Evaluation
Tabiya evaluates each deployment at four levels. The levels follow the AI evaluation framework for the development sector published in 2025 by the Center for Global Development, the Agency Fund, and partners.
| Level | Question | Who leads |
|---|---|---|
| 1. Model | Does the AI model produce the desired responses? | Tabiya |
| 2. Product | Does the product facilitate meaningful interactions? | Tabiya |
| 3. User | Does the product positively support users’ thoughts, feelings, and actions? | Tabiya |
| 4. Impact | Does access to the product improve human development outcomes? | The program team or an external research partner, with Tabiya’s support |
Levels 1 to 3
Section titled “Levels 1 to 3”Tabiya handles the first three levels as part of a deployment.
At level 1, the AI is tested against general benchmarks and against benchmarks specific to the country and program. The code repository includes a set of test conversations that represent different users and situations. They are used to check each new version.
At level 2, Tabiya tracks whether the intended users engage with the product at a satisfactory rate.
At level 3, Tabiya examines whether using the product changes what users think, feel, and do in line with its purpose.
A deployment moves to level 4 after consistent success at levels 1 to 3.
Level 4
Section titled “Level 4”A level 4 evaluation establishes whether the platform causes changes in outcomes such as employment and income. It usually takes the form of a randomized controlled trial with data collection before and after. It needs its own budget line and planning that starts before the deployment begins.
The program team or an external research partner leads a level 4 evaluation, in close coordination with Tabiya.

