EDGE Grant for AI Automation: 4 Tests Before You Apply

The EDGE Grant, defined

The EDGE Grant is Enterprise Singapore’s single grant programme for SME projects, which reimburses eligible SMEs for part of the cost of an approved activity after it is delivered and paid for.

Enterprise Singapore launched EDGE on 30 September 2026 to bring the former EDG, MRA and PSG routes into one programme. The catalogue now covers more than 150 activities across eight business areas, including Automation & Digitalisation and Financial Management. You do not apply for “a grant for AI”. You apply for one of those activities, and the activity sets the rules.

The money has limits. Enterprise Singapore’s EDGE FAQ says SMEs receive up to 50% support for most activities, with higher support for selected internationalisation and sustainability activities. Each company has an annual cap of S$100,000, and up to S$30,000 inside that cap can go to single-function digital solutions, integrated enterprise systems and pre-scoped equipment.

How it works, from first idea to reimbursement

Every EDGE application follows the same order. Skipping ahead is the most common way to lose eligibility.

• 1. Check the company. The applicant must be registered and operating in Singapore, with at least 30% direct or indirect ownership by Singapore citizens or permanent residents. The EDGE FAQ defines an SME as a group with annual sales turnover no higher than S$100 million, or no more than 200 workers.
• 2. Describe the job, then find its activity. Name the bottleneck, the systems touched, the human approvals and the result you will measure. Then match it to a published activity using the Enterprise Singapore catalogue or the BizSG recommender, or ask an SME Centre when nothing matches cleanly.
• 3. Gather the documents that activity asks for. For Large-scale Digital Transformation that means a project proposal, vendor proposals and quotations, the last financial year’s audited or management-certified statements, and outcome projections.
• 4. Submit before anything is signed or paid. An authorised employee applies through the Business Grants Portal using Corppass. A vendor or third party cannot submit or manage the application for the company.
• 5. Wait, or proceed at your own risk. Costs after submission may be supported if the application is later approved, but Enterprise Singapore warns that submission does not guarantee approval.
• 6. Deliver, pay in full, then claim. Follow the Letter of Offer, pay the approved expenses, keep demonstrations, reports, training records and payment proof, and submit the claim by its deadline. Reimbursement comes after the claim is approved.

Seen from a distance, those six steps reduce to four tests: an eligible company, a supported activity, clean timing, and credible evidence with enough cash to reach reimbursement. A failed test is usually a design problem, not a dead end. Ownership cannot be rewritten for one form, but a vague scope can be narrowed, maintenance can be separated out, a vendor contract can wait, and the manual process can be measured before automation begins.

Three routes an AI project can take

Use the narrowest published activity that matches the whole job. This is a screening aid, not an approval promise.

The chatbot activity shows how specific a pre-approved route is. Enterprise Singapore’s GenAI Customer Engagement Chatbot uses pre-approved vendors, supports software costs for up to 12 months, gives an application outcome in one week and requires completion within 18 months. Those terms belong to that activity alone. They do not carry over to a custom finance-automation proposal.

Worked example: a distributor’s invoice pile

Picture a typical Singapore distributor, a composite rather than a real client. Every supplier invoice arrives by email. Someone downloads it, keys the fields into a spreadsheet, then checks it line by line against the purchase order and the delivery record. When quantities or prices disagree, the finance lead chases an approval over messages. Once approved, the same data is typed again into the accounting records and a tracking sheet.

The owner’s first instinct is to ask a vendor for “an AI invoice bot”. As a grant application, that framing is weak. Enterprise Singapore excludes isolated task automation from Large-scale Digital Transformation, and turning one paper form into a digital one is the exact non-supported example on its activity page.

So the owner rewrites the scope around the whole operation. The workflow extracts invoice fields and links each document to the right record. Rules match invoice, purchase order and delivery record and flag differences in quantity, price or supplier. Exceptions outside the agreed tolerance go to a named approver with the source documents attached, and the decision is recorded. Approved results write to the accounting record and update the status. At every stage a person still owns a decision: reviewing unreadable documents, ruling on exceptions, approving or rejecting, and controlling access.

Now the numbers. Before anything is signed, the team records how long each step takes today, including the time invoices sit waiting in an inbox, and keeps sample documents, match results and approval timestamps as the baseline. Because this looks like a large-scale project, the application also needs a three-year projection of revenue growth, cost savings and overseas sales where applicable, plus a one-year projection of qualifying new jobs, each built from that measured baseline with the assumptions shown.

On the money side, the owner reads “up to 50%” carefully. Support for most activities is capped at that rate and sits inside the company’s S$100,000 annual cap, and none of it arrives at signing. The distributor must fund the whole project, finish it, pay the vendor in full and claim afterwards. So the owner also tests the case without the grant, against keeping the spreadsheet or hiring another administrator when volume grows. If the workflow only makes sense with a subsidy, it is not ready.

Only then does the owner ask Enterprise Singapore or an SME Centre which activity fits, because only the agency can decide supportability. Biks finance and accounting automation is built around the same boundary: software prepares the match and the evidence, and a person decides exceptions and approvals. That map is useful for choosing a vendor even if the grant route does not fit.

Myth: anything labelled AI qualifies

The technology is supporting evidence; the operational scope decides the fit. A large-scale project must affect growth and efficiency, strengthen core operations with modern technology, and integrate or standardise systems. A standalone bot that changes nothing around it does not meet that bar, however advanced the model.

Myth: you can start while the application is pending

You can, but only after you submit. Any work, deposit or payment to a supplier, vendor or third party before submission is what Enterprise Singapore calls retrospective, and it will not be supported. Committing after submission is allowed at your own risk, since approval is not guaranteed.

Myth: the vendor handles the grant

A vendor can help define the technical scope and write the quotation. The applicant company must own the business case, submit the application, pay and claim, and Enterprise Singapore restricts related-party vendor relationships. If a vendor offers to file on your behalf, that is a red flag, not a service.

Myth: a bigger scope wins a bigger grant

Padding a project to look transformational usually backfires. Large-scale Digital Transformation excludes maintenance costs, and an integrated system that bundles unrelated tools just to reach the function count is a warning sign. When nobody owns the process or can state the baseline, a narrower pilot paid without grant support is often the more sensible first step.

So should you apply?

Run the four tests: eligible company, supported activity, clean timing, and credible evidence with the cash to reach reimbursement. If all four pass, you have a defensible reason to apply. If one fails, fix the design first. For the wider funding picture and the move from EDG, MRA and PSG, see Singapore SME AI grants.