Case study · Baobab Network · Venture investor, Africa
40,000 applications a year, screened in minutes, with no investable company missed.
Baobab Network invests in early-stage African startups and receives more than 40,000 applications a year. We built a system that applies the firm’s own judgement to every application.
The challenge
Too many applications for a team to read properly.
Every application was screened by hand. The team could not give each one the same attention, and the information needed to judge risk was often missing.
- Several days to turn around each application.
- Decisions that varied from one reviewer to the next.
- Incomplete information for judging risk.
- Little view across the whole portfolio.
What we did
Write down how the team judges, build it into a system, and check it against them.
How the team decides
We worked with the investment team to set out the criteria they use, and how they weigh them, when they read an application.
- Their investment criteria written down
- Past decisions used as the test set
Baobab OS
We built a system that researches each applicant, applies the team’s criteria and gives a recommendation with its reasons.
- 38 agents for screening and the investment pipeline
- 20 agents for monitoring and supporting portfolio companies
- Over 120 data points gathered on each applicant, 60 on each portfolio company and 50 to 100 on each investor
- Several models used together, each for the task it does best
- Used through the web, a dashboard and messaging
Checked against the team
Before the team relied on it, its recommendations were compared with the team’s own decisions. Their feedback goes back in so it keeps improving.
- 86% agreement with the team
- No investable company missed
The result
Decisions in minutes, not days, at a fraction of the cost.
Every application now gets the same first screen, applying the team’s own criteria, in minutes instead of days.
The system agrees with the investment team 86% of the time, is right 96% of the time on companies that are not investable, and has not missed an investable company.
The cost of screening each application fell by more than 90%, so the team spends its time on the companies worth meeting.
Every application read and scored against the team’s criteria.
Missing information gathered automatically before a decision.
Monitoring and support across the companies they have backed.
Each recommendation explains why.
The team’s feedback improves future recommendations.
The system and its code belong to Baobab Network.