Director's Briefing: Deploying AI Voice Outreach for Loan Collections and Member Engagement in SACCOs

The Decision Facing Every SACCO Director

Your loan portfolio is growing. Your field officers spend hours each week calling members about overdue payments, new product offers, and KYC updates. Some members answer on the first ring. Others only respond to WhatsApp. A few have no data bundles at all. Meanwhile, your compliance team needs records of every outreach attempt — when it happened, what was said, and whether the member opted out.

This is the operational reality that makes AI voice outreach worth evaluating. Not because it's trendy, but because the alternative — manual calling, scattered spreadsheets, and incomplete audit trails — creates risk and leaves revenue on the table.

Wetu Voice addresses this with a specific architecture: record one message, deliver it across voice calls, SMS, and WhatsApp, and have AI agents that answer questions from your business knowledge while remembering each member's conversation history. This briefing walks through what a director needs to understand, decide, and ask their team before committing budget and credibility.

What Makes This Different from Bulk SMS or Robocalls

Bulk SMS platforms send the same text to everyone. Robocalls play a recording and hang up. Neither remembers that Mrs. Nakato asked about her loan balance last Tuesday, or that Mr. Okello requested a callback after 7 PM.

Wetu Voice works differently:

• Memory across channels: If a member receives a voice call about a new savings product, asks a question via WhatsApp two days later, then gets an SMS reminder, the AI agent recalls the full thread.
• Business knowledge integration: The agent answers from your actual policies — interest rates, repayment schedules, branch locations — not a generic script.
• Channel fallback: Members without data bundles still receive voice calls. Those who prefer WhatsApp get WhatsApp. One recording, three delivery paths.
• Built-in compliance: Opt-out and quiet-hour checks happen on every send. You see who heard, who replied, and what it cost.

For a SACCO, this means your "digital field officer" doesn't reset every interaction. It accumulates context the way your best human officers do — but at scale.

Where the Highest-Value Use Cases Live

Not every outreach problem needs AI voice. The strongest candidates share three traits: they're repetitive, they require consistent messaging, and they benefit from conversation history.

1. Loan Repayment Reminders and Collections

Current state: Field officers call members 3–5 days before due dates, then again on the due date, then after. Notes are scribbled in notebooks or entered inconsistently in the core banking system.

With Wetu Voice: A single recorded message goes out on schedule. The AI agent answers "How much do I owe?" "Can I pay via mobile money?" "What's my balance after this payment?" using your live loan data. Every interaction — promise to pay, dispute, request for restructuring — is logged with timestamps and transcriptions.

Director's question for the team: "Which loan products have the highest manual follow-up burden, and what's our current promise-to-pay conversion rate?"

2. Dormant Account Reactivation

Current state: Branches run quarterly campaigns. Staff call lists of inactive members, often reading from scripts they haven't updated in months.

With Wetu Voice: Segment members by last transaction date, product type, and branch. Deliver tailored reactivation messages — "Your savings account has been quiet for 90 days. Here's what you're missing." The agent handles "How do I reactivate?" "Are there fees?" "Can I do it from my phone?" and routes complex cases to a human.

Director's question: "What percentage of dormant accounts reactivate after our current outreach, and how many staff hours does each campaign consume?"

3. New Product Announcements with Eligibility Checks

Current state: Marketing sends generic SMS blasts. Members call the contact centre asking "Am I eligible?" "What's the rate?" overwhelming the queue.

With Wetu Voice: The announcement goes out. The AI agent answers eligibility questions using your product rules — "Yes, salary loans require 6 months of payslips" — and captures interest signals for your sales team to follow up.

Director's question: "How many inbound queries does each product launch generate, and what's our cost per qualified lead?"

4. KYC and Compliance Updates

Current state: Compliance teams chase members for expired IDs, missing tax numbers, or updated contact details. Deadlines slip. Audit trails are incomplete.

With Wetu Voice: Automated reminders with specific actions — "Your ID expires in 30 days. Visit any branch or upload via WhatsApp." The agent confirms receipt of documents, answers "What documents do you accept?" and escalates non-responsive members.

Director's question: "What's our current KYC completion rate before regulatory deadlines, and how many compliance hours go into manual follow-up?"

The Integration Reality Check

AI voice doesn't float above your systems. It needs data to answer questions and a place to write interaction records. Before approving a pilot, confirm these integration points:

System	What Wetu Voice Needs	What Your Team Must Provide
Core banking (Temenos, Oracle FLEXCUBE, or local)	Loan balances, repayment schedules, member contact info, product rules	API access or secure file drops; data mapping documentation
CRM or member management	Interaction history, segmentation tags, opt-out status	Field definitions; consent records
Mobile money / payment gateway	Payment confirmation, transaction references	Webhook endpoints or reconciliation files
WhatsApp Business API	Verified business account, template approvals	Meta Business Manager access; template library

Critical question for your CTO or IT lead: "Which of these integrations are already API-ready, and which require custom middleware? What's the realistic timeline for each?"

If your core banking system only exports CSV files nightly, the AI agent can't answer "What's my current balance?" in real time. It can still deliver reminders and capture responses — but the value ceiling is lower. Know this before you sign.

Consent, Compliance, and the Regulatory Landscape

Uganda's Data Protection and Privacy Act (2019) and the Financial Institutions Act govern how you contact members. Wetu Voice builds in quiet-hour checks (no calls before 8 AM or after 9 PM), opt-out handling on every channel, and full send logs. But the tool doesn't absolve you of responsibility.

Your compliance officer must verify:

• Consent records: Do you have documented consent for voice, SMS, and WhatsApp separately? A member who agreed to SMS may not have agreed to voice calls.
• Purpose limitation: Outreach must align with the purpose for which data was collected. Loan reminders — yes. Cross-selling insurance — only if consent covers marketing.
• Data minimisation: The AI agent only needs access to fields relevant to its use case. Don't feed it full member profiles if it only needs loan balance and phone number.
• Retention policy: Interaction logs (recordings, transcripts, SMS threads) are personal data. Define retention periods and deletion workflows before launch.

Director's question for compliance: "Can we produce a consent audit trail for every member in our pilot segment within 48 hours if the regulator asks?"

Pilot Design: Small Enough to Learn, Large Enough to Matter

A pilot that's too small gives false confidence. One that's too large exposes the institution to reputational risk if the experience is poor.

Recommended Pilot Structure

Scope: One loan product, one branch or region, 500–1,000 members. Duration: 6–8 weeks. Success metrics (agree before launch):

• Reach rate: % of members contacted successfully across any channel
• Engagement rate: % who responded (pressed a key, replied to SMS/WhatsApp, spoke to agent)
• Self-service resolution: % of queries answered by AI without human escalation
• Promise-to-pay or action rate: % who committed to a next step
• Opt-out rate: % who unsubscribed (should stay below 2%)
• Cost per successful contact: total spend ÷ members reached

Control group: Run the same outreach manually for a comparable segment. Compare outcomes.

Escalation protocol: Define exactly which scenarios route to a human — disputes, hardship requests, legal threats, data correction requests. Test this weekly.

Director's question for operations: "Who owns the pilot daily? Who reviews transcripts every Friday? Who has authority to pause the pilot if member complaints exceed a threshold?"

Training the AI Agent: Your Knowledge, Not Ours

The agent answers from your business knowledge. This means someone must curate and maintain that knowledge base. It's not a one-time upload.

What Goes Into the Knowledge Base

• Product terms: interest rates, fees, tenures, eligibility criteria
• Processes: "How do I apply?" "How long does approval take?" "Where do I submit documents?"
• Policies: late payment penalties, restructuring options, early settlement rules
• Branch and channel info: locations, hours, mobile money paybill numbers, WhatsApp numbers
• Regulatory disclosures: required statements for loan communications

Maintenance Rhythm

• Weekly: Review unanswered questions (the agent flags these). Add answers.
• Monthly: Update for rate changes, new products, policy adjustments.
• Quarterly: Audit a sample of conversations for accuracy and tone.

Director's question for product/marketing: "Who owns the knowledge base? Is it a full-time role, or shared? What's the SLA for updating rates after ALCO decisions?"

If no one owns this, the agent gives outdated answers within weeks. That destroys trust faster than no agent at all.

Measuring What Matters to the Board

Your board will ask: "What did this cost, and what did it return?" Prepare the framework now.

Cost Structure (Illustrative — Actual Pricing Follows Scoping)

• Platform access: monthly or annual licence
• Usage: per-minute voice, per-SMS, per-WhatsApp message
• Integration: one-time setup for API connections, knowledge base configuration
• Training: staff onboarding, pilot management

Return Framework

Metric	Baseline (Manual)	Pilot Target	Board Narrative
Cost per member contacted	UGX X (staff time, airtime)	40–60% reduction	"We reach 3x more members per shilling"
Promise-to-pay rate	Y%	+10–15 percentage points	"Collections efficiency up without hiring"
Staff hours redirected	Z hours/month	70% reduction in routine calls	"Field officers now handle complex cases only"
Compliance audit readiness	Manual spot-checks	100% logged, searchable	"Regulator requests answered in hours, not weeks"

Director's question for finance: "What's our fully loaded cost per field officer hour? That's the baseline we're comparing against."

Common Objections — and How to Test Them

"Our members are too old / too rural / too low-tech for this."

Voice calls work on any phone. No smartphone, no data bundle needed. SMS works everywhere. WhatsApp reaches the growing smartphone segment. The pilot data will show actual reach by channel — don't assume.

"Members will hate talking to a robot."

They hate being called at 10 PM. They hate repeating their story to three different officers. They hate unanswered questions. A well-trained agent that remembers them, respects quiet hours, and escalates smoothly often scores higher satisfaction than rushed human calls. Measure it.

"Our core banking system can't integrate."

Then start with use cases that don't need real-time data: announcements, KYC reminders, reactivation campaigns. Build the integration roadmap in parallel. Don't let perfect be the enemy of better.

"We don't have consent for voice calls."

Then the pilot uses SMS and WhatsApp only. Run a parallel consent campaign — "May we call you with important account updates?" — and measure opt-in rates. This is solvable.

The Team You Need Around This

AI voice isn't an IT project. It's a member experience project with technical dependencies.

Role	Responsibility
Executive Sponsor (you)	Sets success criteria, unblocks resources, owns go/no-go decision
Product Owner (operations/marketing)	Defines use cases, prioritises knowledge base updates, reviews transcripts weekly
Compliance Lead	Validates consent, audits logs, approves scripts, handles regulator queries
Technical Lead	Manages integrations, API keys, webhook endpoints, data flows
Member Experience Analyst	Tags conversations, identifies gaps, reports on sentiment and resolution
Escalation Team (2–3 officers)	Handles routed cases within SLA, feeds insights back to knowledge base

If any of these roles are "whoever has time," the pilot will drift. Assign names.

Questions to Ask Your Team This Week

1. "Which three member outreach processes consume the most staff hours but have the lowest consistency?" — Start there.
2. "Can we extract a clean list of 1,000 members with phone numbers, loan balances, and consent status by Friday?" — Data readiness is the hidden blocker.
3. "What's our current WhatsApp Business verification status?" — If not verified, start the Meta process now. It takes weeks.
4. "Who in compliance can review and approve the pilot consent framework in 48 hours?" — Speed matters.
5. "If the pilot works, what's the procurement process for a 12-month contract?" — Don't let procurement delay scale.

The Go/No-Go Checkpoint

At week 6 of the pilot, you'll have data. Convene the team. Review:

• Reach, engagement, and resolution rates vs. targets
• Opt-out and complaint volumes
• Knowledge base gaps and update velocity
• Integration stability (any missed webhooks, data mismatches)
• Staff feedback — are escalations appropriate? Is the handoff smooth?
• Member feedback — sample 20 respondents for qualitative input

If reach > 70%, self-service resolution > 60%, opt-out < 2%, and the team says "we can maintain this" — scale. If not, diagnose. Maybe the knowledge base is thin. Maybe the segment was wrong. Maybe the integration drops 30% of balance queries. Fix and extend 4 weeks, or pivot use case.

What Good Looks Like at Scale

Six months in, a well-deployed Wetu Voice implementation in a SACCO typically means:

• Morning: Automated repayment reminders go out. 85% reach. 40% engage. 15% promise to pay. Field officers receive a prioritised list of non-responders and complex cases by 9 AM.
• Midday: New savings product announcement hits 12,000 members. AI agent handles 300 eligibility questions. 50 qualified leads routed to sales.
• Afternoon: KYC expiry reminders sent to 2,000 members. 60% click the WhatsApp upload link. Compliance dashboard shows real-time completion rate.
• Evening: Zero calls after 9 PM. Opt-out requests processed instantly. Full audit log ready for any regulator request.

Your field officers spend their day on restructuring negotiations, new member onboarding, and relationship deepening — not dialling numbers.

Next Steps for You

1. Assign the pilot owner this week. Give them this briefing.
2. Schedule a 90-minute working session with operations, compliance, IT, and marketing. Walk through the pilot design table. Fill in your numbers.
3. Request a scoping call with the Wetu Voice team. They'll map your use cases to technical requirements and give you a concrete implementation plan — no generic proposals.
4. Set a board update date 10 weeks out. "We'll have pilot data and a scale recommendation by [date]."

The technology is proven. The variable is your execution discipline. That's the part you control.

Explore Wetu Voice →