This blog is part of the ‘GSMA AgriTech AI Blog Series,’ which highlights AI innovations in agriculture. Throughout the series, we will explore various use cases where AI is transforming agritech operations and farming practices.
Access to agronomic advisory empowers smallholder farmers to plan better, adopt more efficient and sustainable practices, and improve yields. Yet, across much of Africa, access to this type of guidance is often out of reach. While digital agriculture solutions offer significant opportunities to bridge access gaps, barriers such as low digital literacy and limited availability of content in local languages hinder their potential.
Farmerline, an agritech company licensed in over 50 countries, developed Darli AI, a multimodal AI-powered solution that provides agronomic advice to smallholder farmers in local languages via voice and text. This blog explores how Darli AI breaks through language and literacy barriers, and dives into the benefits and difficulties of leveraging AI for the benefit of smallholder farmers.
Bridging the agronomic advisory gap for Africa’s smallholder farmers
Across Africa, smallholder farmers face challenges in accessing relevant and actionable agronomic advice that can help them adapt to a changing climate, identify and treat crop pests and diseases, and ultimately improve productivity and resilience.
Historically, they have relied on two key sources of agronomic advice: government-backed extension officers and private-sector-led field agents. These agents traditionally deliver agronomic advice through in-field visits or group trainings, but these methods often fail to reach many smallholders. Extension officers and field agents are too costly to deploy widely due to high operational costs and sparse farmer distribution. On the other hand, group trainings require significant time commitments from farmers and incur high costs for those offering the sessions. Even when available, in-person advice may not address the specific challenges faced by farmers.
Digital advisory solutions have emerged as a cost-efficient way to reach farmers with agronomic knowledge. Since 2010, the number of digital advisory services grew exponentially and, by 2022, they had already reached close to 300. However, digital literacy, internet connectivity, data affordability, and device ownership remain important barriers to wide access and adoption of digital agriculture solutions.
Low levels of basic and digital literacy, combined with language barriers, create further obstacles to knowledge dissemination. Text-based digital advisory services that are only available in a country’s official languages or in English often exclude farmers with limited literacy or formal education, especially those who only speak local languages or are unable to read or write in any language.
Successfully reaching diverse smallholder farming communities requires tailoring advisory services to accommodate varied languages, literacy levels, and digital skills.
Darli AI: A multimodal AI-powered solution to agronomic advice
Founded in 2013, agritech company Farmerline aims to support farmer livelihoods by leveraging technology to facilitate access to finance, agronomic advisory, and markets.
To overcome literacy constrains faced by farmers, Farmerline initially delivered advisory services through push voice messages in local languages, facilitated through SIM integrations with mobile network operators (MNOs). As the need for more tailored support grew, the company later developed an Interactive Voice Response (IVR) menu that allowed farmers to access relevant agronomic information on-demand in the most widely spoken local languages across West Africa. Besides accessing information on the IVR menu, farmers could request to speak directly with a helpline agent for more detailed or location-specific inquiries.
Recognising the need to scale beyond their existing language offerings, Farmerline introduced AI-powered solution Darli AI, to supercharge its IVR and helpline agent capabilities. This solution was particularly valuable for supporting languages that helpline agents couldn’t speak, allowing farmers to interact directly with the AI model in their native tongues. More recently, Farmerline expanded Darli AI’s functionality to include a chatbot and pest and disease detection model.
Key Features of Darli AI
Named after the Ewe word for “whisper”, Darli AI is designed as a multimodal digital agriculture advisor, providing smallholder farmers with pest and disease diagnosis, crop recommendations, weather information and good agronomic practices.
AI-powered IVR solution: Darli AI incorporates a Natural Language Understanding (NLU) engine with machine learning into Farmerline’s toll-free dedicated IVR number. The model is regularly tested by Farmerline’s team of agronomists to ensure that the information is accurate and relevant.
After dialling the IVR number, a farmer selects their local language and starts interacting with the AI model. A proprietary Automatic Speech Recognition (ASR) model transcribes the farmer’s spoken query and translates it to English for processing. Machine learning algorithms then retrieve relevant information to address the query. A response is crafted in English and translated into the farmer’s local language in audio. The process only takes a couple of seconds, and the interaction can continue until the farmer considers all his questions answered.
Available in 27 languages and compatible with basic phones, the service is highly accessible by remote and illiterate farmers.
Sentiment analysis: To optimise its AI-powered IVR solution, Farmerline developed a sentiment analysis tool that analyses interactions between farmers and the AI model. Using Natural Language Processing (NLP), the tool processes full transcripts of conversations to gauge the emotional tone of a conversation. If the tool detects a negative sentiment or a topic that requires deeper assistance, it flags the issue to a helpline agent for follow-up.
The tool also identifies recurring topics and offers insights into common farmer concerns or content gaps. This helps Farmerline refine the AI model and improve its agronomic advice. The sentiment analysis tool works across both human-to-human and human-to-AI interactions, making it especially valuable for Farmerline’s corporate partners, such as agribusinesses and food companies, who use Farmerline’s call centre to engage farmers for training, input sales, or crop procurement. The tool flags any unresolved issues to their teams, strengthening farmer loyalty and service quality.
Chatbot: Darli AI also includes a chatbot where farmers, with the aid of extension officers, can type questions on topics such as regenerative farming practices, soil preparation, planting techniques, pest and disease management, and post-harvest handling, and receive answers in real time. This service, available via WhatsApp and a web platform, leverages NLP and machine learning to understand user input, match it with a large database of agronomic knowledge, and generate tailored responses. So far, the chatbot is trained in English, with more local language capabilities under development. The chatbot connects with Farmerline’s helpline, allowing complex queries to be referred to human agents when necessary.
This feature is primarily used by more tech-savvy, smartphone users, such as Commercial farmers and crop aggregators.
Pest and disease detection: Embedded into its chatbot, Darli AI’s pest and disease detection tool uses machine learning and image recognition to help farmers and field agents identify crop pests and diseases receive tailored treatment recommendations. This feature currently supports key staple crops grown across Africa, including cassava, cocoa, tomatoes, sugarcane, wheat, coffee, tea, maize, soybeans and rice.
In Ghana, not only can the tool provide a diagnosis and recommend treatment, it can also map a farmer’s location and suggest nearby agro-dealers and agribusinesses for purchasing the recommended inputs—thus helping bridge the gap from advice to action.
According to Farmerline, Darli AI is now available in 60 countries and has supported over 1 million users. The agritech claims that, to date, Darli AI has handled 8.5 million interactions and calls, and more than 6,000 smartphone empowered farmers have engaged with the chatbot via WhatsApp.
Key benefits of using AI for digital advisory
The use of AI has significantly enhanced the inclusivity and efficiency of Farmerline’s digital advisory services. It has also allowed the company to offer services in a scalable and cost-effective way.
Scalability: AI-powered agronomic advisory through IVR and chatbot enables Farmerline to deliver customised advice at a larger scale compared to traditional methods that involve, for example, in-person training, field agents, or helplines. Using Darli AI is not only more cost-effective, as it significantly reduces reliance on personnel, it also reaches a broader and more diverse audience through its support of 27 languages.
Inclusivity: Darli AI supports 27 local, low-resource and high-resource languages and is compatible with basic mobile devices. It addresses a critical market gap, and has enabled Farmerline to become more inclusive, reducing the barriers that often prevent rural farmers from accessing information digitally. Farmerline’s intentional focus on low-resource languages—many of which are spoken by smallholder farmers in West Africa—ensures that even those who cannot read, write, or use smartphones can still access essential agronomic advisory through simple voice interactions.
Reduced drop-off rates: Previously, farmers experienced long wait times to speak with a helpline agent, especially when language barriers existed. For instance, if 10 farmers speaking Twi called the helpline, but only one or two agents spoke the language, delays were inevitable. This situation often led to frustration and high call drop-off rates. With Darli AI, Farmerline can respond to questions such as when to apply fertilizer or when to plant, in real time. This reduces wait times, lowers drop-off rates, and provides a more satisfying experience for farmers.
24/7 access: AI has allowed Farmerline to offer round-the-clock access to agronomic advisory, overcoming the limitations of traditional, human-led services that typically operate only during business hours. While helpline agents are available from 9am to 5pm, Darli AI is accessible 24/7. Farmers can contact the service at any time, including outside working hours, to ask questions and receive guidance in their preferred language.
Considerations for using AI for digital advisory
While AI presents significant potential, implementing Darli AI required Farmerline to navigate several important considerations, including overcoming translation challenges, building speech recognition capabilities for informal dialects, and adapting the system to real-world field conditions.
Limitations of Mainstream Translation Engines: To translate their advisory content into 27 unique languages, Farmerline mainly relies on existing translation engines, such as Google Translate. However, these tools often fail to accurately translate technical terms into local languages like Twi. Some terms, like “mulching,” lack direct equivalents in these languages, which can lead to confusion in translations. Farmerline focused on translating concepts rather than words. For example, instead of using the term “mulching,” Darli AI uses phrases like “putting dead leaves on your soil” to ensure clarity and understanding.
While Farmerline has made significant strides, they draw inspiration from pioneers like Ghana NLP and Lelapa AI in South Africa, who have been leading innovative NLP research for African languages. Their work continues to influence and shape ongoing efforts in the sector.
Taking steps to recognise informal dialects: Beyond translation, Darli AI requires accurate speech recognition. However, most open-source AI models don’t support informal, spoken dialects. To overcome this challenge, Farmerline developed a custom Automatic Speech Recognition (ASR) system in-house, using internal voice data from helpline conversations with rural farmers.
Accounting for real-life conditions in the field: As an AI model, Darli AI may appear robotic in the way it processes data and verbalises text. Transcriptions can be affected by noisy environments in the field, or by fillers or pauses that may occur in speech. To ensure smooth interactions, Farmerline integrated audio preprocessing techniques to remove background noise and handle pauses or speech irregularities, improving both transcription accuracy and user experience.
Outlook
Farmerline is already making continuous improvements to Darli AI to enhance its functionality. Future plans include expanding the number of supported languages from 27 to 55 by the end of 2025, launching a dedicated mobile application to access Darli AI chatbot, and using NVIDIA’s NEMO architecture to develop more sophisticated translation capabilities.
By embracing AI-driven solutions, Farmerline’s goal is to ensure that smallholder farmers can access much-needed guidance and be part of the digital revolution in agriculture.
Read the previous blogs from our AI series:
- ‘Harnessing machine learning to power weather forecasts: Insights from Ignitia’.
- ‘Detecting and managing crop pests and diseases with AI: Insights from Plantix’
- ‘AI-driven smallholder farmer lending in Africa: Insights from Apollo Agriculture’
Stay tuned for more insights into how AI is shaping the future of agriculture.
The GSMA AgriTech Accelerator is funded by the German Federal Ministry for Economic Cooperation and Development (BMZ) and supported by the GSMA and its members.
