Back in July, we wrote about how Layer helped Jobr journey from “Tinder for Jobs” to robust, mobile-first recruiting marketplace.

One major piece of the story was the development of Jobr’s Career Concierge bot. At the time, the automated bot was able to handle about 80% of inquiries. Now, we need to revisit the topic.

So many business leaders are focused on the reported 70% failure rate of chatbots built on Facebook Messenger. And yet, Jobr trusts its bot to handle so many user inquiries no one manages its conversations full time. Why is Jobr succeeding where most other bots have failed?

We reconnected with Jay Feng, Lead Data Scientist at Jobr, to learn how building the automated bot with Layer has pushed the marketplace app forward.

From Manual Support to an Automated Bot

Jobr’s early messaging strategy was meant to pursue  5-star app reviews. These reviews were critical to their customer acquisition funnel, which relied on App Store install ads to drive growth.

The better the rates, the better Jobr’s ads converted and the lower their CAC fell. The pursuit of 5-star ratings was so important for Jobr that the company had its head of customer success manually responding to users through Layer-powered conversations.

“The original hack was to identify users who wrote back with positive sentiments and send them a pre-written message requesting a review,” said Feng. “I would interview our product manager to figure out the best place to start building classifiers for the bot. First, we automated the 5-star reviews, basically by creating a binary classifier in Python and working out a threshold that would help us balance review automation and accuracy in deployment.”

Jobr saw that their original messaging hack was limited in that it only prompted 5-star reviews from users expressing positive sentiments. Feng and the Jobr team started to build out more functionality by creating a bot.

Taking the Next Step with Bot Functionality

“Around 80% of user questions revolved around just 4 or 5 questions. And by automating 4 of our responses, we were able to expand the breadth of users we could reach. Because we had pre-scripted answers from our manual processes, we just trained the bot on these responses to achieve a solid level of accuracy.”

Users were typically asking about how the app’s search functionality works, how they can change their location, how to upload a resume, or just how to use Jobr in general. Feng and his team set a narrow domain and clear goals for its bot based on these standard questions. 

The Layer Customer Conversation Platform gave Feng and his team the flexibility to work around pre-built functions like sending messages, writing functions, send/receive notices, and more. One important feature was the automated push notification system, which has made it easier to close conversation loops with the Career Concierge.

“Every time we accomplish one goal, we start looking forward to the next one,” said Feng. “We implemented notifications with Layer and started Career Concierge a few months after we did the basic chatbot. We wanted to get more people integrated into our app. With our backend logic, we identified users who hadn’t uploaded their resumes, sent them notifications with Layer, and sparked a conversation to get them using the app again. There are many more things we can do with Layer and Career Concierge—asking for reviews was just the start.”

Key Lessons from Jobr’s Automated Bot Success

At this point, the percentage of cases where Jobr’s Career Concierge doesn’t correctly identify questions is low. This keeps the cost of customer acquisition low and customer satisfaction high.

For brands to replicate Jobr’s success with bots, there are a few key lessons to keep in mind.

1. Control the Customer Experience

Third-party platforms let you leverage the functionality of a bot. But they don’t integrate well into a larger mobile strategy.

If you’re onboarding customers within a mobile app, it doesn’t make sense to kick them out of your app and into someone else’s. Roll your own automated bot to address unique use cases and challenges.

2. Keep a Narrow Focus

Feng holds that identifying one metric for the data science project was critical to success.

First, figure out what you can optimize on. Then, build a classifier for that metric, keep iterating and retraining it, and then automate that retraining process. When one practical application is running smoothly (and automatically), you can set a new goal.

3. Own Your Data

In addition to controlling the customer experience, building a bot in-house means keeping 100% ownership of data. If Jobr deployed Facebook Messenger bot, they would have been training Facebook to enter their market.

Worse yet, all data collected on a third-party platform can be used to help your competitors enhance their own offerings. A native bot helps Jobr provide a unique product and maintain a competitive edge.

Get the Most Out of Your Chatbot

Every native chatbot development process will provide unique lessons. But the main takeaway from Jobr’s story is that you can succeed with an automated bot with the right approach.

Jobr’s success proves that brands don’t have to rely on third-party messaging platforms to get the most out of chatbots and customer conversations. Regardless of whether you create your own intelligence engine from natural language processing libraries (like Jobr), or you use an off the shelf AI solution, you still own the entire experience when you deploy via Layer.

If you want to learn more about how the Layer Customer Conversation Platform can help you deploy a native chatbot, contact us today for a free demo.