Martha, the AI agent that cut 8 jobs at Zap, just got smarter

In February 2026, Zap Africa cut 44% of its workforce. The team shrank from 18 to 10 in a single round, wiping out roles in customer support, operations, and marketing. The culprit? Martha AI.

Launched months earlier, Martha is the brainchild of Moore Dagogo Hart, Zap Africa’s CTO and founder of deep-tech lab Cognito Systems. It was designed to automate fintech support at scale.

Now, Cognito has upgraded the system with what it calls an “empathy stack,” a four-layer pipeline that detects intent, language, and sentiment. The result, Moore told Condia, is an agent that can sense frustration and de-escalate it in real time.

“It can surface information about your customers and the kinds of activity they’re having,” he said.

That raises a harder question: if the first version eliminated eight roles, what does a more advanced one mean for the workforce?

“Martha isn’t built to remove jobs,” said Moore. “It enhances the individual who knows how to use AI. One person can now do the work of three or four.”

Inside Martha’s architecture

Martha is built around a four-channel integration strategy designed for high-friction fintech environments. Its most important distribution layer is WhatsApp, where many African customers already resolve financial issues informally.

Instead of forcing users into a new app, Martha plugs directly into WhatsApp, allowing customers to troubleshoot transactions inside a familiar chat interface.

For companies with established digital platforms, Cognito offers two web-based integrations: the widget, a low-code script that generates a floating support bubble on a company’s website, and the API, a deeper integration that embeds Martha directly into an app’s interface, preserving brand control while outsourcing support logic.

Martha also plugs into existing email systems, where it monitors inbound messages, drafts replies, and resolves routine tickets without human intervention.

Its most ambitious expansion, however, is into voice and SMS. Cognito can assign a dedicated phone number or integrate with a company’s existing line. When customers call, they encounter a voice agent trained to interpret tone, understand Nigerian Pidgin, and respond conversationally.

Behind all four entry points — WhatsApp, web, email, and voice — sits a four-layer processing pipeline. First, an empathy layer analyzes sentiment and intent. Next, language detection identifies linguistic context. Then, intent identification isolates the core problem from emotional noise. Finally, a security and PII masking layer strips sensitive data such as names, BVNs, and card numbers before processing.

While large-scale model inference may occur on global infrastructure, the sentiment and security layers remain locally controlled, according to Cognito. This hybrid design allows Martha to respond in under 30 seconds while keeping sensitive financial data within African regulatory boundaries.

Built for Africa

Martha interprets tone, cultural context, and informal speech patterns, including Pidgin and regional variations.

“We’ve built proprietary layers on top that help the system understand African languages and how people here expect to be spoken to,” Moore told Condia.

Congnito said Martha was trained on data from Black Pride Canada and J Bottoms in addition to conversations from Zap Africa, an inclusive design that has made it easy for the named organizations to integrate it into their workflows.

If that expansion holds, it marks a subtle shift. Instead of importing automation tools, an African-built system is being exported outward. The stress-test of the Nigerian market has become a global competitive advantage.

The jobs at risk

When Moore talks about the future of work, he doesn’t frame it as collapse, but restructuring.

“Jobs like entry-level customer support will not be needed as much as they used to be,” he said. “Any job that is repetitive is at risk.”

In the traditional fintech model, large support teams absorbed password resets, failed transfers, and account verification issues. Under an AI-led system, many of those queries are automated before a human ever sees them.

The roles most exposed are junior, process-driven positions. Higher-level support managers, Moore argues, are more likely to survive by supervising the logic, escalation thresholds, and decision trees that govern the AI.

“Learn to use AI to 10x your value and your output at work,” he said.

In Moore’s framing, the goal is no longer to be a task executor, but someone who understands how to configure, audit, and refine automated systems.

Still, he acknowledges the transition is disruptive. Founders, he says, should communicate early if AI adoption will reduce headcount. “It’s about being honest with people,” he said.

He also advocates for severance packages, AI training, and continued access to tools during the transition.

The marketing playbook

Cognito is leaning into founder networks and developer communities, distribution channels that have powered many of Africa’s fast-growing startups.

The approach aligns with what Condia previously described as the “new leapfrog playbook”: products spread through community trust, founder visibility, and integration into platforms users already inhabit, rather than through expensive top-down sales funnels.

For Martha, that means embedding inside WhatsApp, existing fintech apps, and community-driven developer ecosystems.

In this model, adoption is less about cold outreach and more about proximity: build where the users already are, and let network effects compound.