AI for Public Programs: Faster Drafts, Same Accountability

Public outreach carries higher stakes than marketing SMS. Citizens need to know who is asking, what action is required, and what happens if they ignore the message. AI helps teams move faster without lowering that bar.

Good inputs: Pulse barrier summary, approved sender name, legal disclaimers, language preferences (English, Kiswahili, or both), and examples of messages that performed well in past cohorts.

What AI can draft well:

Two SMS variants with amount, deadline, or location in line one.

A short FAQ block for call-center scripts.

A/B test plan with primary metric and guardrails (complaints, opt-outs).

Journey outline — reminder timing, exit when action is recorded.

What humans must always approve:

Sender ID and branding consistency with prior official comms.

Claims about penalties, benefits, or personal data use.

Targeting rules—who receives which message and why.

Frequency caps and quiet hours.

Kenya-shaped checks: Does the link match the program citizens know (e.g. eCitizen, NHIF, county portal)? Is Kiswahili natural, not machine-translated? Is airtime cost considered for USSD steps?

A failure mode to avoid: generic "Dear citizen" blasts that look like phishing. AI drafts should include specific program name, deadline, and one trusted action—mirroring how successful commercial reminders state amount and date first.

Kilele AI sits between Pulse and Hub: it ingests diagnosis, produces execution-ready drafts, and later summarizes Hub results for the next review. The model does not replace program owners, legal reviewers, or field teams—it gives them a faster starting point.

Start with one program and one approved template family. Compare AI-assisted Variant B to your current control for two weeks. If action rate does not move, return to Pulse—trust or process may be the blocker, not wording.