Where Production AI Actually Lives

Why traditional enterprises beat tech companies as AI consulting clients

Most articles about AI consulting point in the same direction: tech companies, startups, digital-native brands.

That direction is wrong.

Based on actual market demand, the real opportunity for AI deployment sits in logistics, shipping, insurance, finance, and manufacturing — the industries everyone calls "boring." And almost no one is competing for that market right now.

The UK government's AI Adoption Research, published in January 2026, found that only one in six UK businesses currently use AI, with natural language processing and text generation accounting for 85% of actual use cases. That kind of work — reading documents, writing reports, answering queries — is exactly what "boring" industries do every day.

The Tech Company AI Problem

Tech companies don't lack AI engineers. Hong Kong Productivity Council's Q1 2026 SME Index shows AI adoption in the information and communications sector at 92%, well ahead of every other industry in Hong Kong. These companies already have internal teams, MLOps pipelines, Kubernetes clusters, and a full technology stack.

They don't need AI consultants. They need more engineers — and that role is already hard enough to fill. IDC projects that by 2026, over 90% of global enterprises will face critical AI skills shortages, with combined losses from delayed products, lost revenue, and reduced competitiveness estimated at US$5.5 trillion. Tech companies and traditional enterprises are competing for the same talent pool.

Worth noting: building in-house doesn't guarantee success either. MIT Project NANDA's July 2025 research — covering over 300 real-world deployments and 150+ executive interviews — found that internally built AI projects succeed at roughly one-third the rate of projects bought from specialized vendors. An in-house team isn't a safeguard; it's just one variable.

The AI Reality in Traditional Enterprises

A Hong Kong shipping company receives dozens of PDF market reports every day, sorted by hand.

This is common across the industry. Global freight forwarding moves roughly US$19 trillion in cargo value each year, yet most operations still run on email, Excel, and manual data entry. Industry research shows leading third-party logistics providers now use AI to process over 60% of customs documents, bills of lading, and freight quotes, cutting manual data entry workload by more than 70%.

An insurance company handles customer inquiries by manually searching through hundreds of pages of policy terms.

Insurance adoption has clearly accelerated: McKinsey's 2025 analysis found the share of insurers with AI fully deployed rose from 8% to 34%. Yet industry-wide straight-through claims processing remains below 10%, and in claims processes that aren't automated, manual handling of unstructured documents can take up to 80% of total processing time. 64% of insurers now list unstructured document processing as a top AI investment priority — exactly the scenario RAG systems are built for.

A construction litigation law firm reviews thousands of pages of contracts by eye.

Legal is similar: a 2026 "AI in Professional Services" survey found 74% of law firms now use AI to assist with document review and 80% use it for legal research. But manual review still accounts for over 70% of total e-discovery spend in litigation, and 43% of legal organizations have no formal AI governance policy in place — the technology is in use; the governance hasn't caught up.

All three companies have real problems, real data, and real budgets.

What they're missing is one thing: someone who can turn the problem into a working system.

The Specific Gap in the Greater Bay Area

Hong Kong and the Greater Bay Area are a microcosm of this exact argument. Financial Secretary Paul Chan announced in June 2026 that the government will allocate an additional HK$300 million this year to expand the "Digital Transformation Support Pilot Programme," helping SMEs adopt AI and cybersecurity solutions. The Hong Kong Trade Development Council has also partnered with Microsoft Hong Kong on an "AI Adoption Programme" aimed at lowering the barrier for SMEs to access AI.

But there's still a gap between policy support and actual implementation. The "AI Adoption Index 2026," jointly published by Deloitte and the University of Hong Kong covering large enterprises in Hong Kong and mainland China, found that only 23% of companies have recorded a quantifiable financial return, just 4% have achieved "full transformation," and nearly 70% remain stuck in pilots or trials. The same survey found that 86% of AI users in Hong Kong rely on self-selected consumer tools for work — feeding documents directly into ChatGPT, for example — entirely outside company policy and data protection controls. Adoption looks widespread on the surface; underneath, it's hollow.

SME AI adoption in financial services and insurance sits at 62%, manufacturing at 60%. The logistics sector is benefiting from the "Smart Port" upgrade driven by Hong Kong's Transport and Logistics Bureau — the Port Community System (PCS) integrates sea, land, and air freight data, paired with the Single Electronic Lock (SELS) scheme to build a cross-border "green channel," with plans to extend into Guangxi and connect to ASEAN markets. The infrastructure and policy are in place. What's still missing is the person who can connect business processes to AI systems.

The Difference Between a Demo and Production

Most AI consultants deliver a demo: a polished ChatGPT wrapper with a PowerPoint deck attached.

This isn't unique to Hong Kong. MIT Project NANDA's research found that among enterprises deploying generative AI, 95% recorded no measurable P&L impact, with only about 5% actually capturing value at scale. The common causes come down to three things: unclear business metrics, data that isn't ready, and solutions never designed to run long-term in the first place.

A production-grade AI system is a different thing entirely:

• A scheduled AWS Lambda news-aggregation pipeline
• A fault-tolerant architecture that supports switching between multiple LLM providers
• A vector-database-driven retrieval system with support for Traditional Chinese embeddings
• A Stripe-integrated subscription SaaS with token cost management
• A system that's continuously monitored, patched, and improved after launch

A real production system doesn't need a consultant on standby every day after delivery. That's what traditional enterprises are actually willing to pay for.

Production-grade doesn't mean infallible: models still make mistakes, high-stakes decisions still need human review, and cost and accuracy still need ongoing monitoring after launch. These limits should be made clear to the client before the project even starts.

MCP and Context Engineering: The Next Entry Point

In 2026, a new set of keywords has started showing up in market demand: MCP (Model Context Protocol) and Context Engineering.

Put simply, this is the technical layer that lets AI agents access a company's internal systems in a structured way — order management, CRM, ERP, internal document libraries.

The growth rate of this layer shows up clearly in the numbers. Anthropic launched MCP in November 2024, when monthly SDK downloads sat around 100,000. By March 2026, that number had grown to 97 million — a 970x increase in 18 months. Gartner projects that by the end of 2026, 40% of enterprise applications will have built-in task-specific AI agents, up from under 5% today. Forrester also projects that 30% of enterprise application vendors will launch their own MCP servers within 2026.

Traditional enterprises happen to be the main battleground for this scenario: they have large numbers of internal systems and clear business processes, but lack the technical capability to connect the two.

That gap is precisely where an AI consultant's asymmetric advantage lies.

Why Almost No One Is Competing in This Market

The reason is simple: most AI engineers don't understand business, and most business consultants don't understand AI.

Very few people can do both of the following:

1. Understand the operational pain points of a logistics company
2. Turn that pain point into a system that runs on AWS, with controlled costs, running reliably after launch

That intersection is the real entry point into AI adoption for traditional industries.

And that entry point is growing fast. The global AI consulting services market was valued at roughly US$11.3–14 billion in 2026, projected to grow at a 26–27% CAGR, potentially reaching around US$73 billion by 2033. Banking and finance currently holds the largest market share, at 22.3%. The market isn't small — it's just that few people know how to enter it.

Closing

The AI consulting market isn't short on theory. It's short on people who can deliver production systems.

Traditional enterprises aren't short on budget. They're short on finding the right person.

This market is bigger than you think, easier to enter than you think, and has less competition than you think.

HKSoka focuses on exactly this intersection — taking traditional enterprises in Hong Kong and the Greater Bay Area from surface-level AI use to production-grade systems with real financial returns. If your company has similar document, process, or staffing pain points, let's talk.