A voice-first AI copilot that cut Keppel’s research and planning time by ~80%

Keppel is a global asset manager and operator based in Singapore. It invests in and runs infrastructure, real estate, and connectivity assets.

Keppel’s teams spend much of their week on research: following the news, comparing investments, and preparing scenarios for the next decision. The AI tools available to them were hard to use and did not meet the company’s compliance requirements.

Keppel wanted one mobile app that would do three jobs. Surface what needs attention today. Answer research questions. Hand routine tasks, like updating the CRM, to an agent.

OnLoop built the app for iOS and Android around four ideas.

• Threads. Each piece of work, and every exchange between people and AI about it, lives in one thread instead of being scattered across chat, email, and documents.
• Voice first. Users speak to call up an agent, for research or for a task like a CRM update.
• Several models per thread. The app uses more than one large language model inside a thread to reduce hallucinations.
• Catch Me Up. A feed reads external news and alerts each user to the developments that matter to their work.

Strategic war-gaming, financial research, and exploring new investment opportunities used to take weeks. With the copilot they take hours, and manual research and planning time fell by about 80%.

Users rated it 8.1 or higher for satisfaction, the highest score of any AI app tested at Keppel.

In early testing, more than 80% of monthly users were active every week.