For retailers, Winter holiday shopping season is like a yearly championship series. How well mobile apps and web sites perform during this crucial November-January period can make or break an entire year.
Can a mobile shopper quickly and reliably connect? Search and compare products? Load a cart and check out? Check account status? Always important, these basic online tasks are crucial during the busiest buying times.
In top world markets, holiday sales alone can account for 20-25% or more of annual revenue. Percentages are lower in Latin America. But projected ecommerce growth of 15-20% a year, and the world’s fastest growth in mobile commerce (36%), make the region one of the hottest online shopping markets in 2020 and beyond.
Like all teams, online retailers come to the annual winter competition with both strengths and areas needing improvement. Winners build on the former and work hard to improve the latter. The bigger the brand, the more challenging (and crucial) to get and stay in top shape for the holidays and the rest of the year.
The reason is simple: Regardless of geography or season, there’s a straight line between high-performing shopping apps and maximizing sales:
- A global study by Google found that as page load time goes from one to five seconds, the probability of a bounce increases 90%.
- Akamai found that a 100-millisecond delay in website load time can hurt conversion rates by 7%.
- Amazon calculates that every 100-millisecond delay leads to a 1% loss in sales. That means a five-second delay can cut sales by 50%!
In Mexico, top brands prepare all year for the big buying season that starts with El Buen Fin (Nov. 16-19) and Black Friday/Cyber Monday, then rolls through Christmas and New Year’s, Dia de los Reyes Magos (Jan. 6), and Valentine’s Day before Easter sales begin.
In this post, we’ll show how AI-driven testing can help a leading Latin American online retailer improve mobile app performance to enable increased sales.
COPPEL: GROWING FAST IN LATIN AMERICA
Coppel.com is Mexico’s third most popular ecommerce site. With nearly 18 million monthly visits, the department store chain surpasses Walmart, Sears, Liverpool, and Sam’s Club, trailing only Mercado Libre and Amazon. Mobile plays a key role: Topping the company’s 13 apps on Google Play is a shopping app with 4.1 stars and more than one million downloads.
Strong online sales have helped make Coppel one of the 30 fastest growing retail brands in the world. Founded in 1941, the Culiacán-based company now has 1,400 stores in Mexico and Argentina, employing around 125,000 people. Both online and brick and mortar outlets sell electronics, home decoration, appliances, clothes and accessories available via easy credit and installments, a big competitive edge.
Coppel is a large, sophisticated user of technology, committed to improving online customer experience as it expands across Latin America. In 2017, the company launched major initiatives to improve its online apps. Efforts have been successful:
electronic sales grew 90% in 2017; Coppel doubled its market share in the category in 2018, and expects strong future growth.
Recently, we noticed comments on issues with connections, page loads, shopping carts and billing. In the spirit of continuous improvement, we took the initiative to conduct performance tests. Our aim was to help Coppel uncover new opportunities for ecommerce optimization that would provide benefits during peak season and year round.
ABOUT THE APP PERFORMANCE SESSIONS
Specifically, we sought to assess the client and client/server performance of Coppel’s popular mobile shopping app. The goal: Identify issues and fixes for problems most negatively impacting visitor experience.
In December, we conducted sessions using the HeadSpin Connected Intelligence Platform. The tools identify, prioritize and visualize mobile app performance improvements using AI-powered issue detection. In total, we ran 60+ tests, spanning four geographic locations, three different mobile devices (Moto, Pixel, Nexus), and two different mobile carriers (Telcel and AT&T).
Note: These were not emulations or simulations, but actual sessions, under real-world conditions. The mobile phones chosen reflected the variety of high-mid-and low-end mobile devices used by actual Coppel customers in the region. We tested for strong as well as weak carrier service coverage.
THE USER EXPERIENCE
The scenario: A mobile shopper wants to browse new accessories for their television, then decides to look at new TV models.
A quick recap: We arrive at Coppel’s bright blue and yellow mobile welcome screen. Our first 26 seconds is consumed by two attempts to permit location tracking. At 55 seconds, we get to Electronics, then choose Televisions and Video, then Accessories (5-second load). We choose a smart remote (6-second load), an HDMI 4K cable (2-second load), an LED light (2-second load). Now we switch sections to browse new TVs (4-second load). We select a Panasonic (15-second load), then get a Server Connection error. We try a new Sony (2-second load), then call it quits at 3:30 seconds.
You can watch the session here.
As you can imagine, the tests produced a lot of actionable data and insights for improving mobile app performance. There’s too much to present here, so we’ll focus on a few of the top issues identified by HeadSpin’s AI-powered platform.
Let’s take a brief look to better understand what we just experienced.
1. Big Picture
First Load Report
December 3, 2019, 1:11-2:17 PM, PST
Devices: Moto, Pixel, Nexus
Carriers: AT&T, Wifi
Locations: LTE (benchmark), LTE-Congested, Poor Connection, 3G
This “forest-level” assessment shows how well the app performs across the network, carriers and locations. It delivers a high-level summary of the most important and actionable insights. These include: key findings and test statistics, benchmarks and comparisons, and high-fidelity video examples of key user sessions, before and after potential improvements.
To ensure statistical significance, the HeadSpin AI engine analyzed several sessions, checking each possible combination of location, device and carrier. Doing this test first helps focus our efforts and provides initial, AI-driven identification and prioritization of issues.
After 76 minutes of intensive testing, three major issues were revealed: Slow servers, domain sharding, and slow TLS. (You can see why they’re problems in the brief explainer for each issue above.)
Now, we’ll drill down and analyze issues in three different dimensions: By Location, Device, and Carrier. That will help us assess individual and combined impact of each issue on user experience. More importantly, it will prioritize the most impactful fixes.
1.1 “3D” overview of the First Load test at Coppel
Right away, the long yellow-orange-brown bars reveal clear issues with Location. (This dimension shows how different network/geographic combinations are impacted by performance issues). In this test, Location includes LTE, LTE-Congested, Poor Connection, and 3G – all in Mexico City.
Let’s drill down further to look at our first problem, Domain Sharding, through this lens.
Key takeaway: As the yellow-orange line at the top right shows, fixing Domain Sharding would yield a 7x benefit. In other words, every peso or dollar invested in fixing the issue would return seven.
The bottom bar shows similar benefit for fixing Slow Server. Let’s drill down on this property to analyze further with the BurstUI.
The mobile device screen at the top left shows what the user is seeing during the testing moment (in this case, permission to track location). The bar graph to its right reveals the combined impact of the four biggest performance issues: 2.19 seconds.
In both the peak of the line graph (middle) and the Issue Cards (below), we see that Loading Animation and Slow Server have the biggest negative impacts on user experience.
Key takeaway: Optimizing for Location will provide the most bang-per-buck in remediating all the problems identified.
1.2 Potential Improvements
Our main mission is to find and fix issues with the biggest drag. So let’s wrap with a quick overview of suggested solutions to the most pressing app-slowing issues we identified.
HeadSpin’s “Potential Improvement” capability shows the net impact of fixing user experience issues. Using the data from the First Load Report, we can see the effect of remedying individual problems or combinations.
For example, this chart shows that 95% of user flows would at least be 3.27 seconds faster by fixing the issues identified by the First Load report. (We can easily analyze the performance impact and percentage of users impacted by other improvements.) While every retailer doesn’t need world-class speed, every fraction of a second translates into real revenue.
2. DEEPER DIVE
Performance session drill-down
December 19, 2019, 6:59:55 PM
Mobile Carrier: Telcel
Device: MOTOROLA, XT1710-02 on Android 8
The First Load report gave us a high-level overview of overall app performance using various combinations of locations, devices and carriers. The results helped identify key issues for Coppel along with prioritized recommendations for remediation.
Now, let’s do a deeper analysis to further pinpoint the major slowdowns we’re working to correct. We’ll do so by digging into one of the performance sessions on an Android device in Mexico City.
Here’s the high-level Waterfall view of overall results that shows the issues impacting the user experience. The middle section with flags shows details of Domain Sharding, down to the server level. The video screen at right shows what the user sees at that and every step.
Here’s the Burst view….
In both views, you can easily spot familiar issues (Connection Reuse, Domain Sharding. HTTP Redirects), which warrant further exploration. Let’s take a quick spin through them:
2.1 PERFORMANCE ISSUES
This technique can be effective with a code base optimized for web applications. But as we see, it’s a problem for mobile apps. Clearly, resources from within Coppel.com must be better served to avoid problems here.
Each new connection incurs a performance cost. When many new connections are made repeatedly, costs adds up, slowing and negatively impacting user experience.
Here, we see that the application created new connections instead of reusing existing ones, resulting in unnecessary TCP/TLS handshakes for the impacted hosts: static.coppel.com, ecommerce.coppel.com, and coppel.com. This accounts for why we saw spinning wheel on screen repeatedly during our test. The good news: Because the issue occurs on a Coppel domain, the fix is within our control.
We see the 301 error coming from Coppel.com. The client needs to resolve for URLs that cannot be reserved properly. More on fixes in a moment.
Client side performance
We also ran the Coppel app through client side performance testing to see how fast it performed on cold-start. Load time was a bit higher than the two seconds recommended, but still quite good. For perspective, here’s how Coppel’s app stacks up against social leaders:
An online retailer doesn’t need the same blistering performance as a global hyperscale giant. But Coppel holds its own here, even beating Twitter. That’s a very solid foundation for online retail success.
2.2 THE FIXES
Specific recommendations are based on proven general fixes, derived by our AI engine. For educational purposes here, we’ll share high-level remedies for the top client/server issues impacting mobile apps identified at Coppel:
Cause: An application using multiple subdomains to serve resources in parallel. Performance issues arise because mobile devices and the Web have different connection pool limits. Each additional connection on a subdomain requires a DNS lookup and new TCP/TLS handshakes, causing significant overhead.
Solution: Request all resources from a single domain. That will reduce performance impact from connection re-use issues, DNS lookup time and packet loss.
This happens when the application creates new TCP connections instead of reusing existing ones. If keep-alive connections aren’t enabled, or the timeout is set too low, client connections terminate prematurely, resulting in unnecessary TCP/TLS handshakes. So the first step here is to check the keep-alive settings on the host server.
These can be costly, especially if they introduce a new host. Doing so can incur additional DNS, TCP, TLS and request-response roundtrips, which add unnecessary delays. The fix starts with a systematic, detailed review of HTTP response status codes.
POSTLOG AND CONCLUSION
Winter holiday season is crucially important for online retailers. Systematically improving the performance of mobile apps and websites helps sellers perform their best during busy times, and throughout the year.
Coppel has created a quick, solid, muti-functional online buying app. It’s fast-starting, but can get bogged down during use. Our tests pinpointed technical issues, including slow animation, connection reuse, domain sharding and HTTP redirects. Each could be fixed and optimized for faster, smoother mobile shopping.
The slogan on the company’s shopping site says: “Make your life better.” By continuing to create better online experiences for its customers, Coppel is furthering its own transformation, strengthening revenues and position in the highly competitive Latin American market. With combined online retail sales in the region forecast to rise from $61.5 billion (US) in 2019 to $82.3 billion (US) by 2022, it’s a winning move.
Accelerate your Automation Skills
Introducing HeadSpin University