Only 19% of Companies Have Actually Scaled AI. The Gap Is Not Technology.

Sixty one percent of the organisations in Deloitte's new Thailand survey have implemented AI. Only 19 percent have actually scaled it across the business. The other 81 percent are still exploring, building proofs of concept, or piloting inside one function. The single biggest thing standing in their way is not compute, budget, or vendor choice. It is people: 71 percent name a lack of technical talent and skills as their main roadblock.

What did the survey actually find?

Deloitte Thailand's sixth annual digital transformation report surveyed 84 leaders across six industries between January and May 2026, with follow up interviews with executives. The headline is that adoption is up and outcomes are not. AI implementation rose to 61 percent this year from 47 percent last year. Digital maturity improved too, with the share of organisations in the "becoming digital" stage jumping 50 percent year over year.

Then the results arrive, and they are thin. Only 9 percent of respondents say their AI implementation achieved the outcomes they planned. Seventy one percent say they partially achieved them, with gaps remaining. Ten percent say they did not achieve them at all, and another 10 percent have no measurement in place to know either way.

So this is not a story about companies refusing to adopt AI. They adopted it. The bridge got built from both ends, and the two halves do not meet in the middle.

Why does the gap keep showing up in the same place?

Look at what respondents say is holding them back, in order: lack of technical talent and skills, 71 percent. Difficulty identifying use cases, 40 percent. Data readiness, 38 percent. Security of AI infrastructure, 38 percent. Lack of an AI strategy, 32 percent. Unclear return on investment, 27 percent.

Notice what is at the bottom of that list. Cultural resistance sits at 5 percent. Lack of executive commitment sits at 8 percent. Nobody is fighting AI. Leadership is not blocking it. The organisations in this survey want AI and cannot operate it, which is a very different problem from the one most AI vendors are selling a solution to.

The first two barriers are also the same barrier wearing two hats. If you cannot say clearly which business problem AI is supposed to solve, you cannot say what skills you need to solve it, so you hire and train against a vague target and end up with capability that does not connect to anything. One IT leader quoted in the report put it plainly: their function struggles to identify meaningful AI use cases because it has limited understanding of specific business needs, and needs guidance from the business units.

What happens when AI stays an IT project?

That quote points at the structural version of this problem. Nearly 70 percent of respondents say enterprise AI is treated primarily as an IT or technology led responsibility rather than a shared, cross functional effort.

When AI sits inside IT, the use cases that get built are the ones IT can build, not the ones the business most needs. Each team builds independently, adoption fragments, and the impact stays uneven. It is the organisational equivalent of two construction crews starting from opposite banks without agreeing on where the middle is.

The reinforcing data point is uncomfortable: only 14 percent of respondents report that AI is formally and fully included in leadership level KPIs. Twenty two percent say it is not part of leadership KPIs at all, and another 14 percent say it has been discussed but never made it in. Without a mandate at the top, AI stays a side project with a side project's resources, which is exactly where most of these initiatives are stuck.

Why is cost reduction the ceiling for most organisations?

Ask these companies what they actually got out of AI and the top three answers are cost reduction at 50 percent, faster or easier development of new systems and software at 48 percent, and efficiency and productivity improvement at 48 percent.

Now look at the bottom of that same list. Increased revenue: 9 percent.

That is the whole story of a cost first mindset. If AI enters an organisation as a way to spend less, it will succeed at spending less and stop there. Deloitte's own commentary in the report makes the point that service interactions traditionally treated as cost centres can be reimagined as growth channels, but that only happens if someone frames AI as a way to build something new rather than as a cheaper way to do the current thing.

The budget numbers say the same thing from another direction. Forty eight percent of digital transformation budget goes to improving business operations, 29 percent to customer experience, and 23 percent to new business models. Efficiency gets the money, so efficiency is what comes back.

Where is the training actually going?

This is the number that should stop any leader reading the report. Nearly 60 percent of organisations surveyed have given AI related training to 30 percent or fewer of their employees. Thirty percent of organisations have trained under a tenth of their workforce.

Set that next to the workforce plan. Fifty one percent of respondents say their strategy for the next three years is to keep headcount flat and upskill and reskill the people they already have. Half of these companies are betting their AI future on their existing staff while training almost none of them.

That is the ambition execution gap in its purest form. Not a technology failure. A promise made at the strategy level and not funded at the people level.

The report is also clear that training alone is not the fix. Deloitte's global research found organisations are roughly twice as likely to exceed their AI return on investment expectations when they redesign workflows and human machine interactions, rather than simply granting access to tools. Separately, only 14 percent of leaders say they are adept at shaping how humans and AI interact, and the ones who deliberately design those interactions are around 2.5 times more likely to report better financial results. Access is not adoption. Handing someone a licence and hoping is not a plan.

What does this mean for Bangladesh?

Thailand is further along this road than Bangladesh is. Its cloud and data foundations are more mature, its enterprises have been through more transformation cycles, and its AI adoption rate is higher. That is exactly what makes this report useful here rather than academic. It is a preview.

And the preview says the wall is not technology. A Bangladeshi company that buys the same tools will hit the same wall, at 71 percent talent and skills, except with a thinner local talent pool to hit it with. The tempting response is to assume the gap closes with better vendors or bigger budgets. The evidence in this report says it does not. It closes with people who can identify a real use case, own it from the business side rather than the IT side, and work alongside the system once it is running.

Bangladesh has a word for exactly this, and every one of us has used it. Ask when the training programme is starting. Ask when the data gets cleaned up. Ask when the pilot becomes a system people actually use. The answer comes back warm, confident, and completely non binding: হবে. Hobe. It will happen.

Seventy one percent partially achieved is just হবে with a dashboard attached.

There is a second, quieter lesson in the speed numbers Deloitte cites. The telephone took 50 years to reach 50 million users. The internet took seven. A leading generative AI tool reached about 100 million in two months. The window between "we should look into this" and "our competitors already did" is now measured in quarters, not years. Bangladeshi organisations do not get the long runway earlier technology waves offered.

This is the problem BIAI, the Bangladesh Institute of Artificial Intelligence, was built around. Not AI awareness, which Bangladesh already has plenty of, but applied capability: people who build and deploy working systems rather than collect certificates for attending lectures. Our curriculum partner Outskill reaches over 10 million learners across 45+ countries and is backed by mentors from Google, Microsoft, Amazon, Meta, and OpenAI, practitioners teaching what they actually do. On the enterprise side, BIAI Consulting starts the same way this report suggests it should start: audit how the business actually runs, find where AI genuinely belongs in it, and train the team that will own it, instead of installing a tool and hoping adoption follows.

So what should a company do first?

Not buy anything. Three moves, in this order.

First, name the business outcome before naming the technology. The 40 percent who cannot identify use cases are not short of AI ideas. They are short of a business problem specific enough to aim at.

Second, move ownership out of IT. If nearly 70 percent of organisations are getting IT led use cases and 71 percent are only partially hitting their targets, those two facts are related. The business function that owns the pain should own the use case. IT should enable the build, not choose what gets built.

Third, train more than the 30 percent. If the plan is to upskill existing staff rather than hire your way out, then the training budget has to match the tooling budget. Otherwise you have bought half a bridge.

The organisations in this survey are not behind because they were slow to adopt AI. They adopted it faster than anyone expected. They are behind because adoption was the easy half of the work, and the half that closes the gap is organisational, not technical. That half does not arrive in a software licence.