsequent consequence — a second-order effect. I think founders use benevolent second-order effects as tools to inspire them of what good could happen if they made a dent in the problem space.
One of ours? Transitioning from UGC (User-Generated-Content) to EGC (Expert-Generated-Content) meant only experts are recommending and so venues offering the best quality in food & drink would now benefit most, which at scale would catalyse the industry into getting back to focusing on their original purpose for joining the industry; to provide great food, drink, and hospitality to people — not what stacked shake, pink photo wall, or 4-for-1 they might need.
How we’d make money
Venues were our customers; diners were our users. Cream would generate revenue by creating tools (bookings, CRM) that helped venues monetise this new, high converting traffic for a fraction of the cost, whilst monetising our content that we were successful with already. Capturing 10% of each market, combined, was estimated at $50M/year after 5–7 (really good, optimistic) years.
How we’d grow
We had growth loops built into the product (more on those below). If we grew, we’d invest earnings back into Cream by building better tools for venues to improve operations and lower costs whilst using diner data to create an even better search experience.
Our content was solid and our food shows (how we began) were on domestic & international Virgin Flights already. With some luck, we could package our new, highest performing recommendations as branded tourism content to generate revenue whilst placing it in both domestic and foreign markets to grow and scale our presence.
We were accepted into and completed the RMIT Activator Program in January 2020, receiving $25,000 in capital to go with our own contributions, and began building.
Go time
In 8 months, our non-technical team of 3 (1 full-time, 1 part-time, 1 casual) along with a local development agency designed, built, and filled the app with the tonne of content required with very little, renamed ourselves to ‘Cream’ (the best recs from the best chefs → best of the best → Cream of the crop → Cream) and launched in November 2019 (many lessons learned here on speed to market). The product management, UX, UI, and branding skills learned here were invaluable.
How it went from launch to March 2020
From November 2019 to March 2020, not too badly; we had some traction and trickles of revenue.
Traction
Content
We achieved over 10,000 recommendations Australia-wide from 250+ top chefs & restaurateurs that was growing weekly — our supply-side growth loop was working. Evidence of EGC being higher quality was clear and after refining the way we generated them, they exceeded expectations. Weeks spent creating beautiful venue profiles are etched into my retinas; I’m a sucker for product.
Users
Our go-to-market was simple — measure and learn from our audience of 2,000 who signed up to our pre-release group and our 30,000 Instagram followers, then optimise/remove/focus and go again. We spent zero on user acquisition here because it didn’t make sense pushing unknown, unforgiving users to a product that probably/definitely needs work. Of those, 1,400 downloaded our app and we were growing at 20% MoM.
Metrics showed our MAU/DAU ratio was 23% (not bad) by the end of month three after simple fixes, like when deep linking broke across iOS and Android so links only opened the app’s home screen instead of venue profiles. We eclipsed our combined MAU for December and January in one email after we fixed it. This is just one of many ongoing tweaks we made.
What was working?
- External triggers. If we emailed or posted content on IG (we regularly experimented with different types and followed the data), it got people into the app thus showing product channel fit was beginning to occur.
- Chef Growth Loop. It made sense for chefs to provide recommendations and great recommendations, which invited more chefs and venues to the platform. We were still ‘Wizard-of-Oz’ing this but it worked really well.
What wasn’t working?
- Our Diner growth loop (pictured below); Diners weren’t sharing recommendations they discovered enough with their friends, instead calling or booking or just saving them to their profile for later. Calling and booking is great for venues, but not for user growth.
- Internal triggers — not enough users were thinking to open Cream when they were hungry, only when we sent an email or an IG post.
Revenue
We launched without a specific revenue model (a mistake) but followed the build>measure>learn loop closely. Our initial plan was to run very leanly, relying on strong content partnerships to bring new users to the platform cheaply & quickly whilst using the revenue from those partnerships to pay for overheads until we had enough users to achieve break-even with in-app products. That didn’t work.
We tested bookings first by placing chat/call buttons with 3rd party services on venue profiles as an inexpensive proxy to building our own bookings tech. In month one, over 50% of active users tried to call/chat to a venue so we got booking affiliate partnerships, built the bookings feature, and launched in December. We kept using Facebook Messenger for chat.
Discovering, however, how broken the restaurant bookings industry was for venues and that we need much more scale for them to keep us alive, we wanted to transition away from them as soon as we could and made steps in that direction.
A week-long revenue sprint in mid-January, 2020, led us to choose chatbots after already having them in our roadmap and understanding they could solve both our and our customer’s problems.
Employing OKRs
Back in December, 2019, we discovered OKRs (Objectives and Key Results) and implemented them in late January after listening to Christina Wodke’s ‘Radical Focus’, recommended to me by a close school friend, Ed Morgan; an Agile Coach at Spotify NY (it’s a recommended book for employees in specific areas during onboarding and I highly recommend).
In three months, we wanted to prove Cream was awesome for our customers and users, that it’s an exciting business for us to work on, and had measurable numbers that, if we achieved them, would prove it.
Our next bet
If we could direct traffic to a chatbot that provided great customer service and increased conversions that was cheaper than a human doing the same task, venues would pay for that. Finding users pressed ‘chat’ at a 2:1 ratio to ‘call’ in the app, coupled with price testing that demonstrated this revenue stream would keep us alive, we pulled the trigger.
Contrary to how we built Cream (product-led), we chose a sales-led route for ‘CreamChat’ (unique name, I know) and made 20 sales in the first week at $199/month + set up fee with another 40+ looking promising. When a product solves an acute pain, selling is more simple.
We estimated achieving $40,000 MRR (not extraordinary but OK for a team of three) without hiring any staff in the next six months (September 2020, hence the $10k OKR after 3 months) if our sales & growth loops were solid. Rather than try and get the $40k in three months, we preferred to make sure our product was great first.
We knew that if a product is <$2,000 per year per customer, it has to be self-serve:
- For marketing you can afford inbound sales only (no outbound)
- No sales team and no support; but
- Because it’s not a complex or expensive product you can get conversions on the same day.
A new opportunity
We had finished the RMIT Activator program and the combination of having:
- A product in market with users;
- Revenue; and
- A full team including a CTO
increase the likelihood of being accepted into the Skalata Ventures Seed Program. I was attracted to Skalata firstly for the six-month program that would improve both my and my team’s learning velocity; companies progress faster in learning environments like theirs because of access to capital, press, talent, and creative insights from the dense network of people working on similar problems. Secondly, the $100,000 in seed funding was attractive.
We had applied in an earlier intake and were unsuccessful, but the process of applying for accelerators is unbelievably valuable, Skalata’s was no different (I usually complete Y-Combinator’s application process just to realign and focus), and Rohan Workman and Maxine Lee (who run the program) are people I wanted to surround myself and my team with.
When I shared our progress with Rohan and Maxine since our last application, they told me to apply and we got through various stages to the final 16 when things changed.
So there we were
Acquisition, activation, and retention were showing signs of life, we’d altered our direction after learning and listening to our customers, we’d added a CTO to the team, we were in the early stages of proving revenue with promising monthly media partnerships on digital and radio on the close horizon to boost our audience and revenue, and joining Skalata was a real possibility to breathe more life into us.
COVID-19 hit the following week and most came to a halt.
What happened from COVID-19 onward
The app’s traction flatlined because legally Melbournians couldn’t dine out. We called venues and together canceled all chatbots. 50% of our team had to step away as their regular incomes were impacted, including our CTO, disabling our ability to pivot at all. We were understandably unsuccessful with Skalata.
It was a good reminder that pre-revenue startups feed off positive energy in place of revenue. When this goes it loses oxygen, forward motion, and the fire begins to fade.
Sanjay and I were the only team left and flipped from selling to supporting venues by pushing takeaway on the app — it felt like using a cup to put out a fire.
Were we just unlucky? Sure, a pandemic is an anomaly (every century in fact) and many great businesses struggle in the early days and beyond, but progress stopped so quickly that it had to be more than lack of luck and I felt something wasn’t right.
Thinking and reflecting in the weeks to come, I confirmed this uncomfortable pill to swallow.
An important point to make
COVID-19 was not what ended Cream. This isn’t a ‘woe-is-me’ piece. COVID-19 didn’t help, but it simply accelerated our learnings about the problems the product had. It was a gift, really.
Cream’s problems
I’ve found Brian Balfour’s ‘Four Fits’ model an effective framework to assess products and what I used with Cream.
We didn’t have model/market fit
EGC doesn’t generate the exponential amount of fast, free traffic & virality that UGC does as fewer people are generating content and so cannot get its revenue models (selling attention) working fast enough before running out of runway. Chatbots, along with smart media partnerships to accelerate our audience growth, could’ve paid for the app until it reached break-even scale but we didn’t have the runway to see it out, the market was dead, and apps are particularly resource-hungry — food search alone takes a significant amount of work. The app could’ve been a lead-gen tool for chat subscriptions, but a really expensive and slow-moving one.
We didn’t have product/market fit
Without the free traffic and virality UGC creates, venues didn’t receive enough value to:
- Drive customers to their Cream profile; and
- Proactively share sufficient recommendations needed for us to grow quickly; which
- Created a lot of work for us.
We had solid enough diner retention and conversion metrics and felt we were close to product/market fit with users, but definitely didn’t have it.
Here’s how Marc Andreessen of A16Z defines product/market fit:
“The customers are buying the product just as fast as you can make it — or usage is growing just as fast as you can add more servers. Money from customers is piling up in your company checking account. You’re hiring sales and customer support staff as fast as you can."
We had good usage metrics but didn’t have this.
Discovering the best place to eat isn’t an acute problem, it’s a nice-to-have, and so diners will never pay us to access this information and thus, never pay our bills.
We had product/channel fit but bad timing
With email and social. We worked tirelessly on crafting posts and emails that converted (60% email open rates, 22% CTR) but didn’t have enough money in the bank to take advantage of those early enough and COVID-19 cut our opportunities short. We were about to take begin exciting monthly partnerships in digital print (a monthly ‘Cream of the Crop’ section in the Sunday paper featuring the best recs from top chefs) and radio (monthly segment of us and a top chef talking about the best places to eat). Who loves food recommendations? Melburnians.
If we had have been a web product, our recommendations and chef profiles could’ve been indexed by Google with actions in the product also being indexed, driving us up in the Search Engine Results Pages (SERPs AKA page 1 of Google).
Unfortunately, SEO (Search Engine Optimisation) is a long, compounding growth tactic and it would’ve taken ages to get to page 1 because of how competitive food SEO is — Zomato, Broadsheet, Trip Advisor, etc spend a tonne of time & cash there and trying to compete in a channel where time & money wins isn’t the best growth tactic for an early-stage startup; go CAC-shopping somewhere else!
The best place to hide a dead body? Page 2 of Google.
We didn’t have channel/model fit for the app
Discovery plays sit on the low end of the ARPU<>CAC (Average Revenue Per User and Customer Acquisition Cost) spectrum and so rely on low CAC channels like SEO (Search Engine Optimisation) and virality for growth, which we just didn’t have — this weakness in our product was a core learning of ours.
An example of our potential growth loop is below — it mirrors how Pinterest grows and continues to.
The hard decision
At this point, we had two separate products; Cream App and CreamChat.
Ultimately, we made the difficult decision to stop Cream App because:
- The above problems outlined with our different fits;
- We were out of money, mobile apps are expensive to run with no MRR, hospitality was looking dire and the market was reportedly shrinking by 15%;
- Food discovery isn’t a big enough opportunity for investment — especially with the level of traction we had; and
- Helping people discover the best food & drink isn’t an acute enough problem. There are bigger problems to solve and better ways to help people, especially in the hospitality industry.
As for CreamChat
We returned to customer discovery in May 2020, rediscovering that fixing margins in hospitality is the hair-on-fire problem — that means using tech better. With our knowledge and proof of sales with chatbots, we’ve been building our own high-quality chatbot that doesn’t need human intervention at all: it’ll help venues increase revenue whilst lowering wage costs by not having to pay a staff member to do online customer service or conversions.
Let’s chat if you want to hear more about how this conversational AI of ours could put an extra $22,000 a year in a venue’s account.
What’s next
“Success is going from failure to failure without loss of enthusiasm.”
— anonymous
I feel wiser from the experience realising how little I know and how much there is to learn but incredibly energised for the next challenge.
I was looking for new opportunities that leverage and improve the product management, growth, strategy, brand marketing, and UX skills learned from my time on Cream. See this and other work on my site I’m constantly tinkering with.
Outside of this, I’m following a “river of inquiry” — a term coined by Blackbird and Samantha Wong — in virtual healthcare:
Since July 2020 and with my sister, Dr. Lucy Nijam, I began Joey Health: a virtual healthcare project. Going to the GP is inconvenient, costs too much, and our healthcare system can’t scale quality. So we’ve redesigned the GP consult to deliver 75% of it with software so you can access a GP from anywhere and over time, reduce the cost of a GP consult by up to 70% by using technology instead. Chat to me if you’d like to hear more.
Special thanks
I’ve learned more from this experience than any other in my life.
I want to deeply thank those who helped me throughout Cream’s journey. There are too many and I look forward to paying it back in spades in the future across your projects and my own.
The 150+ chefs & restaurateurs who helped and supported us from the beginning. Definitely not confined to and in no particular order: Nick Kutcher, Jesse Gerner, Karen Batson, Matt Lane, Michael Bascetta, Casey Wall, Manu Potoi, Shannon Martinez, Alex Ghaddab, Nic Coulter, Lisa van Haandel, John van Haandel, Kelly O’Loghlen, Jason Lui, Dave Mack, Iain Ling.
Sanjay De Silva for being an exceptional co-founder since hatching the idea in 2015 over a coffee in Trunk Diner with Barun Chatterjee and I, and an even better friend since we were 13yrs old in country Victoria.
Barun Chatterjee for being a great co-founder with forever-timely counsel, skill, and wit.
Liv van Haandel for her excellent, contagious co-founding energy and for being the best product evangelist I can think of.
Darcy Tuppen and Tessa Mansfield-Hung for being the best crew members we ever had, and then some.
Josh Nijam for his unflinching guidance, Dr. Lucy Nijam for her unending support, and Mum & Dad for their unwavering confidence in me.
My close friends for listening, for sharing, for letting me interview them, for being early users, and for the positivity.
Loris Campanile, Matt Hayward, and Callan Delbridge for engineering our dreams into reality and being great people throughout the ups and downs.
Stephan Brown for his infectious, proactive engineering mind, friendship, and ongoing work on our next product.
Derek Blank for his technical work, guidance, support, and walks around Albert Park Lake.
Special thanks to Adam Schwab, Simon Holland, Jane Newton, Rod Hamilton, Megan Flamer, Rohan Workman, Maxine Lee, Marc Hammond, Luke McInnes, Erika Geraerts, Sam Chapman, Susie Robinson, Jodie Crocker, Leanne Clancey, Rebecca Bellan, Hugh Stephens, Josh Brown, Michael Boyd, Amy Whitfield, Saxon Quinn, Tim Williamson, Ed Morgan, Will Egan, Brendan Wilde, Ashish Kumar, Christopher Doyle for your help, your support, and for your ears.
The RMIT Activator community especially Paz Pisarski, Julie Stevens, Adam Seedsman, Sukanya Banerjee, Matthew Frith, and Matt Salier for their undying support and provision of both a space to call home and opportunities to take.
Any questions or feedback, I really welcome them. Reach me on Twitter or at thomas dot nijam at gmail dot com.
