Background was founded in 2016 to build chatbots to help Kenyan farmers improve their businesses. While initial user research led to a chatbot to link farmers with buyers, testing revealed that the infrequency of crop cycles limited adoption of a marketing-only tool. Farmers first wanted a better means of filtering through large farmer Facebook groups to find relevant cultivation information about their particular location and crop. The company has since begun developing a suite of Facebook Messenger chatbots to meet this demand.

Why messaging apps?’s founders saw that farmers were already using large Facebook groups to exchange information. However, within groups of thousands, users could not keep pace with or search through the information to extract what was relevant to them. The team believed chatbots could provide efficiency by filtering group information, so they tested 20 prototypes on Facebook Messenger and Telegram. felt the Telegram API enabled bots to have more meaningful functionality within groups, but no farmers had heard of the app, and downloading and training created costs for farmers and Farmers’ awareness and existing use of Facebook groups and Messenger enabled more efficient user acquisition.

How it works continuously tests new business models and chatbots, recently deciding to offer a suite of Facebook Messenger chatbots built on a single unified database. The first is the Africa Farmers Club (AFC) bot, which is integrated with the AFC Facebook group. created the group in August 2017 to generate agronomic content, then saw an opportunity to capitalize on its growth to more than 37,000 members in three months. The AFC chatbot allows group members to input their crop and location via Facebook Messenger, which generates daily messages from the chatbot with links to relevant group content and nearby farmers, as well as games that allow farmers to earn “tokens” and compete with other group members.

Users can also use the chatbot to search for farmers and information by crop or location, prompting the chatbot to return links to content and member profiles. uses natural language processing (NLP) to extract terms from the text of Facebook group posts, label them by crop, and enable the bot-based search. Tagging by more complex criteria, such as the post’s intent (e.g., ask a question, share a story, etc.) is done manually, but expects to increasingly automate these tasks as it expands its dataset. no longer applies NLP to parse unstructured user input to the chatbots, because in testing users preferred a button-based interface, and NLP technology struggled with text written in a mix of languages.

Results and reflections

As of November 2017, the AFC group averaged 28,000 active users, 2,000 posts, 19,000 comments and 70,000 reactions monthly. does not share chatbot user data, but reports retention is high. has begun testing the next chatbot in its suite, which will be licensed to agribusinesses for use as a “digital field agent.” The chatbot can answer questions, explain products, accept mobile payments, and process and deliver information like soil testing results through an API, emulating human field agents in a more cost-effective and scalable way. In addition to AFC is now prototyping a full suite of chatbots all built on the same database to include other services such as connecting buyers.