AI Automation Funding for Singapore SMEs: EDGE Grant 2026
Published 22 November 2026 · 10 min read
PSG and EDG ceased on 29 September 2026. How Singapore SMEs now fund AI automation under the EDGE Grant, plus PDPA duties and a scoping checklist.
The short answer
If you are a Singapore SME looking to fund an AI or automation project, the scheme you were probably told to use no longer exists. Enterprise Singapore confirms that the Enterprise Development Grant (EDG), the Market Readiness Assistance (MRA) grant and the Productivity Solutions Grant (PSG) ceased on 29 September 2026, and that applications from 30 September onwards go through a single replacement called the EDGE Grant.
EDGE offers up to 70% support for SMEs and up to 50% for non-SMEs, capped at S$100,000 of total grant support per company per year across all activities, disbursed on a reimbursement basis. "Automation & Digitalisation" is one of its eight business areas. The practical consequence for anyone planning an AI project is that the old PSG mental model — pick a pre-approved tool from a list, claim 50% — is not the model you should be budgeting around any more. If you want help scoping a project that actually survives a grant assessment and a PDPA review, that is the work we do in AI operations automation.
What actually changed on 30 September 2026
For most of the last decade, Singapore SMEs funded technology in two tracks. Off-the-shelf software went through PSG — up to 50% of eligible costs, a S$30,000 ceiling, and a pre-approved vendor supplying a pre-scoped package. Larger bespoke projects went through EDG at up to 50% for local SMEs on a project-assessment basis. Overseas expansion went through MRA.
Enterprise Singapore's Business Refresh Package factsheet signalled the merger earlier in 2026, stating that from the second half of 2026 "businesses can look forward to a new grant – EDGE which streamlines three existing business grants – the Market Readiness Assistance (MRA), Productivity Solutions Grant (PSG), and Enterprise Development Grant (EDG) – into a single grant scheme", supporting all Singapore businesses, including non-SMEs, up to S$100,000 per year.
That has now happened. The EDGE Grant page describes coverage of eight business areas and over 150 activities, with the areas named on the page including Automation & Digitalisation, Business Strategy, Financial Management, Innovation, Internationalisation, Standards and Sustainability.
Old versus new, side by side
| Item | PSG / EDG (until 29 Sep 2026) | EDGE Grant (from 30 Sep 2026) |
|---|---|---|
| Status | Ceased to new applications | Current scheme |
| SME support level | PSG up to 50%; EDG up to 50% of eligible costs | Up to 70% for SMEs |
| Non-SMEs | PSG was SME-only | Up to 50% |
| Cap | PSG up to S$30,000 | Up to S$100,000 per year across all activities |
| Cap reset | Not framed as an annual refresh | Refreshed on 1 April each year |
| Payment | Claim after deployment and full payment | Reimbursement basis, claim after completion and full payment |
| In-flight projects | — | Existing EDG/MRA/PSG submissions continue to be processed; claims on completion |
Two details change how you sequence a project. The cap refreshes on 1 April, so a large programme can sometimes be split across two grant years. And reimbursement means you fund the project from working capital and recover afterwards — cash flow, not the headline percentage, is usually the binding constraint.
What the EDGE page does and does not say about AI
Be careful here, because a great deal of what is currently written about Singapore AI funding is either out of date or wishful. As published, the EDGE Grant page states the general support levels and the annual cap, and adds that support levels differ by activity and that you should refer to each activity for details. It does not, in its main page text, state a minimum project size, confirm that bespoke or custom-built AI systems are eligible, or set out whether pre-approved vendor lists still apply in the way they did under PSG.
So the honest position in November 2026: Automation & Digitalisation is explicitly an EDGE business area, and EDGE describes helping businesses automate operations and adopt new digital solutions — but the activity-level eligibility of a custom AI build must be confirmed against the specific activity listing and the scheme FAQ before you commit budget. Treat any vendor who states flatly that "your AI project is 70% funded" as making a claim they have not checked. Stratgik is a technology firm, not a grant consultancy or a law firm, and grant decisions rest with Enterprise Singapore.
The PDPA side, which is where most AI automation projects actually stall
Funding is the easy half. An AI automation system in a Singapore business is almost always processing personal data — customer records in a support agent, employee data in an HR workflow, supplier contacts in a finance process — and the Personal Data Protection Commission has now published direct guidance on it.
The PDPC's Advisory Guidelines on Use of Personal Data in Generative AI, issued 20 July 2026, is advisory and not legally binding, but it is the clearest available statement of how the regulator reads the PDPA in an AI context. Three points matter most to a buyer:
- Consent is the fallback, not the default. The guidelines note that organisations may consider relying on exceptions to the Consent Obligation or on deemed consent, but that where those alternatives are not applicable, "they must obtain consent". The exceptions commonly reached for — business improvement and research — are set out in the PDPC's separate Advisory Guidelines on Use of Personal Data in AI Recommendation and Decision Systems, and they are narrower than most people assume.
- Your generic privacy policy is probably not enough. For large-scale model training or fine-tuning, the guidelines state that "General Notifications are an insufficient means of obtaining consent to use User Data for Generative AI development" and that organisations "must provide AI-Specific Notifications for this purpose". They also encourage telling people how to decline or withdraw consent to AI training use.
- Deploying someone else's model still creates duties. The deployment guidance asks system deployers to be disciplined about specifying the intended purpose and the amount of personal data required, to track and designate responsibility over new data sources, to educate end users on what personal data should be input, and to develop clear written policies — noting that publishing those policies demonstrates accountability.
There is also a trap specific to AI projects that scrape or buy data. The guidelines treat the publicly available exception as genuinely conditional: paywalls, subscriptions and registration requirements are digital barriers, and the organisation must assess that data is publicly available before relying on the exception. "We found it on the internet" is not an assessment.
If your automation touches customer or employee records, run the data-protection design work in parallel with the build, not after it. Our Singapore PDPA compliance implementation work exists because retrofitting consent notices, retention rules and access controls into a live AI workflow costs far more than designing them in.
A realistic sequence for a funded AI automation project
The order below is deliberately grant-aware. The single most common way SMEs lose grant support is paying a vendor before applying, which was explicitly fatal under PSG and remains risky under a reimbursement model.
- Pick one process, measured. Not "AI for the company". One process with a current cost you can state: hours per week, error rate, response time. Our manual process cost calculator is a quick way to put a number on it.
- Decide build versus buy before you decide funding. A configured off-the-shelf tool and a custom system have very different grant paths, timelines and risk. Work it through with the build vs buy assessment.
- Map the personal data. What enters the system, what the model sees, where outputs are stored, who can read them, how long they are kept, and which PDPA basis you are relying on.
- Scope and cost the project properly. A written scope with deliverables, acceptance criteria and a maintenance plan. Grant assessors and your own finance team want the same document. The AI project cost estimator gives a defensible starting range.
- Apply before you pay. Submit through the Business Grants Portal with Corppass, under the relevant EDGE activity, and confirm activity-level eligibility rather than assuming it.
- Plan the cash flow. Reimbursement means you carry the full cost until claims are approved after completion and full payment. Budget for that gap.
- Decide who runs it on day 91. An AI workflow is a system with monitoring, drift, model updates and an audit trail — which is why most of ours end up under managed technology rather than handed over and forgotten.
Checklist before you sign anything
| Check | Why it matters |
|---|---|
| Baseline metric written down | Without it you cannot prove ROI or justify the claim |
| Activity-level EDGE eligibility confirmed | The headline 70% is not uniform across activities |
| No deposit paid pre-application | Retrospective spend is the classic disqualifier |
| Annual cap position known | S$100,000 is shared across all activities and refreshes 1 April |
| PDPA basis chosen and documented | Consent, or a specific exception you have actually tested |
| AI-specific notice drafted | A general privacy policy may not suffice for AI training use |
| Data residency and sub-processors listed | Cross-border transfer duties apply to your model vendor too |
| Human review point defined | Decides where the system may act and where it must only recommend |
| Exit and ownership terms | Who owns the prompts, the pipeline, the fine-tuned weights and the data |
| Run cost for 12 months | Inference, licences, monitoring and support, not just the build |
What AI automation is genuinely worth doing first
Across Singapore SMEs, the processes that reward automation earliest share three traits: high volume, structured-enough inputs, and a tolerable cost of being occasionally wrong with a human check in the loop. In practice that tends to mean document-heavy finance work, first-line customer support triage, quotation and order processing, and internal data retrieval — rather than anything that makes an unreviewed decision about a person.
The processes to avoid first are the mirror image: low volume, judgement-heavy, or regulated decisions about individuals. Those are where the PDPC's recommendation-and-decision-systems guidance bites hardest, and where a failed pilot poisons appetite for the next three projects. For a structured read on which of your processes qualify, use the AI automation opportunity scanner, and the Singapore market page sets out how we engage locally.
Frequently asked questions
Is the PSG grant still available for AI software in Singapore?
No. Enterprise Singapore's EDGE Grant page states that EDG, MRA and PSG ceased on 29 September 2026, and that from 30 September onwards businesses apply for grant support under the EDGE Grant. Projects already submitted or in progress under the old schemes continue to be processed, and claims can be submitted on completion.
How much of an AI automation project can an SME recover under EDGE?
The published support levels are up to 70% for SMEs and up to 50% for non-SMEs, with a cap of up to S$100,000 in total grant support per year across all activities, refreshed on 1 April each year. The page also states that support levels differ by activity, so the 70% figure is a ceiling rather than a flat rate, and you should confirm the level for the specific activity you are applying under.
Are custom-built AI systems eligible, or only off-the-shelf tools?
Automation & Digitalisation is one of the EDGE business areas, but the main scheme page does not state whether bespoke or custom AI builds qualify, nor whether pre-approved vendor lists still apply as they did under PSG. Confirm this against the specific activity listing and the scheme FAQ before committing budget. We will not tell you a custom build is funded when the published scheme text does not say so.
Do we need consent to use customer data in an AI system?
Often, yes. The PDPC's Advisory Guidelines on Use of Personal Data in Generative AI, issued 20 July 2026, state that organisations may consider relying on exceptions or deemed consent, but must obtain consent where those alternatives do not apply. For large-scale training or fine-tuning, the guidelines say general notifications are insufficient and AI-specific notifications must be provided. The guidelines are advisory and the PDPA itself prevails, so take legal advice on your specific facts — Stratgik is a technology firm, not a law firm.
Does it count if we only deploy a model someone else trained?
Yes, duties still attach. The deployment guidance asks system deployers to specify the intended purpose and the data actually required, to track and assign responsibility for new data sources, to educate users on what personal data should be entered, and to maintain clear written policies, with publication of those policies cited as a way to demonstrate accountability.
What does an AI automation project realistically cost to run, not just to build?
Budget for inference or API usage, model and platform licences, monitoring, periodic evaluation as models and data drift, and a named owner. A build cost with no run cost beside it is an incomplete number, which is why we quote both and why most clients move the running of these systems onto a managed arrangement after launch.
Related
If you are weighing up a first or next automation project, start with AI operations automation, see how engagements are structured in how we engage, and read the PDPA compliance implementation page if customer or employee data is involved.
Grant rules, support levels and activity lists change, and the scheme that replaced PSG is only weeks old. Verify figures against Enterprise Singapore before you rely on them. If you want a second opinion on whether a specific process is worth automating, what it should cost and what the PDPA exposure looks like, book a free 30-minute session with the Stratgik team and we will tell you plainly if the answer is "not yet".
Keep reading
More from Stratgik

21 Nov 2026
AI Ecommerce Automation in the UK: Rules, Cost and ROI 2026
Automating UK ecommerce support and returns with AI: the 2013 Regulations deadlines, DMCC and DUAA duties, real build co...

20 Nov 2026
Fixed-Price vs Time and Materials for Custom Software 2026
A 2026 guide to fixed-price vs time-and-materials software contracts: where each works, the clauses that matter, and who...

19 Nov 2026
MVP Development Cost in the UK in 2026: Rates and Rules
What an MVP really costs to build in the UK in 2026: developer-day bands, contract rates, IR35, R&D relief and the Febru...

