AgentBI — linking agricultural agents to businesses
agentbi
Nate Ventures · Burgeon Strategy · CGIAR · Apollo Agriculture
Apollo Agriculture · Kenya
Nate Ventures in partnership with Burgeon Strategy, CGIAR and Apollo Agriculture
Linking agricultural agents to businesses. Based on evidence, not referrals.
Agribusinesses spend significant time finding, screening, and replacing agents — relying on informal referrals that favour known people over strong candidates. AgentBI is the trusted screening and matching layer between them.
Agribusinesses like Seed Sasa screen 200 applicants down to 20 — through costly in-person training days. AgentBI collapses that to a curated shortlist before anyone travels.
The MVP focus
Selection
One acute, recurring pain: helping agribusinesses identify and select better-matched agents. Selection combines high operational pain, recurring demand, natural data generation, and measurable outcomes.
The learning loop
3 signals
Posting + applying populates the marketplace. Contacting a recommended agent validates the platform. The rehire question after assignment sharpens the next season's shortlist.
Where AgentBI fits in the hiring funnel
Example from Seed Sasa. AgentBI solves Steps 1 and 2 — final selection stays with the agribusiness.
Step 1 — Sourcing: 200
Sourcing & first contact
Sourced via referrals, posters, churches, and community networks. Pool includes many unqualified or weak-fit candidates.
Bias risk: referrals favour known people
✓ AgentBI opens sourcing beyond local networks
Step 2 — Screening: 120→30
In-person training as screen
Seed Sasa invites the full pool to training (~60% attend). Requires venues, comms, follow-up, and managing attendance.
Logistical nightmare — and cost-heavy
✓ AgentBI shortlists before anyone travels
Step 3 — Final selection: 30→20
Interviews & community fit
Assessed on communication, trust, and ability to mobilise farmers. Final picks based on direct interaction.
Human judgment is critical here
AgentBI does not replace this step
Agribusiness Flow
Post a job → review a curated shortlist → contact recommended agents
3 screens · click through to explore
My Jobs
Agent Shortlist
Agent Profile
Screen 01 of 03
Default landing: "My Job Listings" — not "Find an Agent"
The agribusiness starts with their active listings, not a search box. This reframes the task: instead of "go find someone", the platform says "here's the demand you've created — we're working on it for you."
Behavioural Insight
Starting with their own job listings rather than an agent directory anchors the agribusiness in their immediate need. It also signals the platform is already working — applications are accumulating.
Design Tactic
Show application counts immediately ("287 Applicants · 52 Shortlisted") to signal platform effort. The number does the persuasion — it shows work has already been done on their behalf.
9:41▲ ◈ ⬛
agentbi
🔔 👤
My Jobs
Browse
My Agents
More
My Job Listings
Manage your listings and applicants
● ActiveInactive
Maize Sales Agent
Active · 2 days left
GreenFields Ltd
Sales & Marketing🌽 Maize🥔 Potatoes🌾 Wheat
📍 Kajiado, Mavoko ⏱ 1–3 years
Cotton Farm Specialist
Active · 6 days left
GreenFields Ltd
Operations🔵 Rice🫛 PeasBartor
My Jobs
Agent Shortlist
Agent Profile
Screen 02 of 03
"We reviewed 287 applicants and surfaced these 52."
The applicant pool page is where agribusinesses spend most of their time. The platform curates a shortlist but keeps the full list visible — this preserves the sense of control while dramatically reducing decision effort.
Labour Illusion — Behavioural Insight
Research shows platforms that show "we reviewed 287, surfaced 8" are rated as significantly more trustworthy than those that just show a list. The number is not decorative — it is the primary trust mechanism. Without it, the shortlist looks like it may just have been all candidates that applied.
Choice Architecture
Curated shortlist shown by default (52 of 287). "See why" expandable explanations anchor each recommendation in stated priorities. Contact actions sit right next to each card with no friction.
9:41▲ ◈ ⬛
Maize Sales Agent
287 Applications
We reviewed all applicants and shortlisted the most relevant agents based on experience, knowledge, and past performance.
Shortlisted (52)
All (287)
CO
Chidi Okafor
Crop Sales Specialist
98%
See why ▾
Digital capabilityTakes InitiativeStrong Local Network
⏱ 3 years📍 Kajiado🌐 2 langs🌽 Kamba
Recommended by AgribusinessesCommunity
CF
Cotton Farm Specialist
Crop Sales Specialist
91%
Strong Local NetworkTakes Initiative
⏱ 5+ years📍 Nakuru
My Jobs
Agent Shortlist
Agent Profile
Screen 03 of 03
The platform makes its screening work visible
The agent profile shows not just who the person is, but why they were recommended — their match percentage anchored in the agribusiness's stated priorities, not a generic score.
Trust Calibration — Behavioural Insight
"For this role your selected priorities are: Strong Local Network, Product Knowledge, Strong Farmer Mobilisation." Tiered match labels (●●●●○ dots) replace raw rankings to keep shortlists manageable. "Not a good fit" affordances let agribusinesses shape what they see — giving them control rather than a black box.
Platform Rule
Post-assignment: the agribusiness answers "Would you work with this agent again?" before posting the next job. Framed around their own future hiring, not a public review — this removes strategic under-rating risk while keeping the feedback loop intact.
9:41▲ ◈ ⬛
← Maize Sales Agent
CO
Chidi Okafor
Crop Sales Specialist
98%
Match
For this role your selected priorities are
Strong Local Network
Product Knowledge
Strong Farmer Mobilisation
Previous Experience
Sales & Marketing2–3 years
Fertiliser / Seeds2–3 years
Training / Extension< 1 year
Recommended by AgribusinessesCommunity
Agent Flow
Browse jobs → apply → knowledge check → see your match score
4 screens · click through to explore
Find Job
About Your Work
Knowledge Check
Application Sent
Screen 01 of 04
Show value before asking for effort
Agents browse real job opportunities before being asked for any profile information. Profile completion, references, and other effortful actions are deferred to the moment of application — when motivation is highest.
Behavioural Insight
Research shows job seekers rarely invest effort in profile-building before they see a relevant opportunity. Users switch from "search mode" to "application mode" only when they find a role they want. This is the motivation window — the platform asks for effort here, not before.
Design Tactic
GreenFields Ltd is shown as a "Verified employer" with "87% of agents would work here again" and "12 agents hired last season" — social proof that builds trust in the opportunity before the agent commits to applying.
9:41▲ ◈ ⬛
English
Kiswahili
For Agribusiness >
🔍⚙
Recent
Sales
Maize
Ka...
GreenFields Ltd
✓
Crop Sales Specialist
🔖
87% of agents would work here again
12 agents hired last season
Sales & Marketing🌽 Maize🥔 Potatoes🌾 Wheat
📍 Kajiado, Mavoko⏱ 1–3 years
KSh 35,000
Acre Africa
✓
Insurance Field Agent
Insurance🌱 Maize
📍 Nakuru⏱ 6mo+
Find Job
About Your Work
Knowledge Check
Application Sent
Screen 02 of 04
Profile-building at the moment of motivation
Profile completion happens during the application — when the agent already wants the job. The platform uses simple chip-select inputs (no open text) to keep cognitive load low. "This is only required the first time" removes anxiety about repeated effort.
Behavioural Insight — Reduce Cognitive Load
One question per screen, multiple-choice over open text, 6th-grade reading level. The application is split into simple profile inputs (Steps 1–5) and more engaging scenario questions — effort is distributed, not front-loaded.
Progress Effect
Profile saved for future applications. The agent can return any time. This creates a "sunk cost" that works in their favour — each completed application makes the next one faster, increasing platform stickiness.
9:41▲ ◈ ⬛
←
Step 1/5
✕
About your work
Tell us about the kind of work you have done
Sales & Marketing
Training / Extension
Input Distribution
Crop Insurance
Field Research
Farmer Groups
ℹ This is only required the first time. The platform saves this to build your profile.
Your progress is automatically saved. You can return anytime under My Applications.
Find Job
About Your Work
Knowledge Check
Application Sent
Screen 03 of 04
Lightweight signals differentiate candidates
In informal labour markets, employers have little information to distinguish applicants. Even small, structured signals significantly improve decision-making. Scenario-based questions reveal judgment under realistic field conditions — not just knowledge.
Behavioural Insight — Signal Architecture
Short, role-specific knowledge checks tied to crops or activities create simple, comparable signals of experience without long assessments. The questions rotate so agents cannot memorise answers — maintaining signal value over time.
Design Tactic
Dark mode used for the knowledge check screen — IL research shows dark mode reduces distraction and increases focus for deliberate tasks requiring judgment. Light mode for quick profile inputs; dark mode for scenario questions.
9:41▲ ◈ ⬛
←
Question 2 of 2
✕
Quiz / Agriculture Knowledge Check
Question 2 of 2 · 100%
What is a common pest that affects maize crops?
Select all that apply
Fall armyworm
Cotton bollworm
Rice weevil
Not sure
The questions will rotate, so agents cannot memorise the answers
Find Job
About Your Work
Knowledge Check
Application Sent
Screen 04 of 04
The match score is revealed as a reward — and a guide
The agent sees their match score only after completing the application. This is a deliberate design choice: showing the score at the start would create anchoring and strategic behaviour. Revealed at the end, it functions as both a reward and an actionable guide for improvement.
Behavioural Insight — Salient Improvement Actions
Job seekers have limited attention and often don't know which actions improve their chances. Prompting improvements is most effective right after relevant actions — not at random points. The platform surfaces specific, actionable steps immediately after submission, when motivation is highest.
Reputation Loop
Profile saved for future applications — the agent's next application is faster. Post-assignment, they receive a rehire signal that feeds back into their ranking. Every completed assignment is an opportunity to improve their position on future shortlists.
9:41▲ ◈ ⬛
Step 6/6 · Review Your Application · Check your details before you submit
78%
Your match score
Good match! Improve your score to increase your chances.
CV & Personal Details78% chance of getting job
Experience · Sales & Marketing✓
Experience · Sales & Marketing✓
Work location✓
Application submitted
Profile Insights Unlocked
After submitting, the agent sees their profile is now saved — future applications are faster. Similar jobs are surfaced immediately, keeping them engaged with the platform rather than returning to job boards.
Platform Learning
Every completed application generates data the platform needs: experience signals, scenario responses, contact behaviour, and eventually post-job feedback. The hiring flow creates the data rather than asking for it upfront.
"Agribusinesses post a real opportunity, and agents apply for a job — building their profile in the process."
Why this is foundational
Without active opportunities, agents won't show up. Without agents responding, agribusinesses don't see value. Every screen is designed to encourage this loop on first use. Profile-building happens as a by-product, in the window of motivation.
02
Contacting recommended agents
"An agribusiness contacts at least one of the recommended agents from the shortlist."
Why this is the trust-conversion moment
Looking at a shortlist is interest. Reaching out to a recommended agent is the first moment where agribusinesses act on the platform's recommendation, turning a suggestion into a real hiring decision.
03
The rehire question
"After a completed assignment, the supervising agribusiness answers 'would you work with this agent again?' — yes or no."
Why this is the learning behaviour
Rehire is intended to become one of the highest-value signals in the ranking architecture. Every answered rehire question helps sharpen the next season's shortlist. Without this loop, recommendations cannot improve meaningfully over time.
Six Interaction Design Principles
How the product reduces cognitive effort and improves usability
01
Show value before asking for effort
Agents browse jobs before building a profile. Profile completion, references, and effortful actions are deferred to the moment of application, when motivation is highest.
02
Make actions effortless
Simple inputs: multiple-choice questions, single-tap responses, and structured chip selections instead of open text. One question per screen, 6th-grade reading level throughout.
03
Help users understand why recommendations appear
"Find agents for your need" reframes the agribusiness task around the outcome they want. Tiered match labels replace numerical rankings. "See why" explanations make the platform's logic legible rather than a black box.
04
Frame decisions around the user's own interest
The rehire question is framed around the agribusiness's own future hiring, not a public recommendation — removing strategic under-rating risk. "Not a good fit" affordances give control without losing feedback.
05
Use effort as a meaningful signal
Application effort is not removed entirely. Completion, follow-through, and responsiveness are useful indicators of reliability. Mandatory post-assignment feedback before posting the next job keeps the loop intact.
06
Design for realistic data capture
Every signal shown in the product has a realistic collection path. The MVP avoids features that require agribusinesses to upload data they don't currently maintain — data is generated through normal use.