- Playbook
Agent BI: A practical guide for building durable last‑mile agent networks
The Alliance of Bioversity International & CIAT: Millicent Omala (PhD), Nathanial Peterson (PhD), and Diane Ombasyi
Busara: Rahab Kariuki, Murtaza Fazlehussein, Jackline Chemtai, Raya Shatry, Melissa Malala, Quinta Otieno, and Elijah Baraza
- November 20, 2026
- 4:12 pm
SECTOR
PROJECT TYPE
Location
BEHAVIORAL THEME
OVERVIEW
This playbook, developed by the Alliance of Bioversity International and CIAT, in collaboration with Busara, examines why last-mile agent networks in agriculture and other sectors struggle with recruitment, performance, retention, and accountability, and presents a behaviourally informed approach to addressing these challenges. The playbook argues that weak agent performance is often a reflection of weak systems, rather than individual shortcomings. It introduces AgentBI, a digital platform designed to help organisations recruit, assess, train, support, monitor, and retain agents while building institutional knowledge that can improve network performance over time.
Research questions
The playbook does not present a single set of formal research questions. Its central questions include:
- Why do last-mile agent networks continue to underperform despite investments in recruitment, training, and technology?
- What motivates different types of agents, and how should organisations recruit and manage them?
- How can behavioural research improve agent selection, training, performance tracking, and retention?
- How can technology support more effective agent management and improve last-mile service delivery?
- Can the AgentBI approach be applied beyond agriculture, including health, financial services, climate information, and renewable energy?
Methods
The playbook draws on the Enabling Market Intelligence and Building Engagement (EMBE) project, implemented by the Alliance of Bioversity International and CIAT with Busara.
The underlying work included:
-
Engagement with more than 20 agribusinesses to understand the challenges facing their agent networks.
-
A survey of 207 agents in Kenya, used to identify behavioural patterns, motivations, aspirations, and distinct agent typologies.
-
Behavioural measurement across agent networks involving multiple employers.
-
Field deployments with SeedSasa in Kenya and Everlend in Uganda to test platform features and assess their relationship with agent recruitment and performance.
-
Application of behavioural research, predictive analytics, and operational learning to develop recommendations for agent selection, training, incentives, supervision, and retention.
The playbook brings this evidence together with practical guidance, examples, and the AgentBI platform framework. The evidence is not presented as equally conclusive throughout. Some findings, particularly those from the field deployments, are described as preliminary.
THEMATIC AREAS
Key findings
- Agent networks are weakened by system failures, not just individual performance. Organisations often invest in recruitment, training, and devices without designing the supervision, support, and performance-management systems needed to make these investments work.
- Agents have different motivations and career ambitions. The analysis of 207 agents identified four typologies: Value Chain Coordinators (32.9%), Commercial Agents (26.6%), Knowledge Champions (24.6%), and Community Mobilizers (15.9%). A single recruitment or incentive model cannot effectively serve all four groups.
- Commission is not the primary motivation for most agents. Just 5.3% of the surveyed agents were primarily motivated by sales and commission, while 51.2% were primarily motivated by community and farmer success. This challenges the assumption that increasing commissions is the main way to improve agent performance.
- Retention depends on more than pay and motivation. Operational challenges, inflexible incentives, weak field support, and limited career progression contribute to agent disengagement. The playbook reports that businesses engaged at the start of the EMBE project estimated average seasonal attrition at around 40%.
- Digital tools are most effective when integrated into organisational workflows. The SeedSasa and Everlend deployments provided early evidence that platform-based assessment could help distinguish candidates and improve aspects of agent deployment. However, technology alone is insufficient. Its impact depends on operational reliability, supervisory engagement, and integration into existing systems.
Implications for policy & development
- Invest in the management systems behind last-mile delivery. Funding should cover supervision, performance tracking, operational support, and retention, not only recruitment, training, and equipment.
- Design agent programmes around different motivations and capabilities. Recruitment, placement, incentives, and career pathways should reflect the different roles agents play rather than treating them as a uniform workforce.
- Move towards outcome-linked incentives. Compensation structures should reward sustained farmer relationships and longer-term outcomes, rather than focusing exclusively on immediate transactions or sales.
- Build institutional learning into agent networks. Agent performance records and operational data should remain accessible across seasons and organisational changes, allowing programmes to improve recruitment, training, and deployment decisions over time.
- Extend behaviourally informed agent management beyond agriculture. The playbook identifies opportunities to apply the AgentBI approach in community health, financial services, renewable energy, and climate information services, where frontline workers face similar challenges of trust, follow-up, and performance management.
How to cite: Omala, M., Kariuki, R., Fazlehussein, M., Peterson, N., Chemtai, J., Shatry, R., Malala, M., Ombasyi, D., Otieno, Q., & Baraza, E. (2026). Agent BI: A practical guide for building durable last-mile agent networks. CIAT. Busara. https://doi.org/10.62372/PMQE4273