The honest view from inside technical leadership
AI News

AI News

Daily AI news with an engineering leadership take, each story credited to its original AI News report.

AI in the wild · AI News

Sarawak’s Voice AI Helpdesk Trial Took 4,056 Calls, and Only 1,225 of Them Were Real ICT Incidents

SAINS and 8nabler are planning more voice AI trials before deciding on a commercial deployment across Sarawak government services. The proof of concept behind that decision reports call volume. It does not report outcomes.

Line-art illustration of a telephone handset beside a funnel narrowing from calls to incidents, representing the Sarawak voice AI helpdesk trial

Sarawak’s voice AI helpdesk trial received 4,056 calls over two months, and 1,225 of them were genuine ICT incidents from civil servants, 8nabler CEO Benjamin Foong told The Borneo Post in an interview published on October 18, 2026. Foong gave no breakdown of how many of those incidents the AI resolved, passed to human agents or lost to callers who hung up. Sarawak Information Systems Sdn Bhd (SAINS), the state’s government IT company, and 8nabler now plan further trials before deciding whether to deploy voice AI across government services.

The Sarawak Voice AI Trial Served One Internal Helpdesk

The proof of concept put 8nabler’s voice AI on the internal ICT support line at SAINS. According to 8nabler’s own page for the partnership, the assistant serves SAINS staff, is not a public citizen-facing service and has not been rolled out to any agency. Callers spoke in mixed Bahasa Malaysia and English, and the system classified each problem and routed it to hardware, software or network queues. It ran around the clock, including weekends and public holidays. On the strength of that pilot, SAINS and 8nabler signed a memorandum of understanding on stage at the Digital Government Exchange 2026 (DGX26) in Kuching on September 10, 2026. 8nabler states that the MoU commits neither side to a system, a scope or a deployment, and sets a joint 12-week working period to assess what is practical.

8nabler’s Headline Figure Counts Calls, and 70% of Them Were Not Incidents

8nabler’s MoU page presents the pilot as 2,028 internal support calls and 4,702 AI-handled minutes a month, handled end to end with no human in the loop. Two months at 2,028 calls a month gives the 4,056 total. Foong’s figure of 1,225 genuine incidents means about 30% of calls were real ICT problems. The other 2,831 calls, by Foong’s account to The Borneo Post as reported by AI News, either contained no actual request or fell outside first-level helpdesk support. Those numbers describe traffic. They do not show how many staff got their problem fixed by the AI, how many were transferred and how many gave up, and SAINS and 8nabler have published nothing on that.

The Pilot Exposed Accent and Vocabulary Gaps That Public Services Would Multiply

Foong said the system had trouble with some Sarawak accents and with department names, internal application names and acronyms missing from its original training data. 8nabler’s page calls these calibration gaps in local vocabulary, accent and script order. The next phase is meant to test Iban, Mandarin and Tamil alongside mixed-language conversations, and 8nabler’s page sets out four themes to explore. They cover hosting models, voice processing and call records inside Sarawak, data handling aligned with Malaysia’s PDPA with personal data masked before it reaches a model and an audit trail on every conversation turn, wider language coverage, and a shared service layer across agencies. 8nabler presents each as a question for the working period, and none is a confirmed feature. Foong said decisions on welfare eligibility or individual rights would stay with human officers.

Why This Matters for Engineering Leadership

The Sarawak trial is a clean example of a pilot metric that cannot support the decision it is being used for. Call volume tells you the phone line worked. It says nothing about whether the AI helped anyone. Before you sign off on any voice or chat AI pilot, require a funnel with five numbers: contacts received, contacts that were real requests, requests resolved by the AI with the caller confirming the fix, requests escalated to a human, and contacts abandoned. Add a sample of escalated and resolved calls reviewed by your own support staff, because a ticket marked resolved by the AI is the vendor’s classification until someone checks it. Then compare against the human helpdesk on the same funnel for the same weeks. A vendor that can report 4,702 minutes a month can report how many of 1,225 incidents it closed. If that number is missing from the proposal, ask for it before the trial expands to callers who cannot simply walk over to the IT desk.

FAQ

What is the SAINS and 8nabler voice AI trial?

It is a proof of concept in which 8nabler’s voice AI answered the internal ICT helpdesk at Sarawak Information Systems Sdn Bhd (SAINS). It served SAINS staff only and has not been extended to the public or other agencies.

How many calls did the Sarawak voice AI handle?

It received 4,056 calls over two months, which 8nabler presents as 2,028 calls and 4,702 minutes a month. 8nabler CEO Benjamin Foong told The Borneo Post that 1,225 of those calls were genuine ICT incidents.

How many incidents did the voice AI resolve?

No figure has been published. Foong did not break down how many incidents the AI resolved, escalated to human agents or lost when callers hung up.

What problems did the trial find?

The system struggled with some Sarawak accents, department names, internal application names and acronyms. 8nabler describes the gaps as calibration of local vocabulary, accent and script order.

Will Sarawak deploy voice AI for public services?

Not yet decided. The MoU signed in September 2026 commits SAINS and 8nabler to a 12-week exploration with no commitment to a system, scope or deployment, and more trials come first.

How should teams measure a voice AI helpdesk pilot?

Report a funnel: calls received, calls that were real requests, requests resolved by the AI, escalations, and abandoned calls. Call volume alone says nothing about whether the AI helped.

Originally reported by Muhammad Zulhusni for AI News, October 20, 2026: Sarawak expands voice AI trials for government services. Primary sources: 8nabler, SAINS × 8nabler Memorandum of Understanding page (DGX26, September 10, 2026); The Borneo Post, “SAINS, 8nabler sign MoU to expand Voice AI for public services” (September 11, 2026). Incident figures are from Benjamin Foong’s October 18, 2026 interview with The Borneo Post, as reported by AI News.

Read more AI News