14 minute read
Tempting Thoughts to A/B Test In Trying Times
Today’s economic forecast is uncertain. However, there are simple testing ideas you can try to make the most of a difficult situation.
Your first inkling would be to cut costs: it’s hard to argue against that. But, make sure that your efforts to save are not damaging growth and retention. A second axis to consider is that your customers themselves have less interest in spending. What can you do when the household budget becomes tight? Finally, always consider the new opportunities that a recession fosters.
When decisions become difficult to make, A/B tests offer an accurate and reliable estimate. This allows everyone to agree which solution will benefit the business the most.
What are examples of cost savings that you want to A/B test?
The most straightforward way to save costs is to look at your suppliers and find cheaper ones. So far, you’ve avoided doing it: your suppliers offer the best quality. With pressure on margins, you are reconsidering. This will likely trigger an internal debate.
Will the drop in quality affect customer satisfaction? Some customers might notice and churn to competitors. No one knows the outcome yet.
If you switch suppliers and see a drop in purchases, you can’t necessarily assume that this is because of your change alone. There is a recession after all.
The best approach is to run an A/B test. Try the new provider for half of your customers. Ask all of them about their feedback and compare. Some customers will complain about the cheaper provider. However, if only a few notice, and if the savings are more valuable to you than their business at the moment, it is worth switching overall.
Degrading quality to save on margin is an arbitrage: a decision with two effects, and one of them will be negative. Arbitrages are notoriously hard to decide. You know which direction metrics will go (less satisfaction, less retention, more margin), but you can’t tell precisely how much. A/B testing gives more accurate measures for both. You can compare the two effects and pick the option with the least negative impact.
This is not just for external vendors. You might be tempted to cut certain customer support functions, or other seemingly less essential services. That can be a painful experience, one that you don’t want to risk without clear justification.
A way to measure the impact of those functions, and likely the most objective, is to run an A/B test. Take away or degrade service for a segment of users, and look at their satisfaction score and return rate. You don’t have to get rid of the department entirely. Depending on whether you want to run a 90/10, a 80/20 or a 50/50 test, you can scale down customer service to 10%, 20%, or 50% fewer hours to match your new demand for agents. That test will tell you if the changes have deteriorated the customers’ experience more than expected. With a clear signal on the deterioration, independent of the economic situation, you can make an informed decision on whether it’s worth scaling that service down further.
If you want to introduce a new service during difficult times, like offering repairs, or training new customers, then a test is useful. If that service needs additional support and the overall budget is too much, you can ask for a partial budget instead. You can run an A/B test scaled to the resources that you could obtain. For example, if you can only get 10% of the budget for supporting the service for all clients, you can run a test where you only tell 10% of users about the new service. Once it ran for long enough, use the results of that test to iterate, or convince executives that such a service is worth more than its costs. Even when the rest of the company is trying to scale down, new ideas can be approved if they help short-term profitability.
Either way, those are difficult conversations to have. Emotional pleas tend to have less weight than hard evidence, and an A/B test is one of the hardest pieces of evidence you can provide.
When looking to remove customer service functions, trust the customers to handle the most common features, or automate the most predictable ones.
Dealing with cancellations is likely a very large part of your customer contact workforce. This is often because you are trying to save sales by having one of your agents speak to them. You might not know how many orders, bookings, or reservations that step actually saves you.
Skipping human contact should save you a lot in salaries, and make your service more convenient and accessible. Which is worth more? To know that precisely, you will need an A/B test to compare.
You can give every customer at once the ability to cancel their order. However, with a recession, you would see more cancellations anyway. Only a controlled test with the same economic condition will tell you the effect of that decision.
Another common approach to reducing customer service needs is to identify common customer questions and give an automated answer. It can help in many cases, but it doesn’t always hit the mark. You must have overheard someone increasingly angry at their phone yelling “Human!”, “Agent!”, “Speak to a real person!!!” You can cut down on those and customer service calls that tie up your lines.
Are automated responses effective? Some are. Personally, my bank has a really helpful process. I am led to believe this is not typical. I know that they do it well because they’ve A/B tested their process—a lot.
If your organization has a system like automated customer handling, but you haven’t tested it, now is a good time to do so. If you want to introduce automation, try it out. Slowly roll out the feature to a small subset of users. You might discover that you’re saving time for both your team and your customers. We strongly recommend that you verify the claims of your vendors, or A/B-test on-line what your machine learning team has tested off-line. Is that system actually saving you time and money? Do you end up with fewer, less angry, customers on the phone?
All those cost-saving transformations can be painful. You don’t want to introduce them with a lack of empathy. An A/B test might not be a very heart-warming process in its face, but it allows you to say to your stakeholders raising concerns: “We’ve heard your fears, we’ve taken them into consideration, here’s the impact that it had when we tested it. They are legitimate, real, measurable issues, but they represent less value (economically, brand-wise, or otherwise) for the organization, than the alternative.”
Charge Customers Less
Another common reaction to seeing difficult times ahead is to try to increase revenue, possibly by charging customers more. You can try that, but your customers might not enjoy inflated prices as much as you do. If you want to, please A/B test it. Expect surprises.
Advertize Price Cuts
A more promising idea is to see if you can charge customers less, i.e. reduce the sticker shock. If your customers are more price sensitive, then lowering your prices could help with increasing demand. Without a detailed price sensitivity study, it’s unclear how much discount would be optimal. Running many tests to find the ideal price point should be a constant operation for your product marketing team.
A more important thing to test compared to price changes is how you communicate about rebates. Cheap is great! Promotions with emails detailing the program, bright yellow stickers, and clear rebate amounts tend to attract more customers. Your brand might not permit blatant promotions, so something subtle might work best. You can even consider introducing a new promotion into existing loyalty programs. Knowing which customers are price sensitive can be more important than how price sensitive they are.
Lower Certain Prices
If people usually fill their order with many items, you don’t have to rebate everything in the store. Supermarkets have a more effective way to attract customers. They identify products that people buy often and which prices are well known. Supermarkets also identify products that are unusually price sensitive for customers and advertise those at discounts.
That might be an interesting strategy, but don’t assume that it will work for you. Test different rebates on different products to see what is more effective. Supermarkets have run those tests for decades.
Milk, bread, and toilet paper are common examples for retail. What should be cheap depends a lot on what service you offer. Travel agents can make the hotel night cheap, but suggest complementary line-dodging tickets to attractions at full price. Restaurants tend to offer entrées at cost, but mark up wine and deserts. Video gaming consoles are notoriously cheaper than they cost, because platforms hope they’ll recoup by selling games .
These are just a few examples. There are so many ways to reduce your unit margins without lowering all of your prices. Consider the following solutions, more e-commerce-specific.
Buy Now Pay Later (BNPL)
The easiest way to lower the sticker price is a finance plan to split it in four or twelve payments. There are many payment partners that offer convenient integration to your payment process. They offer your customers instant credit, more often known as Buy Now, Pay Later (BNPL). Incurring consumer debt can be concerning in the long-term, but it’s an effective way to smooth out the bad times until the better ones return.
BNPL providers handle risk-management, collections, even foreign exchange. All that has a cost—most of the time for you, the seller. Circumstances indicate that you probably want to offer a solution like that, but you shouldn’t integrate blindly. Is the fee worth the increase in sales, or will that displace existing sales and hit your margin? Does the service work, or are they experiencing many payment issues because they don’t cover your clients’ favorite payment provider?
These are all fair questions. They can’t be answered by simply integrating with that solution, and comparing your sales from month to month. Indeed, with direr economic conditions, you should expect less sales, more price-sensitive customers overall. You also can’t just assume that BNPL will save the day: they have a cost. The increase in sales might not be worth it all things considered.
This is, again, an arbitrage between two opposite mechanisms: increase payment fee, more customers thanks to convenient credit. Your best bet remains an A/B test.
A recession doesn’t just mean that people buy less: they also buy more of certain things.
Previous recession has shown that gambling, alcohol and other “vices” tend to do well during difficult times. However, “vices” are not the only product that works best during a recession. Cheaper options (like camping over hotels) more durable products (straight razors over disposable) tend to do well too. Products with less services included likely work well too. Meal kits might replace restaurants. At-home hair-dyes take over colorists, etc.
If you start suggesting those, there could be some hesitation. A chain of hairdressing salons might pause before promoting home kits, fearing it would hurt their in-store offering. It’s hard for anyone to anticipate the combined effect of a recession and a push to self-made solutions.
This is why, if you see any similar discussion, we recommend that you echo both arguments. List the metrics that each side mentions, and use those to define an A/B test. Both sides are likely true: yes, now is the time to offer cheaper options to keep most clients, and yes, that could hurt other offers. However, neither side can tell precisely by how much. With the accuracy and reliability of an A/B test, you can test the idea and get everyone to agree which solution will benefit the business the most, at least in the short term.
Cater Your Offer to the Growing Number of Job Seekers
Finally, one of the most concerning aspects of recession is the increase in people losing their job. More people without a job are looking for one. In many developed countries, this group has almost disappeared recently. You can tailor an offer to them: offer job-seeker discounts, for instance.
Some stakeholders might be worried that a discount program like that, without strict control, would be subject to abuse, especially if the program is widely advertised. Others might be worried that controls would be expensive, or counter-productive. You can compare alternatives using an A/B/C/D test to distinguish the different effects. Test presenting the offer with and without strict controls, not mentioning but offering the option if they ask, or not offering the option at all.
Help, But Test Good Deeds
Job seekers are likely keen to receive support when looking for a job. For instance, dry cleaners have been known to offer free ironing before an interview. Such initiatives have a clear value, but a cost too. You might fear that a similar initiative won’t be as useful as it seems. Alternatively, you might want to defend those but fear backlash as your organization is cutting costs. Either way, offering to launch such programs as A/B tests is the best option to measure how useful they are.
Testing itself might face internal criticism. Some people see such actions as charity, and believe that good deeds should not be tested. This was a widely held position until recently. Thankfully, Esther Duflo and Abhijit Banerjee started testing the impact of international aid with Random Control Trials (RCT), i.e. large scale A/B tests. They showed which type of assistance worked, and which was a waste of precious aid money.
That work was incredibly controversial initially. It proved eye-opening, and transformative for international assistance. The findings ended up being widely praised. The approach even earned them a Nobel Prize.
Challenge the idea that altruism is inherently good, and that it should not be tested. Not all good deeds work. If you want to help effectively, you should verify that you are alleviating more than your conscience.
If you are convinced that aid will help, and refusing it is unethical, A/B testing offers the best pattern to address that question. At the end of the test, roll-out the best solution to everyone. In many cases, this is what clinical tests do, and what Duflo and Banerjee have done. They delayed the assistance by a few months, enough time to have clear test results. The test guarantees the impact, and justifies the effort. This brings trust and ensures that the help will continue to be given in confidence.
Tough times offer opportunities. Most of those new ideas are great, but will need to be introduced gradually. Objective feedback will offer the support needed to grow them. Uncertain initiatives need reliable control to make sure that they fulfill their promises and don’t cost more than they bring.
Bad times are worrisome. They require resilience, empathy, and many adaptations. There’s a lot you can do. Lower your cost base, change your prices to adapt to tightening price sensitivity, and explore new opportunities. For all those changes, you don’t want to ignore disagreements. They are often valid points. The best approach is to listen, measure the concerns and offer to run an A/B test to compare the different effects.
Start Delivering Features That Matter, Faster
We’re happy to answer any questions or provide more information on your software development experimentation journey. Split can get you going with feature flags built upon a patented data attribution engine. Start doing canary releases and gradual rollouts backed by the industry’s best insights. Begin your free trial, request a demo, or get Split certified through our Split Arcade. Split Arcade includes product explainer videos, clickable product tutorials, manipulatable code examples, and interactive challenges. Breathe a sigh of release with Split!