Customer case study
Kuda reviewed 3% of customer conversations. Now it reviews 98% and saves 50+ hours a month.
Sampling roughly 3% of around 350,000 monthly conversations, Kuda’s QA team was surfacing issues days after they occurred. With Rulebase, the team reviews 98% of conversations across every channel, at the moment, saves over 50 hours of manual work every month, and has had more than 20,000 high-risk conversations flagged and routed automatically.
Kuda at a glance
- Customer
- Kuda — digital bank, 8M+ customers
- Industry
- Financial services / neobank
- Region
- Nigeria (Lagos), serving customers across Africa and the UK
- Team
- Service Measurement & Analytics, Customer Experience
- Scale
- ~350,000 support conversations a month
- Rulebase used for
- AutoQA, high-risk alerts, agent coaching, weekly reporting
The team behind 350,000 conversations a month
Kuda is one of the fastest growing fintechs in the world, serving over 8 million customers.
Its customer experience team handles around 350,000 support conversations every month across multiple channels, spanning frontline and back-office operations.
The Service Measurement and Analytics team, led by Francis Madubuogu within Kuda’s Customer Experience Department, owns quality assurance, performance management, and CX reporting for the operation.
Before Rulebase, the team ran on spreadsheets, then moved to a semi-automated QA tool that assigned conversations at random and covered roughly 3% of volume.
To keep pace with that scale, Kuda set out what it needed:
- Significantly wider coverage through an AI-first approach.
- Full integration across the support tool stack, with a vendor willing to build new integrations as Kuda adopts new tools.
- Full language coverage, so voice calls in Yoruba, Igbo, and Hausa could be transcribed, translated, and reviewed without depending on which languages an auditor spoke.
- Issues surfaced in the moment, rather than reviewed days after the conversation closed.
- Insight that drives action: which agents need coaching, which teams need support, and what to fix this week.
“We’ve gone from auditing 3% of interactions to having eyes on nearly all of them.”
Francis MadubuoguService Measurement & Analytics, Customer Experience, Kuda
The challenge
1. A 3% sample of a growing conversation volume
With coverage at roughly 3%, early signals of customer frustration often surfaced too late to act on.
2. Review cycles ran behind the conversation
By the time an issue was caught, it was often several days old, long enough for the same mistake to recur before anyone noticed.
Previously, an issue that went uncaught in the 3% sample could surface later as an escalation to senior leadership, by which point the customer had already had a poor experience. Rulebase now surfaces those cases while the conversation is still live.
“Those delays lead to escalations that reach top executives, but most importantly they affect the experience of the customer. Nobody wants to be on the back foot when that happens.”
Francis Madubuogu
3. The tooling created work instead of removing it
The previous solution integrated with only one support channel; analysts fetched the rest manually. Locating a single conversation worth auditing took up to 15 minutes. Conversations are predominantly in English, but calls held in Yoruba, Igbo, or Hausa could only be audited when a QA specialist who spoke that language was available, which limited who could review what. Reporting was manual, and the team spent more time processing data than acting on it.
First “aha!”
For Francis, it took two weeks.
Within the first two weeks, Rulebase was surfacing issues that manual review might sometimes miss, including agents deviating from the correct resolution path, verified against Kuda’s own internal SOPs and knowledge base.
Then came the volume. Over 20,000 high-risk conversations have been flagged for review since launch, conversations that would otherwise have gone unexamined.
“Despite the usual onboarding learning curve, we immediately started seeing issues that humans might sometimes miss. The value was there from the start and has only grown as we’ve calibrated the system.”
Francis Madubuogu
Results and ROI
Rulebase reviews 98% of conversations across every channel, flags high-risk issues in the moment, routes them to the right team in Slack, and turns QA data into targeted agent coaching. Two of those line items are pure time back: over 40 hours a month no longer spent hunting for voice calls to audit, plus the 2 to 4 hours a week that used to go into building the weekly report — more than 50 hours a month in total.
| Metric | Before | After (Rulebase) | Change |
|---|---|---|---|
| Conversation coverage | ~3% sample | 98% | Over 30x |
| High-risk conversations surfaced | Not systematically captured | 20,000+ flagged and routed | Net new visibility |
| Time to catch an issue | Several days after the fact | In the moment, routed via Slack | Immediate intervention |
| Finding conversations to audit | Up to 15 min each, manual | Eliminated — over 40 hours saved every month on voice calls alone | Fully automated |
| Weekly insight and reporting | 4 to 6 hours of manual analysis | ~2 hours a week | Roughly two-thirds less effort |
| Channel coverage | 1 integrated channel | All channels | Full integration |
| Agent coaching | Manual, dependent on an auditor happening across a relevant interaction | Per-agent, evidence-backed | Systematic |
“We actually have a clear lens into what’s happening and where we need to improve.”
Francis Madubuogu
What Rulebase does for Kuda
| Capability | What it does for Kuda |
|---|---|
| High-risk alerts | Flags conversations matching Kuda’s risk conditions in the moment, assigns them to the right team, and posts to Slack so work starts immediately. |
| Resolution-path QA | References Kuda’s internal SOPs and connected data sources to verify agents followed the correct resolution for each specific issue, not just grammar and tone. |
| Full-channel coverage | Integrated across every support channel, removing the manual hunt for conversations to audit. |
| Local-language coverage | Transcribes and translates voice calls in Yoruba, Igbo, and Hausa, so any auditor can review any call regardless of the language it was held in. |
| Ask the data | Team leads query Rulebase directly on what an agent is struggling with in a given week, powering targeted, evidence-backed one-on-ones. |
| Effort detection | Surfaces high-effort conversations where an agent spends too long on a single issue, creating an immediate coaching opportunity. |
| Weekly reporting | Delivers key issues, agent struggles, and trends every week, so Mondays start with a plan rather than a scramble. |
Qualitative wins
The QA team moved from oversight to problem-solving. “Instead of just passing feedback down to agents, we’re now actively helping resolve customer issues in real time.”
Coaching became systematic rather than incidental. Previously, targeted feedback depended on an auditor manually surfacing a relevant interaction. Now it is evidence-backed and tied to specific conversations or an observed trend, enabling personalised coaching for each agent and team.
Systemic patterns get escalated and resolved. Trends identified by Rulebase are taken to stakeholder review and addressed at process level, rather than handled case by case.
“What stood out was the agility of the Rulebase team. They felt more like a collaborator willing to build with us than a vendor selling a rigid tool. That partnership was the deciding factor. I’m glad we made that bet.”
Francis MadubuoguService Measurement & Analytics, Customer Experience, Kuda
The bottom line
Asked what breaks first if Rulebase disappeared tomorrow:
“I don’t want to think about it, to be honest. We’d go back to reactive work that doesn’t really help our organisation or our customers in the moment.”
And in one sentence, what does Rulebase do for Kuda?
“They help us serve customers better. Really.”
Francis Madubuogu
Key questions about ROI
How much of its customer conversations does Kuda review with Rulebase?
Kuda reviews 98% of its customer conversations with Rulebase, across every support channel. Before Rulebase, its QA team sampled roughly 3% of volume, so coverage increased by over 30x.
How many support conversations does Kuda handle each month?
Kuda’s customer experience team handles around 350,000 support conversations every month across multiple channels, spanning frontline and back-office operations. Kuda serves over 8 million customers.
What measurable results did Kuda get from Rulebase?
Conversation coverage went from a ~3% sample to 98%. Over 20,000 high-risk conversations have been flagged and routed since launch. Manually locating conversations to audit — up to 15 minutes each — was eliminated, saving over 40 hours every month on voice calls alone, and weekly insight and reporting dropped from 4 to 6 hours of manual analysis to about 2 hours a week. Together that is more than 50 hours of manual work returned to the team every month.
How does Rulebase review voice calls in Yoruba, Igbo, and Hausa?
Rulebase transcribes and translates voice calls held in Yoruba, Igbo, and Hausa, so any auditor at Kuda can review any call regardless of the language it was held in. Previously those calls could only be audited when a QA specialist who spoke that language was available.
How long did it take Kuda to see value from Rulebase?
About two weeks. Within the first two weeks, Rulebase was surfacing issues that manual review might sometimes miss, including agents deviating from the correct resolution path, verified against Kuda’s own internal SOPs and knowledge base.
What does Rulebase do differently from a traditional QA tool?
Rulebase reviews conversations against a company’s own SOPs and connected data sources to check whether the agent followed the correct resolution path, rather than only scoring grammar and tone. It flags high-risk conversations in the moment, routes them to the owning team in Slack, and turns QA findings into per-agent coaching.
Review every conversation, not a sample
If your team is auditing a fraction of what your customers actually experience, we’d love to chat. , or read more about AutoQA, Proactive Intelligence, and automatic coaching.