Grants 5 min read

The ECI Grant: Why Your First AI Project Should Be Small, Not Ambitious

Singapore's Enterprise Compute Initiative funds serious AI scale-up — but only for companies that have already proven they can build something that works. Here's the two-phase approach that gets you there.

The grant most SMEs read about backwards

Singapore’s Enterprise Compute Initiative (ECI) is one of the more generous AI support schemes available right now — up to S$150 million set aside at Budget 2025, delivered as cloud credits, tools and training from a major cloud provider, plus consulting costs co-funded at 70% (up to S$105,000 per project). The instinct, when a business owner first reads this, is to think of it as seed funding: “let’s use ECI to build our first AI project.” That instinct gets the order backwards, and it’s the single most common reason promising applications don’t go anywhere.

What ECI is actually asking you to prove

Buried in the eligibility criteria is a requirement that changes the whole strategy: applicants need “prior experience building a custom AI solution for a proof of concept or wider deployment.” In plain terms, DISG isn’t funding your first attempt at AI. It’s funding your next one — specifically, a bigger, harder use case from a company that’s already shown it can execute. Alongside that, you’ll need at least 10 Singapore-based staff and an in-house tech team of two or more in software, AI or data roles. This isn’t a programme designed for a company’s very first AI experiment.

The two-phase approach that actually works

The businesses that get the most out of ECI tend to follow the same pattern, whether they planned it deliberately or stumbled into it:

  • Phase 1 — build something small and real. A modest, self-funded proof of concept, tested on your own data by your own team.
  • Phase 2 — apply for ECI with a bigger, specific use case. With that proof of concept in hand, you’re no longer pitching an idea — you’re asking to scale something you’ve already validated. That’s a materially stronger application, assessed well against DISG’s stated preference for complex, scalable use cases.

Skipping straight to phase 2 without phase 1 isn’t just a weaker pitch — for most first-time-AI companies, it’s a bar they simply won’t clear yet.

What kind of project actually counts as phase 1

The projects that work best as a phase 1 proof of concept share two traits: they use data you already have, and they solve one specific, narrow problem rather than trying to transform the whole business at once. In practice, most Singapore SMEs land in one of these categories:

  • Customer service triage. A model that reads incoming emails, WhatsApp messages or support tickets and automatically tags or routes them by topic and urgency — well short of a full chatbot, but enough to prove the approach works on your actual customer language.
  • Document and invoice processing. Extracting line items, amounts and dates from supplier invoices or delivery orders, so your finance team stops retyping the same numbers into your accounting system every week.
  • Demand and inventory forecasting. A simple model trained on your past sales data to predict next month’s stock needs for your top products — useful on its own, and a natural stepping stone to a bigger, SKU-wide forecasting system later.
  • Lead scoring and sales prioritisation. Ranking inbound leads or quote requests by how likely they are to convert, based on patterns in your historical CRM data, so your sales team calls the right people first.
  • Internal knowledge search. A tool that lets staff ask plain-language questions over your own SOPs, product catalogues or past quotations, instead of digging through folders and group chats.
  • Quality or anomaly flagging. Flagging unusual orders, returns, or transactions that deviate from your normal patterns — a lightweight first step toward fraud or defect detection.

None of these need to be sophisticated. A model that’s right 80% of the time on a narrow, well-defined task is a far stronger phase 1 proof point than an ambitious platform that’s only half-built.

What happens once you’re in

If accepted, you’re matched to one of three Cloud Service Providers — Google Cloud, Microsoft, or AWS — each running its own version of the programme (Google’s AI Cloud Takeoff, AWS’s AI Springboard, Microsoft’s Agentic AI Accelerator), with cloud credits ranging from roughly S$150,000 to S$350,000 depending on the provider, plus the co-funded consulting. The actual build then happens over up to six months, delivered through a Consulting Partner from that CSP’s own pre-qualified list — a mix of major consultancies and specialist system integrators.

Where we fit into this

At Oasis, we’re increasingly doing both halves of this journey for clients: scoping and building the phase 1 proof of concept that gets a business ECI-ready, then preparing and filing the actual application once there’s something real to point to. We’re upfront about one thing: once you’re accepted, the formally funded build work sits with whichever Consulting Partner your assigned cloud provider pairs you with — not with us by default. What we do stay on for is making sure that funded build actually integrates cleanly into your business — your website, your CRM, your existing systems — and that the investment doesn’t end up as a well-funded pilot sitting unused in a cloud dashboard.

The practical takeaway

If you’ve been waiting for “the right grant” before touching AI at all, ECI isn’t it — not yet. Build something small first, from the list above. Prove it works. Then come back to the bigger funding with a stronger case than almost anyone applying cold.

Keep reading