Understanding customer lifetime value (LTV) is one of the most complex and important analyses a business can tackle. Every part of your organization affects the outcome of the calculation: acquisition costs, revenue, customer service, and returns.
Learn about:
- Calculating customer lifetime spend. How can you measure customers who've purchased and their distribution of lifetime purchases over time.
- Real-world decisions based on lifetime value. How can you use LTV data to focus marketing and drive repeat purchases?
- Modern web-based analytics tools. What tools can help you foster collaboration, explore data in greater depth, and ensure cross-company use of data to drive smarter growth?
2. • Traffic acquisition
• Sales targeting and
costing
• Re-marketing efforts
• Customer support focus
• Product decision making
Customer lifetime
value sits at the
center of every
important business
decision.
3. Roadmap
• Where to start understanding loyalty
• How to perform basic cohort analysis
• Simple ways to predict LTV better
• Extensions and improvements to your model
• Applications across the business
• Q & A
4. Who We are
Al Ghorai
VP of Finance, Planning and
Analytics, thredUp
Colin Zima
Chief Analytics Officer,
Looker
5. About thredUp
thredUP is the nation’s leading (r)ecommerce marketplace for buying and
selling the highest-quality used women’s and kids clothing.
We are reinventing the way consumers shed old clothing and shop for
clothing online.
6. About Looker
Looker is a data exploration solution
that operates in the database
to enable anyone across an organization
to explore data in all its detail.
14. Engagement Metric
3500
3000
2500
2000
1500
1000
500
100%
80%
60%
40%
20%
0%
Engagement Rates of Users with 2+ Purchases
jul
aug
0
Engagement Profile by Cohort
Dead
Lost
At Risk
Engaged
Uber Engaged
15. tU TV Campaign
Orders / User
103%
113%
160%
140%
120%
100%
80%
60%
40%
20%
0%
pre campaign campaign
control
target
104%
Rev / User
146%
160%
140%
120%
100%
80%
60%
40%
20%
0%
pre campaign campaign
control
target
10% lift in order rate from users in campaign 40% lift in revenue over same campaign
18. Measure at the Widget Level
gross revenue $10.00
returns ($2.00)
discounts ($1.00)
other revenue $0.50
cash $7.50
material cogs $2.00
labor $1.50
other cogs $0.75
cogs $4.25
profit $3.25
Every point along the P&L is an opportunity to differentiate a user
Isn’t this the real measure of a customer?
20. Why Does This All Matter?
• Driving users towards inflection points
– How can product create “better” users
• Smarter marketing spend
– What channels have the highest value users
– What can we pay to re-engage a user
• Cultivating support for better users
– Focus time on your most valuable customers
23. Thank You
Al Ghorai al@thredup.com
Colin Zima colin@looker.com
Notes de l'éditeur
This presentation will not dive deep into any single topic, but rather show a lay of the land in attacking customer lifetime value and it’s applications
Al -
% of registered customers that have ever purchased; examining loyalty and value by different attributes (gender, age, location); identifying funnel drop-off
Finding inflection points of intent is crucial for driving usage. Much like Twitter found that 20 users drives long-term usage, every business has a critical interaction point where users “become loyal”. This may be different for every business – 1 purchase, 2 different items, buying outside a holiday – but analyzing historical data is crucial to finding these signals.
Repeat purchase rate by first-item purchase somes how some buyer types may drive more loyalty on future purchases. Subsetting users to experiment with different items may drive more future purchase intent (ie. Buying staples rather than one-off items)
TO DO – one slide each graph. Highlight the highlights more nicely. Al to send new graphics.
Trying to put the data in ‘looker’ format instead of excel….
For thredUP order mix is an indicator of loyalty. A customer buying for herself and her kids is a more loyal user of thredUP.
Also movement from web based platforms to mobile apps can in some cases be viewed as a higher level of engagement for a user
Repeat customers coming back to us organically instead of thru paid/referral and affiliate channels
Lastly, discounting behavior shouldn’t be forgotten as an indicator of loyalty, repeat (or even new) users that buy without discount surely have a head start on LTV!
TO DO – one slide each graph. Highlight the highlights more nicely. Al to send new graphics.
Trying to put the data in ‘looker’ format instead of excel….
For thredUP order mix is an indicator of loyalty. A customer buying for herself and her kids is a more loyal user of thredUP.
Also movement from web based platforms to mobile apps can in some cases be viewed as a higher level of engagement for a user
Repeat customers coming back to us organically instead of thru paid/referral and affiliate channels
Lastly, discounting behavior shouldn’t be forgotten as an indicator of loyalty, repeat (or even new) users that buy without discount surely have a head start on LTV!
Classic cohort analysis allows for simple LTV calculation. How much is each user spending each month since registration. This can be extended to first purchase or any other appropriate level to value customers at different purchase levels (ie. Value of a visitor, value of a registered user, value of someone adding a product to cart, value of a purchaser). This analysis could also be used to value users for re-marketing (ie. Value of user post purchase, value of a user that purchased but has not booked in 6 months, etc).
After deriving historical estimates for LTV, it is important to put the averages in the context of history. Cohort analysis allows us to dig into groups of alike users and model how the business has changed over time. In the above example, we see so positive trends in early spend (revenue increasing in Months 1, 2, 3), while revenue flattens in Months 4-6 and may even be trending down for later months. In this example, a shift in customer mix (different sources or markets) could be driving more one-time customers while long-term user experience may be diminished. This type of analysis allows us to track that our cohorts ‘look like previous customers’ such that our LTV predictions remain valid moving forwards. In the case above, we may adjust estimates for LTV upwards due to the large increase in early lifetime spend, but temper our customer lives to fewer months or lower spend estimates as customers season.
Early lifetime conversion rates or spend, increasing over time, but there appears to be downtrend in long-term loyalty. This is a simple way to extend the first cohort analysis. Are the real trends in ROAS that impact when spend comes in and how it ages over time.
We think there is absolutely implied future value. We struggle with how to quantify that implied value beyond the all to easy simplification of just giving them a 2 year life.
We wanted a metric that indicated the overall level of engagement of a user, and we wanted it at the user level so we could roll it across a variety of dimensions.
So while we might not give a large future lifetime to a Lost user in the Jan 2014 cohort (and the above graph would do this too). We will give a good future lifetime to that uber-engaged users from the Aug 2012 cohort. I think this methodology helps rationalize longer lifetimes for loyal customers.
If marketing is doing a large winback campaign for at-risk or lost user, we can use the metric to do a month over month comparison of enagement rates
The metric also helps us with lifecycle marketing (thanks Jordan!)
I throw the picture of James in there as he likes to joke that the revenue lift was from his performance on the morning talk shows in support of the campaign. We can kill it.
Note how different subsets of users can have entirely different spending profiles. While differences like those between California and Maine might be the result of thin data or small groups of distinct customers, difference between traffic sources (like Google search and Facebook display) can be drastically different. To effectively track ROAS, it often is important to use independent evaluation assumptions for these disparate sources by repeating the previous analysis with users bucketed into different groups. It’s also important to note that even when LTV calculations may return the same ultimate value for Google and Facebook above, the different payback paces mean that at any given point of time since download, the measure of success will be different. In the above example, 50% of Google CLV is returned in the first 2.5 months, while for Facebook, 50% of CLV is not realized until month 8. This data can also feed into financial models and cash flow models.
In the previous example, we have been predicting customer lifetime value using our most stable metric – revenue. Businesses cannot run on revenue alone, so understanding the cost side of CLV calculation is equally important. Unfortunately not every item in the store holds the same margin – especially when promotions are common.
Imagine, for example, a business with customers that spend $500 over their average lifetime. Underlying this estimate is several customers that may spend upwards of $10,000 and many customers worth $0 or even negative amounts when costs are taken into account. Understanding the underlying purchase distribution – margin per item, promo code usage, or return rates – for each bucket of users can have enormous impacts on ultimate lifetime values. While two different groups (say our Google and Facebook users above) may have similar LTVs on a revenue basis, if the return rate is 2x from Facebook users, we again will need to adjust our business decision making to reflect the reduced profitability and increased costs of this group. Note the importance of leverage in the lower left.
Can add more on costs if we want to.
At thredUP we are using looker to help us build a P&L at the item level.
If we can understand these accounts at the item level and
If we can link item to user id then
We can roll up granular costs to the user level and get a much better understanding of a customers VALUE
By the same token, customers do not purchase in a vacuum. No matter the business, customers interact and with that interaction comes virality. Virality estimates do not have to be large to have enormous impacts on customer acquisition decisions. If every 3 customers bring in a new customer, any given customer you buy is actually worth 50% more than you think! If every 2 customers refers a friend, LTV is double!! There’s a reason people love virality.
Virality calculations can be tricky, but understanding customer sources through survey or promo codes can be an effective way to estimate the benefit of your existent user base on customer growth. Need a good way to visualize the math of adding a multiplier to our calculations above.
Would love to break this into 3-4 slides where Al, you can talk about how this data feeds into your business directly. I am happy to fill in the gaps for examples that may not apply. Would like to hint at sales/marketing (how much to pay and which channels, payback times, cashflow burn, etc), product (decision making on pushing funnel dropoff points with AB testing or alternate flows, remarketing, credit decisions), customer support (tailoring service to whales, amount to devote to retaining users in SaaS businesses).