On con­struc­tion sites, shipyards and dams in 15 coun­tries, Ailyt­ics uses arti­fi­cial intel­li­gence (AI) to ana­lyse video from more than 1,400 cam­eras and flag safety haz­ards. The Singa­pore start-up says it had flagged 4.2 mil­lion haz­ards across 500 mil­lion square metres of work­space as of Janu­ary, includ­ing at Changi Air­port’s Ter­minal 5, a dam renov­a­tion in Japan and a bridge project in Aus­tralia.

While this is a less vis­ible use of AI than a chat­bot, it may say more about the com­mer­cial test facing Singa­pore’s AI start-ups.

Singa­pore drew 94% of South­east Asia’s ven­ture fund­ing in the first half of 2026, accord­ing to Tracxn. Whether that money pro­duces last­ing com­pan­ies will depend partly on AI start-ups’ abil­ity to stay use­ful as model pro­viders add fea­tures of their own.

This is why investors are increas­ingly look­ing bey­ond pitch decks and demon­stra­tions for proof that an AI start-up has solved a cus­tomer’s prob­lem more than once. Some are even ask­ing to speak dir­ectly with cus­tom­ers, Dar­ren Mowry, vice pres­id­ent of global start-ups and investor eco­sys­tems at Google Cloud, tells DigitalEdge.

Jeremy Tan, co-founder and man­aging part­ner of Tin Men Cap­ital, applies a fur­ther test, ask­ing whether a product is import­ant enough for a cus­tomer to pay to fix the prob­lem it addresses. His Singa­pore-based ven­ture firm backs B2B and fron­tier tech­no­logy com­pan­ies across South­east Asia, includ­ing Ailyt­ics.

“AI start-ups that focus on a luc­rat­ive pain point — medi­cine, not vit­amin — by going deep into one ver­tical have a much higher chance of con­vin­cing cus­tom­ers to adopt bey­ond a pilot,” says Tan. By medi­cine, he means a product that solves an urgent and costly prob­lem. A vit­amin may help, but the cus­tomer can get by without it.

An AI model, adds Tan, is only one part of the sale. Cus­tom­ers must also be able to use the AI solu­tion without dis­rupt­ing their busi­ness or expos­ing sens­it­ive data. “Deliv­er­ing value for an enter­prise involves much more than an AI model. Work­flow, data secur­ity, repeat­ab­il­ity, explain­ab­il­ity and reli­ab­il­ity all determ­ine whether a deploy­ment suc­ceeds.”

The ver­tical bet

Tan points to Groundup.ai and Pints AI, both in Tin Men Cap­ital’s port­fo­lio, to illus­trate his focus on com­pan­ies that go deep into a spe­cific industry.

Groundup.ai uses AI to mon­itor indus­trial equip­ment. It has signed US$45 mil­lion ($57.14 mil­lion) in oper­a­tional con­tracts over the past 18 months after rais­ing US$6.7 mil­lion over its life­time. Its largest con­tract, worth more than US$30 mil­lion over sev­eral years, was a repeat order from an exist­ing cus­tomer. As for Pints AI, it has helped 12 reg­u­lated fin­an­cial insti­tu­tions in four coun­tries gen­er­ate a com­bined US$10 mil­lion in sav­ings in under two years, adds Tan.

The legal and insur­ance indus­tries are also attract­ing AI start-ups. Mowry cites Har­vey, which applies AI to legal work that was pre­vi­ously manual. Mean­while, in insur­ance, AI start-ups are help­ing under­writers assess risk and pre­pare quotes faster. “There are agen­tic AI start-ups going into that space and redu­cing the amount of time it takes an under­writer to eval­u­ate risk and gen­er­ate premi­ums

and quotes [even though it is not a glam­or­ous use of AI],” he says.

Bey­ond wrap­pers

Mowry con­trasts those com­pan­ies with the founders he met two years ago, many of whom were build­ing what the industry calls wrap­pers, or thin soft­ware lay­ers around a large lan­guage model. “The vast major­ity of my con­ver­sa­tions [with AI start-ups] now are at the agen­tic AI layer. It will not be about wrap­pers any­more,” he adds.

Singa­pore cam­era-net­work start-up Lyte­house is among the AI com­pan­ies Mowry has in mind when he talks about mov­ing bey­ond wrap­pers. It uses exist­ing CCTV sys­tems to help retail­ers and indus­trial oper­at­ors, includ­ing Fair­Price Group, man­age theft, queues and safety com­pli­ance. It is expand­ing into logist­ics, real estate and indus­trial sites over­seas.

An AI start-up’s defence against a model pro­vider, there­fore, lies in pro­pri­et­ary data, industry and reg­u­lat­ory expert­ise, dis­tri­bu­tion, and the cus­tomer rela­tion­ship built through imple­ment­a­tion. “None of these are things that a model pro­vider adds in a release,” says Tan, adding that Ailyt­ics and Groundup.ai have access to data from live work­s­ites and indus­trial equip­ment that can­not be found online or bought off the shelf.

Des­pite South­east Asia’s many lan­guages, Tan notes that lan­guage alone offers little advant­age. “It is a hygiene factor and one we can­not do without, but it is not a dif­fer­en­ti­ator. A com­pany whose advant­age rests mainly on a coun­try’s lan­guage or reg­u­la­tions may build a strong local busi­ness but find it harder to expand bey­ond that mar­ket.”

Built to leave Singa­pore

When assess­ing a poten­tial start-up invest­ment, Tin Men Cap­ital looks for founders who intend to expand bey­ond Singa­pore from day one and have begun lay­ing the ground­work to do so. One sign is an AI product that addresses a prob­lem that is “not hyper-local” and has a size­able mar­ket bey­ond South­east Asia, Tan says.

That ambi­tion is com­mon among founders of Singa­pore start-ups, says Mowry, who con­trasts it with Amer­ican founders. “The mar­ket in the US is big enough that if they just built a busi­ness for the US, they’re prob­ably going to be okay. [But] when I meet founders here in Singa­pore, the reac­tion and the response is an imme­di­ate ‘go global’ [as they don’t have that option to the same extent].”

Another cri­terion is product-mar­ket fit, or evid­ence that a start-up is provid­ing something that addresses a real mar­ket need. “Rev­enue is not a good indic­ator of this. Without get­ting into tech­nic­al­it­ies, we want to see that around 40% of cus­tom­ers would kick and scream if the product were pulled out of the mar­ket,” adds Tan.

Tin Men Cap­ital also looks for “repeat­able and defens­ible gross mar­gins of about 80% for pure soft­ware and 70% for busi­nesses that com­bine soft­ware with hard­ware”, says Tan. “Cus­tom­ers in Singa­pore have high bar­gain­ing power. If the gross mar­gins work here, they should hold in other developed mar­kets such as Aus­tralia and Japan.”

When growth is not enough

Not everything wear­ing an AI label clears that bar. Tan is wary of firms whose rev­enue comes mainly from help­ing enter­prises imple­ment AI but present that work as a scal­able soft­ware product. “They are not ver­tical plays, the mar­gins do not behave like soft­ware, and fast user growth in that model is not the same thing as demand for a product,” says Tan.

He con­tin­ues: “The same cau­tion applies to any­thing sit­ting hori­zont­ally across every industry without going deep into one. Going back to the earlier point — if the offer­ing is a vit­amin rather than a medi­cine, the pilot rarely becomes a pur­chase order.”

Mowry ran his own ver­sion of that fil­ter when assess­ing start-ups. “It’s com­plex­ity, cost or eco­nom­ics, and then it’s scale.” On eco­nom­ics, he is less inter­ested in the cost of AI tokens than in what it costs a cus­tomer to com­plete a task. “Toke­nom­ics is yes­ter­day’s world. In an agen­tic AI world, we need to be think­ing about the cost of the task or the cost of the action,” he explains.

Singa­pore has become the region’s main des­tin­a­tion for AI ven­ture money. Whether it remains so will depend on its local star­tups turn­ing AI into products cus­tom­ers pay for repeatedly, rather than projects that demand more staff with each sale.

‍