AI for SDG: agent-based analysis of the Seed Value Chain in rural Mali?


This research proposal describes the design and development of an agent-based system for farmers in rural Mali. The envisaged system is used as a simulation tool to (i) analyze bottlenecks in the current system and advise the local stakeholders how to optimize the current value chain. The goal of the research project is to enhance food security in Mali, in line with Sustainable Development Goals SDG2: Zero Hunger.


Background to the problem

The seed value chain in Mali is a complex one. It is on one hand traditional, in which farmers exchange seeds with each other, without financial transactions, and on the other hand commercial, based on hybrid, laboratory-improved seeds. The traditional value chain consists of local seeds that have lower yields than the improved seeds. Still, many farmers rely on them, as they are available without investments, no pesticides are necessary. They are resistant to drought and can be — in contrast to the hybrid seeds that only breed two or three generations– replanted/reused without limitations. They are the basis of the daily food consumption for large part of the population.

The commercial seeds are interesting for farmers, as they give higher yields, and they can be sold on the commercial markets. They give farmers some real revenues. Unfortunately, the seed value chain for hybrid seeds is hampered by a lack of information and by inefficient channels for communication. For many smallholder seed producing farmers this is a serious problem. Farmers do not have access to market information; they have difficulties in finding customers and are often not up to date about market prices.

The main stakeholders are seed producers — usually smallholders — who buy and sell seeds. Often, they are confronted with a sudden lack of demand for seeds, leading to plunging market prices, especially in the occasional good rainy season and a good harvest. Especially the poorest farmers in rain-fed agriculture, in remote regions are the most vulnerable for price and demand fluctuations, as they are compelled to sell rapidly, even at the lowest prices, to dispose on cash.

AOPP, — an association of farmer organizations that reaches up to 3 million people in the whole of Mali —  is concerned with the livelihoods of their member OPs (organisations paysannes). AOPP as a national umbrella organization supports about 250 local farmer organizations, enhancing communication and trying to provide them with timely, accurate and relevant information about all segments of the value chain. To improve communication, AOPP is developing an online seed information system.

There is a local market for cereal seeds, produced (smallholder) farmers in Mali. Smallholders produce millet, maize, sorghum, fonio, rice in a number of varieties and sell these on the market.


The complete seed market (value chain) is not efficient. Some years there is overproduction, and the seed prices go down, so farmers have to sell at a loss. Sometimes the demand is high, and the production is insufficient to feed the population.


The farmers take decisions every year: e.g. what to invest, how much to sow, which product. Information to support their decisions is often lacking, which makes planning for them difficult.


Given the distributed nature of the value network, the decisions made at the local level, and how this (information) influences the value chain at the aggregated level. The research question is: can we design an agent-based system to analyze existing barriers and optimize decision-making in a distributed network of agents with a limited amount of information-based decision-making.

So the idea is to build an agent-based simulation of this network, for the different stakeholders: smallholders, buyers, central entity. (there are more stakeholders, but for the sake of simplicity).

The variables are: product type, quantity per farmer, yield (in e.g. kg), land in hectares, price, price range f fluctuations, price of investments (pesticide, fertilizer)

Time etc. # of orders, size of orders, (other costs e.g. transport, stock, certification, marketing costs, transaction costs). Decisions that farmers take

At the beginning of the rainy season (April-June):

(i) what to plant, how much to sow/plant, how much money to invest,

At the end of the season:

(ii) when to sell, how much to sell etc.

A first agent-based simulation could clarify a number of questions:

  • Are farmers willing to change their decision what to sow etc?
  • What influences this decision
  • How would timely information (on price, order, what other farmers are doing) influence decision-making?

Local experts about seeds in Mali are: Seydou Tangara and Souleymane Diarra from AOPP farmer organization.

Distributed decision-making by local farmers

Currently, the value chain for seed producing (smallholder) farmers in Mali is hampered by a number of serious barriers. One of the is lack of information to take good decisions. Before the start of the  rainy season, (which starts in June/July) smallholder seed producers in Mali make investments in pesticides and fertilizer (this is an upfront investment for them, before they start seeding) They buy seeds (base seeds, called in French “semences de base”). With this they produce second generation seeds (called R1) or third generation (R2) seeds, in the rainy season (June- October). After harvesting, they bring the seeds in sacks to the cooperative to make them sell at the (small) local markets or at the large-scale national market. There is a market demand for seeds at the aggregated level (seed market, large quantities)

These are on order for the next year. So a big seed buyer can e.g. order 20.000kg of millet in February 2020 for November 2020. There are also small customers who buy on the local market, a few kg without ordering upfront.

Problem: farmers need a planning tool to know how much the demand will be next year.

Unknown variables are:  fluctuating price of seeds, due to over- or underproduction.

Other barriers: certification delay problems

Delays in certification. Sometimes this takes too long and farmers cannot sell without a certificate for their seeds. Certification is done by a research institute. This can take months or a year sometimes. In the case the certification is not given, or too late, the farmers have to sell their seeds at a lower price, at the local market.

Crops can be depreciated when they are contaminated with no-certified crops. In that case no certificate is obtained

The envisaged  system

With an envisaged AI model, farmers will be able to:

  • Analyse the bottleneck in the value chain: where is the barrier/delay in information/value flow?
  • Help the farmers in decision-making what to plan for next year (making scenarios about price and delay, about the choice for the best crop).
  • Understand the relevance of AI models for a rural context and improve the model by providing us with expert information.