Unlocking Growth: Estimating LTV of a 2-sided, co-dependent Market Place

There was a quote about on-demand economy that I read from a Sherpa Ventures deck a couple of years ago, that I found funny yet very insightful—

“On Demand Economy connects our village past to our economy future ”

While I’m typing this, my Postmate is about to arrive with my food from CurryUpNow (swear by their Kathi Roll), I get a push notification from Uber asking me to rate the last ride and I just signed up for Handy because I need help to install a new light fixture (hey, I’m an Electrical engineer, what about that?), and I receive a text message from Kiwi asking me to answer a couple of questions when I haven’t even signed up yet... Phew!

Hang on, the Instacart guy is at the door.

Why’s ODE important?

Everything that we’ve done in the past — hiring the locksmith around the block to fix locks on a new door, to employing a nanny who came highly recommended by the neighbor who lives across the street, is now available at our fingertips with every channel amplified several times over. We have a multitude of options when it comes to finding a service for our needs (food delivery, transport, electrical, plumbing, even waiting in the line at a restaurant!) and a treasure trove of feedback on each option from people who are separated from us by 3, 4 degrees, yet made trustable because they were up-voted by other reviewers…

Why’s it critical to understand the important elements of an ODE?

Products and services in this genre typically employ a dynamic marketplace, where supply (an uber driver for example) is matched up with demand (rider) at preset, established terms. These terms differ from time to time, influenced by fluctuations in either supply or demand.

Growing a marketplace of this kind comes with a set of challenges. There are distinct areas within the growth funnel that are all equally important:

1. Supply acquisition funnel— Organic, Paid, Referral, etc
2. Supply on-boarding — New user on-boarding, lapsed user re-onboarding, etc
3. Supply engagement, retention post on-boarding
4. Demand acquisition funnel — Organic, Paid, Referral, Incentivized, Retargeting, etc
5. Demand first-time-user-experience (FTUE)
6. Demand retention post installation
7. Conversion / Monetization & feeding back into the growth funnel

Each of these are complex models on their own. Demand retention, post installation for instance, would need a deep understanding of the source from which demand came from (was it from a paid targeted campaign, or a incentivized push that received a lot of users who weren’t likely to stay anyway). Each of these ‘types’ of cohorts would retain in a specific manner, and understanding that upfront and building a feature set that flexes itself around these dynamic changes in the marketplace could be the difference between staying alive or sinking.

Great, what is the single most important tool to build to prioritize growth?

That said, there’s one model that rules them all. It is the user LTV model, and in itself is a powerful tool that informs the Product Manager about the most important thing that matters to the business now. The tool helps identify which aspect of the growth funnel needs most attention, and allows PMs to craft nifty strategies to grow their product (example 101: showing users the value of the product, and then opening up the gates to refer the product to all of their friends, fuels viral growth).

An LTV model for a 1 part economy is quite simple. The gaming industry is a perfect example.

LTV of a player = Area under the daily retention curve (integral between limits 0, # of days that the team is confident a converted user will stay) X average revenue per daily active user. 

Simply put, it is # of days a player is expected to stay X average $ spent every day. The primary variable of this model is uni-dimensional, i.e, player retention. Different ‘types’ of cohorts will have different LTVs because their user retention curves will vary. For instance, a cohort from a highly targeted, paid acquisition campaign will likely retain much better, and convert at a higher rate, and potentially spend a lot more, thereby leading to a much higher LTV. A cohort that is driven by an incentivized campaign, will lead to sign ups from users who are reward-motivated and less likely to stay longer than it takes to claim the reward, thereby leading to lower LTVs. So long as the LTV is greater than the cost of acquiring the user from that particular source, the business model sustains itself, and the product has the potential to grow.

Sounds simple, why can’t we do the same thing for an ODE?

An LTV model for a 2-sided marketplace is far more complicated. There are multiple dimensions, inter-related to each other. On one hand, there might be a lot of demand, and very little supply, in which case the business has to aggressively recruit supply (at a certain cost per acquisition); on the other hand, there might be a lot of supply but not enough demand, in which case they’ll have to go after demand acquisition (at a certain cost per acquisition, albeit, cheaper than supply acquisition, not $0).

In either case, there are lost marketplace transactions (read: rides), that mean lower monetization opportunities. Ultimately the business has to understand the LTV of each additional unit of supply to be able to quantify the maximum they can spend on user acquisition for each additional user they on-board in the break-even scenario.

Ok, let’s start estimating LTV for a 2-sided ODE, shall we?

This is my attempt at simplifying the problem to the easiest ‘executable’ model that can be taken to the next level if required. I must admit upfront that it lacks the sophistication of machine learning models that companies have built around answering this type of question, but it answers basic questions that can help teams prioritize their roadmap for sustainable growth in a new territory while positioning themselves to be competitive. At a minimum, hopefully, it gives us enough information to appreciate just how complex a 2-sided ODE is and how startups are navigating their own journeys tackling the problem of growth.

Driving Variables

1. Supply Retention Curves: This essentially is the rate at which ‘Supply’ (example: Drivers) return to the marketplace. These will be used to project Daily Active Supply Units.
2. Demand Retention Curves: This is the rate at which ‘Demand’ (example: Riders) return to the marketplace. This is used to project the number of Daily Active Demand Units. Retention of demand is dependent on the availability of supply itself. It is easier to simplify it as the blended cohort retention of all ‘types’ of users, irrespective of (1) Source of cohorts (2) Lack of OR surplus of supply at different time periods in their life time. We can make the model more accurate by learning more about the correlation between demand retention and supply fluctuations to make demand retention dependent on supply factors. Simply multiplying the retention curve for the cohort with a factor that is driven by supply fluctuations will make it co-dependent on supply retention.
3. Marketplace Transaction Dynamics: Rides/Rider per day and Drives/Driver per day. Again, simplified to be constant every day. This can be applied to the food delivery business model, or ‘handy’ esque on-demand contracting. Upon maturity of the business model, these are variables that are likely to become stable/predictable. Transaction velocity is critical for continued customer satisfaction and products have been known to regulate this, via price increases to ensure that paying customers have the best user experience in times of short supply. Conversely, businesses have experimented on the price elasticity of their customers.
4. Channel Information (Organic/Paid): This is an understanding of the number of users that different channels source on a daily basis. Installs organically driven from the app store, or organic driver sign ups on a daily basis, or paid referrals, are examples of this. The more detailed we can get on this, the easier it is to understand which channels are more influential in inflecting growth.

How does the model work?

The objective is to estimate the maximum amount that the business can spend on acquiring one additional unit of supply, while maintaining its break-even position, i.e, all things being equal, how much revenue does an additional unit of supply bring to the business. The model is built on the following:

Static Data

• Supply: Organic Installs per day
• Demand: Organic + Paid User Acquisition + Other Sources into one static data constant
• Marketplace Dynamics: Rides/Rider/Day, Drives/Driver/Day

Changing Parameters

• Driver Acquisition Cost — this includes joining bonus + cost of on-boarding + first week retention, etc
• Drivers Acquired/Day — this includes the # of drivers on-boarded via paid acquisition every day for the launch period.

Constraints

• Driver Acquisition Cost — not to exceed an upper limit ($3,500 in this case)
• Drivers/Day — not to exceed an upper limit (300 in this case)
• Final Margin — within a narrow band of limits. Ideally in a break-even case, this is $0.
• Maximum Drivers in new market — not to exceed an upper limit (5000 in this case based on transportation industry reports, etc)

Objective Function

The model (uploaded here) solves (via GRG non-linear Solver) for maximizing overall revenue, constrained by keeping the margin as close to $0 as possible, to calculate what is the maximum that can be spent on acquiring each additional unit of supply. In other words, the LTV of an additional unit of supply. The model projects

• Daily active riders based on rider retention and rider growth
• Daily active drivers based on organic growth plus automatically computing additional driver sign-ups required to maintain ride fulfillment % at an economically viable rate
• Rides fulfilled and revenue generated per ride
• Lost revenue opportunity due to lack of supply / slower supply growth

With the assumptions on static data and constraints from this report, Supply LTV computes out to $2,700. The model predicts an EOY daily run-rate of $0.5 M, or a monthly run rate of $16M, which was approximately what Uber did at the end of 2014 for the month of December.

Takeaways

By simply building a bottoms-up business model, we can quickly observe the following:

• Stretching the spend on sign-ups and on-boarding will eat away at margins; Reducing costs, or increasing Driver LTV via network effects will build a competitive advantage: From the model, we can see that driver on-boarding/sign ups is a huge catalyst to growth. It is also the biggest puzzle in the business model. While it is extremely expensive to acquire new drivers, every additional one adds $2.7k to top line revenue. Any optimizations around reducing the cost of driver on-boarding after recruitment, such as a mentorship program, or revenue sharing with referrals will help put more $s to use for recruiting. At certain points, the business has to send customers away, because it cannot scale up supply to meet demand, as can be seen from D213 onwards.

Paid driver acquisition to meet demand becomes too expensive, thus having to turn away customers

• Supply growth will be the biggest revenue growth lever and also the toughest problem; Look at other high-engagement industries such as gaming for boosting supply retention and engagement: Driver Retention is another catalyst to growth, and can also become a potential problem. Playing around with the model, you can see that if driver retention curves drop 100–200 BPS @ D180, driver LTV drops down 12% or $324. Optimizations around driver feedback, making them feel part of community and being cared for, bonus programs for high engagement, weekly return starting bonus, etc are ways to increase supply retention. The gaming industry is a great place to look at for techniques to improve user retention and boosting engagement.

If the business isn’t able to scale up supply economically, there’s $s left on the table

• Demand retention is important to growth, but not as influential as supply: This is due to the fact that there’s a lot more demand than supply, and demand retention is much stronger, and theoretically could be trending upwards rather than downwards after D1. In simple terms, there are a lot more rides that can be fulfilled, hence demand will not be that big an issue unless retention is extremely poor (which will show up if there is stiff competition from another provider that offers better prices/services). A 200 BPS decrease in D180 on the demand side, hardly changes the LTV of an additional unit of supply.

While this LTV model over simplifies a lot of these aspects, there’s a lot that can be learnt from quickly glancing at the KPIs and variables that affect growth. Every PM should be encouraged to build a model that helps put product growth into perspective, so that s/he can focus on the right parts of the funnel and optimize it to position the product for sustainable growth.

ODE is an industry that is poised to grow rapidly, but companies that want to establish a sustainable, successful business model will have start by building one on paper :)