Three years on the front line of compute supply and enterprise adoption reveal a clear pattern: which industries actually commit budget, and which show up for a proof of concept (PoC) and then disappear. The data doesn't lie. This article distills the front-line observations that Darren Su, CEO of Horizon AI, shared in his interview on Black Swan Academy, episode 286 (full video, in Mandarin).
The three fastest-moving industries share one trait: visible data
E-commerce and marketing adopt earliest. Commerce is full of measurable steps — creative production, ad optimization, conversion rates — so the effect of adopting AI tools shows up directly in the numbers. Decisions are fast and individual spend is small, which makes these teams ideal early users.
Healthcare and biotech are the best-fit industry. Every step of the pharmaceutical process is required by regulation to be fully documented, so digitization is inherently high — whenever someone poses a clear optimization problem, there is a tool to meet it. A real example: a biotech company needed to find the best ratio among five ingredients. It used to rely on traditional computation, grinding through the numbers slowly, and after adopting AI its compute spend peaked at NT$2.6 million (about US$80,000) a month. With a more sensible compute cost structure, the same work now runs at NT$500,000 a month. For biotech, where costs are heavy and time is money, getting an answer a month or two earlier is worth far more than the compute bill.
Finance has demand that is highly varied but clear: quant trading teams use compute to backtest strategy ideas found overseas, banks adopt AI for transaction-state recognition and fraud risk control, and support teams roll out AI customer service. Each problem has a specialist provider working on it.
The common thread across all three: the question is not "is this industry suited to AI" but "can this industry see the optimization in its data." AI is a value-add tool. It doesn't discriminate by industry; it looks for a clear problem to solve.
Messy data is no reason not to start
"Our data isn't cleaned up yet" is the most common reason to wait, especially in fields like healthcare where records and paper reports are mixed together. The reality is that cleaning and labeling data is not hard, just tedious — it takes people and process, not genius. The mature approach is for the adoption partner to bring in a dedicated data team or a specialist vendor (for example, providers whose recognition systems handle shipping slips or medical records) to take that tedious work off your plate, rather than letting it block the whole project.
The biggest variable in success or failure: culture, not budget
Give the same tools and the same budget to two companies, and the results diverge sharply. The biggest variable is organizational culture.
A lean team of five to ten people adopts AI fast — the person across the table can make the call, so you finish the conversation today and go live next week. A large enterprise needs a longer partnership: more departments, more layers, and the governance requirements of a public company mean every step carries communication cost. Neither is better or worse; the adoption strategy simply has to differ. Small companies push for speed, while large ones need to find a department willing to champion the effort internally, then use one success to persuade the rest.
The market's temperature is shifting too. In the first year, many companies ran a PoC and then went quiet — not because the tools failed, but because they hadn't settled on the problem. Since the second half of last year, demand has become noticeably clearer, and also sharply polarized: one group holds off no matter what, while another that has tasted the benefit pours in more resources to press its advantage. Of two companies in the same business, the one that moves first optimizes its processes at lower cost and overtakes the other in a short time — something that is genuinely happening now.
The right first step: spend NT$5,000 to save NT$10,000, then scale
Many AI adoption proposals open with a price tag of NT$500,000 or NT$1 million. Companies may have the money, but they are wary — especially those that have spent it before and failed.
The more sensible path is to experience the benefit of AI at the smallest possible cost first: start with a plan of around NT$5,000 (about US$150) a month, validate the optimization of one clear process, feel the "spend NT$5,000 to save NT$10,000" effect for yourself, and only then move to the second stage of scaling compute — with security, data deployment, and self-hosted environments as the third. Every stage of this three-year path should be funded by the value the previous stage proved.
One fact is often overlooked: the value of AI compounds. As data keeps feeding the system, the solution grows more accurate and the switching cost rises — which means that when you choose an adoption partner, what matters is not whose deck looks best, but who will still be sitting in your operations review three years from now.
Horizon AI is an enterprise AI adoption partner within the KONST group. It turns the group's advantages in compute and infrastructure into AI adoption that enterprises can actually use — end-to-end, from assessment and custom development to system integration, operations, and governance. Book an AI readiness assessment →
FAQ
Which industries should adopt AI first?
The most digitized industries, with the most complete data, move fastest: e-commerce and marketing have a short feedback loop on performance data, healthcare and biotech have a thorough documentation culture with a matching AI tool at every step, and finance has clear demand in quant trading, fraud risk control, and customer service. The common thread is optimization you can see in the data.
What is the most common reason enterprise AI adoption fails?
A proof of concept (PoC) gets built in the first year and then goes nowhere — not because the tools are hard to use, but because the company hasn't yet framed the problem clearly, and there weren't enough market examples to learn from. The next reason is organizational culture: the more people and departments involved, the more resistance to adoption, and the longer the partnership required.
How much budget should the first step of AI adoption take?
Far less than most companies assume. A reasonable start is a plan of a few thousand New Taiwan dollars a month to validate the optimization of one clear process. Once you feel the 'spend NT$5,000, save NT$10,000' effect, you can scale up gradually toward compute, security, and self-hosted environments. Committing NT$500,000 or NT$1 million from the outset carries far more risk than necessary.
Our data is messy — can we still adopt AI?
Yes, and messy data is the norm. The logic of data labeling and cleaning is not difficult; it is just tedious and labor-intensive. A mature adoption partner assigns a dedicated data team or a partner vendor to handle paper documents, reports, and unstructured data. It should never be a reason not to start.
Start with a free AI readiness assessment
In one hour, we map your data, systems, and use cases and give you an actionable AI adoption blueprint.
Book an assessment