Earlier this year I gave a talk at the Onlinemarketing Rockstars conference in Hamburg on the big trends in the Marketing industry.
My main point has been, that Loyalty Marketing is still a hugely underserved market in the ad-tech industry and bears plenty of potential for retailers, but also other online players, to improve their sales & ROI
Spending money on retaining existing customers (filling the holes in the funnel) leads in most of the times to a higher ROI than just acquiring new customers.
The big question for many companies is, how to measure the long-term effectiveness of their marketing dollars and the long-term prospects of their business model.
Here are some thoughts that might help you on this path:
- The main metrics to focus on, no matter what business you are in, is the CLV (Customer Lifetime Value). How much more revenue does a customer generate over his lifetime minus the costs that can be attributed directly to the customer (marketing, COGS, payment costs, etc.)
- Focus on the right metrics at the right time. It's not a bad thing to start optimizing for CPA (Cost -per-Action) in the very beginning. It is however a bad thing for your business if you continue to optimize on CPA even after you have sufficient data to optimize for your ROI or your CLV
An overview of how to move from optimizing on CPA to CLV
over the lifetime of your business cycle.
- However, 95% of the time companies compute a completely inaccurate estimate for the CLV of their customers. They get it wrong.
- Some of the wrong methods include (and I have seen all of them)
- Taking the average revenue per user (ARPU model)from the past and forecasting the same number.
- Example: A customer signed up with your store and spend €30 in the first 3 months (€10 per month). For your business model you assume that the Average revenue per month for the next couple of years will be €10.
- Why is this wrong? - You don't account for churn, reactivation costs (e.g. vouchers, retargeting, mailings)
- Taking past user cohorts and using a linear regression model to forecast future behavior
- Example: You take all the users that signed up last January and see what their revenue has been after 10 month (in October). After averaging this out for some of your monthly cohorts, you estimate the monthly revenues of your current cohort into the future.
- Why is this wrong? - Although this method is way more accurate than the previous one, many factors that affect the CLV are not accounted for. Seasonality, the mix of the cohort, or for example the products you offer on the website may be completely different than the year before. Many times the early adopters are your best cohorts and using that data leads to overly optimistic CLV projections for the future
- So what is then the best way to calculate the CLV of your customers/ users/ visitors?
- The key is to forecast the CLV on USER level - so an individual CLV for every single person. This also allows for the first time to target users individually
- Apply machine-learning techniques and probabilistic models so that recent changes on your website, seasonality, macroeconomic environment can be included into the forecasts
- Include more factors in your CLV than only the revenue. Some of the factors can be
- Transaction-related factors (Av. basket value, time of the purchase, day of the purchase, transaction frequency)
- On-Site behavior (time-on-site, bounce rate, frequency of visits over time)
- Soft-factors (browser, device, marketing channel, etc.)
This will go a long way in accurately forecasting the CLV of your customer base - and better decision-making on when to invest in marketing and where.
Average margin error when using different methods to forecast CLV (courtesy of custora.com)
In my upcoming post I will suggest a couple of effective ways to actively increase the CLV of your user base in a few steps.
More infos on this topic in my Speech at the OMR 2013 in Hamburg
Leave a Reply