Your team uses AI. The company still does not. Start with a study
Monday morning meeting. Marketing says Copilot made English faster. Sales opens free ChatGPT and shows three follow-up emails. Finance holds a POC: scan an invoice, SKUs come out mostly right. The boss nods: “So we are using AI.” The operations manager in the corner stays quiet. Customers still chase quotes on WhatsApp. Customer service still copies cells into the old system. Everyone at the table types faster. Shipping and posting invoices are still slow.
Departments each run tools and POCs. It looks busy. Company results move when AI meets how you actually quote, approve and post. If it does not meet that path, only individuals got faster.
We wrote earlier that buying ChatGPT is not a strategy. This piece is the step after that: even paid seats and departmental pilots do not mean the company is using AI. Dropping a giant AI suite before you know which process actually needs it is a high-risk move. A Hong Kong SME can start with an AI consulting study, then implement.
People got faster. The company path can stay still.
The HKPC / Standard Chartered SME Index for Q1 2026 says 55% of local SMEs have used AI or plan to within a year. Among those already using it, only about a third pay for tools; most stay on free versions. Chatbots, document tools and OCR lead. The same survey notes that most use still sits on repetitive paperwork, with a gap before the potential is used. Those figures are easy to read as “we have caught up.” An operations manager often sees the other picture: personal accounts, quotes still waiting for a signature, invoices still typed in by hand.
A 2025 MIT study of enterprise generative AI makes the split plainer. Generic tools such as ChatGPT and Copilot feel useful to individuals, immediately. Custom or vendor enterprise systems often show no measurable, lasting P&L effect in the sample: about 60% of firms evaluated such tools, about 20% reached a pilot, about 5% reached everyday production. The report points at a mismatch with day-to-day work, not at model scores. Use the numbers fairly: they measure “no visible P&L,” not “all AI is useless.”
In a Hong Kong office it looks like this. Sales uses AI to draft a reply, then still waits for the boss to approve a discount, then still has customer service move the same words into MYOB. A finance POC reads an invoice well, then someone still matches the PO and still keys the old system. Both teams can report “we have AI.” The shipping and reconciliation the company wanted faster this year can be unchanged.
POCs and giant suites can both drain the budget
A pilot is nothing to be ashamed of. The risk is finishing it without asking: can it see the customer in WhatsApp, the warehouse and the accounts? How will you know it worked — for example, how much faster a quote leaves, how many fewer posting errors? Three departments, three pilots. The boss thinks the firm is ahead. The operations manager cannot say which pilot serves which part of the business. When more budget is asked for, everyone says theirs comes first. The company still has no one picture of progress.
On the other side, a vendor arrives with an “all-company AI platform”: six months, replace the systems, every department at once. It sounds tidy. After go-live, WhatsApp quotes still do not enter, the old accounts still need a person, staff go back to Excel to go faster. The money is spent; reversing is hard. HKPC’s 2025 workplace AI readiness survey lists integration with existing systems, and cost, among the main barriers. Buying the giant suite before a study is signing a large, expensive system before you know how you actually work — and it is hard to reverse.
A consulting study asks which process should move this year
When Frasertec Limited walks in, we do not open a product list first. We sit with the operations manager and department heads and list every AI already in the building: free accounts, paid accounts, departmental pilots, vendor quotes. Then we pick one path you actually want faster this year — for example enquiry to the books — and walk it. Where people wait, copy cells, mismatch numbers. Once you can count that, you know which stretch AI should meet, and which stretch can wait.
The study has to be readable by the boss and by the manager: how long a quote takes now, how many leave in a day, how many posting errors in a week; which numbers will show movement afterwards. Without those, the project becomes “everyone should use AI,” and three months later you still cannot say where the time went. Hong Kong’s privacy commissioner also reminds firms to name which generative-AI tools staff may use, and what must not be pasted into public tools. The study asks that too: where data stays, and what never goes on a public prompt.
Changing the systems can wait. Some steps an off-the-shelf tool can cover. Some need the Excel, WhatsApp and accounts you already use. Signing a giant suite in a hurry only adds a more expensive pilot. After the study, you can try one department, one process, and measure results there.
If people already use AI, a department already ran a POC, or a large-platform quote is on the desk, bring the slowest company process. That is enough. Free 30 minutes — we will sort whether you need another tool, or a formal study first.
References
- HKPC: Standard Chartered Hong Kong SME Leading Business Index, Q1 2026 press release (55% have used or plan AI; about 32% of users pay for tools)
- Same index, report PDF
- HKPC: AI Readiness in Workplace Survey 2025 (integration and cost among barriers)
- Fortune on MIT NANDA, The GenAI Divide (generic tools help individuals; most enterprise pilots show no measured P&L)
- MIT NANDA: State of AI in Business 2025 (PDF)
- PCPD: checklist on employee use of generative AI
