The Microsoft 365 Copilot licences have been handed out. And yet the only feedback coming back from the floor is some version of “I opened it, but I have no idea what to use it for.” Among Japanese manufacturers operating in Thailand, this is one of the most common conversations we are having right now. When Copilot business use stalls, the cause is rarely the capability of the tool. It is the absence of concrete work scenarios to point people at. This article works through the four everyday entry points where those scenarios actually live, which are Word, Excel, Teams and Outlook. If you would first like to go back a step and understand generative AI itself, our companion piece on what generative AI is and how it works is a good place to start.
Deploying Copilot and Using Copilot Are Two Different Problems
Over the past two years the conversation around Copilot has shifted. The question used to be whether to buy it. Now it is why nothing happens after you have bought it. Before going into the how, it is worth fixing where the market actually stands, in numbers.
Paid Copilot seats have passed 20 million, and large accounts have grown 4 times
In the FY26 Q3 results announced on 30 April 2026, Microsoft CEO Satya Nadella disclosed that paid Microsoft 365 Copilot seats had passed 20 million. Microsoft also stated that the number of large customers holding 50,000 seats or more had grown 4 times compared with a year earlier. Microsoft 365 commercial cloud revenue grew 19% in the same quarter, and in the segment commentary published for investors, Microsoft 365 Copilot was named alongside Microsoft 365 E5 as a driver of growth in revenue per user.
In other words, the decision to buy Copilot is no longer unusual. These figures mark a shift in the centre of gravity of the market, away from small pilots inside a single department and towards enterprise-wide agreements covering tens of thousands of users. Japanese manufacturers in Thailand are feeling that shift second-hand. In a growing number of cases, the head office in Japan signs a global agreement and the licences simply arrive at the local entity.
There is a wide gap between seats bought and seats used
Growth in licence numbers, however, does not automatically translate into growth in usage. Independent surveys report that only somewhere between 20% and 30% of purchased licences are actually being used on a weekly basis, and that workplace penetration sits in the low-to-mid thirties as a percentage. These are not primary figures published by Microsoft, so they should not be treated as settled fact. But they are not far from what we see on the ground either.
Among the companies we support, it is common to find that the IT department has licensed every employee, while the people genuinely opening Copilot each week are a handful of staff in general affairs and corporate planning. The licence cost continues to be billed per head every month, but the benefit lands on only a fraction of those heads. Let that run for six months and the conversation at the next renewal becomes predictable. Nobody can see the effect, so the seat count gets cut.
Handing out seats that go unused is not a failed deployment, it is a missing usage design
The important thing to establish here is that this situation should not be labelled a failed deployment. The deliverables of a deployment project are the contract, the permission model, the tenant configuration and the usage guidelines. In most of the cases we see, all of that was completed correctly. What is missing is the step that was supposed to follow, which is to define specifically who uses Copilot, for which task, and in what way.
This is a question about what happens after the tool is handed over, not a purchasing question. Think about installing a new measuring instrument in a plant. Nobody imagines that buying the instrument and bolting it down will improve measurement accuracy on its own. The value appears once you decide who measures what at which process step, what the pass and fail criteria are, and where the record is kept. Copilot behaves exactly the same way.
The deployment decision itself, meaning the comparison of pricing models, the permission design and how many seats of which plan to contract, is covered in detail in our article on how to plan a Microsoft Copilot rollout. This article picks up one step later, from the point where the licences are already in your hands.
What this article covers is the part that comes after the rollout
From here on, we take the applications that people open every working day, which are Word, Excel, Teams and Outlook, and translate what Copilot can do in each of them into concrete work scenarios. After that we look at the uses that are specific to manufacturing, the structural reasons adoption stalls, the questions that come up when rolling out at a Thai site, and the order in which to get started.
Copilot for Word Document Drafting
Word is the first entry point. Document work is where the benefit of Copilot is felt fastest, which also makes it the right place to start when you want people to try the tool for the first time.
Remove the time spent staring at a blank page
Give Copilot in Word the purpose of the document and its key points, and it will produce a structured first draft. An instruction such as “write safety training material for Thai staff covering pre-work inspection, protective equipment and the reporting procedure, about two A4 pages” comes back as a draft with headings and body text already in place.
The point is not to publish what Copilot produced. The most expensive part of writing a document is the time spent facing a blank page, working out the structure and getting the first paragraph down. Hand that part to the machine, and let people concentrate on whether the content is correct and whether it reflects the specifics of your own company. That is the realistic way to use it.
Get the gist of a long document before reading it
Open an existing document, ask for a summary, and Copilot will compress the argument and the main points into something you can read in a minute. Think of the 30-page policy document from head office, the draft contract that arrived from a customer, or the industry association report, all of them documents you previously had to read from end to end simply to decide whether they were worth reading at all.
A summary is an input to a decision, not the decision. For documents where a single word changes the meaning, contracts being the obvious case, use the summary to locate the relevant clause and then read that clause in the original. Make that the standing rule.
Use it as a first-pass translation
At a Thai site where Japanese, English and Thai are all in daily use, translation accounts for a large share of the workload. Rolling out an internal notice written in Japanese into English and Thai versions, or letting a Japanese expatriate manager understand a government document that arrived in Thai, are typical examples.
Machine output should not go outside the company unedited. But as a first-pass translation for internal distribution or for simply understanding what a document says, the quality is already good enough for practical work. Documents that used to wait for the translator to become available now start circulating the same day, and that change alters the rhythm of a business more than people expect.
Copilot Excel Use Starts With Natural Language Data Analysis
Excel is next. For the administrative functions of a manufacturer, Excel is effectively the core system, so any time saved here converts directly into capacity for the whole department.
Ask for aggregation and analysis without writing a formula
Select a table, give Copilot an instruction in plain language, and it will aggregate, sort, filter and run simple trend analysis. An instruction like “summarise this order list by customer and by month, and pull out the customers whose year-on-year figure has declined” goes through without touching a function or building a pivot table.
Until now, the only people who could do this were the handful of staff in the department who were strong in Excel. Requests concentrated on them, and when they took leave, the reporting stopped. The biggest value of Copilot Excel use is not the raw time saved. It is that it breaks this dependency on specific individuals.
Delegate the data cleaning
Data coming up from the shop floor is rarely tidy. Part numbers mixing full-width and half-width characters, customer names spelled several different ways, columns where you cannot tell whether a date is a date or a string. This preparation work consumes most of the time in any analysis task, and yet it is almost invisible as an achievement.
Copilot handles this reshaping from a plain-language instruction. Something like “normalise all part numbers in column A to half-width and strip trailing spaces” finishes faster than it takes to remember the steps of the replace dialog.
Numerical accuracy stays with people
There is a boundary to respect. Do not hand a process where numerical accuracy is the top priority, such as cost calculation or inventory valuation, to Copilot in its entirety. Generative AI is a mechanism for returning plausible results. It is not a mechanism for guaranteeing that a calculation is correct.
The practical division of labour is to leave the calculation itself to existing systems or to formulas you have already validated, and to give Copilot the explanation of the trend and the drafting of the report around those numbers. Tool-independent approaches to automating Excel work in general are a separate topic, but as far as Copilot is concerned, this is where the safe line sits.
Copilot Teams Meeting Notes Remove the Work After the Meeting
Teams is the third entry point. The benefit becomes much easier to understand if you stop thinking about the meeting itself and look instead at the work that happens before and after it.
Extract minutes and decisions automatically from the recording
Record and transcribe a Teams meeting, then run Copilot over it, and you can automatically extract a summary of the discussion, the decisions taken, the points left unresolved and the tasks assigned to each participant. People who could not attend can catch up on the essentials, and meetings can run without appointing someone to take minutes.
In a case study published by a Japanese vendor, a global company that had been taking several days to produce minutes for English-language meetings was able to have them ready immediately after the meeting ended. At a Thai site where Japanese, English and Thai are mixed inside a single meeting, that effect is larger still.
Nippon Steel logged roughly 20,000 AI meeting notes in one month after expanding
Nippon Steel provides a useful sense of scale. According to the customer story published by Microsoft, in the month following the company’s expanded enterprise-wide rollout in February 2025, AI meeting notes for Teams meetings were used approximately 20,000 times. Over the same period, more than 50,000 prompts were submitted to Copilot Chat, largely for tasks such as searching internal files.
Looked at another way, this tells you that post-meeting processing, an activity nobody had ever bothered to count as work, was in fact occurring that many times across the organisation. Even at 15 minutes saved per instance, the monthly total is substantial.
If you want to compare meeting-minutes AI across tools
Automated minutes is a field with many dedicated tools beyond Copilot. If you also run meeting platforms other than Teams, or if transcription accuracy and speaker separation are hard requirements, see our cross-tool comparison of AI meeting minutes solutions. In this article, Teams Copilot is positioned as the option for companies already on Microsoft 365 that want to keep everything inside the platform they already pay for.
Outlook Is About Summarising Long Threads and Drafting Replies
Outlook is the fourth entry point. The benefit looks modest on paper, but it is the area that touches the largest number of people.
Summarise threads that have piled up
Open a long-running email thread, ask for a summary, and Copilot returns the history, the current point of contention and the items left pending. This lands hardest when you come back from leave to several hundred unread messages, or when you inherit a case midway and cannot work out how it got to where it is.
In the Nippon Steel case, roughly 4,500 email thread summaries were run in the month after the expanded rollout. That is second only to AI meeting notes in volume, which says a good deal about how much of the working day email processing occupies.
Hand over reply drafts and calendar entries
Generating a draft reply to an incoming message is also good enough for practical use. An instruction such as “ask to reschedule, in polite English” comes back as something you can work from. Treat it as a draft rather than a finished item, and make it a rule that a person reads it before it is sent.
Copilot can also take a date and time mentioned in the body of an email and create the calendar entry from a single instruction. It is a small feature, but it is exactly this accumulation of small savings that turns Copilot into something people open every day.
Copilot Business Use Points Specific to Manufacturing
Everything above applies to administrative work in any industry. From here we move into the uses that are particular to manufacturing.
How Nippon Steel expanded in stages
Nippon Steel began evaluating Copilot in February 2024 and ran a pilot of 300 seats for four months from April of the same year. On the strength of those results, the company expanded to 4,400 seats at the licence renewal in January 2025, giving every one of its more than 100 departments the opportunity to use Copilot. The rollout later grew to 11,000 seats across the group.
What stands out is that the company did not hand seats to everyone at once. It ran 300 seats for four months, came away with concrete usage patterns, and only then expanded. That sequence is decisive in avoiding the handed-out-but-unused problem discussed below.
Programming support for equipment troubleshooting
The interesting detail in the Nippon Steel case is who sits at the centre of usage, namely citizen data scientists and members of the company’s Azure OpenAI Service evaluation community. Copilot is not confined to administrative document work there. It is also used to support the development of image-analysis programs for equipment troubleshooting.
On a production site, small analysis needs that no off-the-shelf software covers come up constantly. Someone wants to isolate the vibration waveform of one particular machine. Someone wants to count only the surface area of defects in inspection images. These requirements are too small to outsource, and the engineers on site do not have spare hours for programming. Copilot works as the tool that fills exactly this gap.
Comparing specification documents, an unglamorous but heavy task
The other use Nippon Steel cites is comparing differences between specification documents. A specification before and after revision, an internal standard against a customer requirement specification, a Japanese version against an English one. In manufacturing, the task of putting two nearly identical documents side by side and finding what changed comes up constantly.
The work demands sustained concentration, is rarely recognised as value-adding, and produces expensive rework when something is missed. Have Copilot do the first pass and have a person verify the points it flags, and you reduce both the workload and the miss rate at the same time.
Across all of these initiatives, the company expects efficiency gains on the order of tens of thousands of hours per year.
Why the Seats You Handed Out Are Not Being Used
Now back to the gap we opened with. The cause is not a lack of motivation among staff. It sits in a few structural conditions.
No work scenarios have been put in front of people
This is the single biggest cause. A member of staff who receives a licence with nothing more than “it is useful, give it a try” has to discover on their own where in their job Copilot fits. That is genuinely difficult work to ask of someone.
What works is presenting three or four concrete work scenarios per department, each with an example prompt. For quality assurance, that means going down to the level of “translate a customer complaint report into English”, “draft a corrective action report” and “summarise an audit record”. The difference between an abstract feature tour and a prompt someone can paste tomorrow morning shows up immediately in first-use rates.
People never reach a first success
With generative AI, the quality of the output varies enormously with the quality of the instruction. Someone who gives a vague instruction first, gets a disappointing answer, and stops experimenting will conclude the tool is not much good. In most cases that verdict belongs to the instruction, not to the tool.
Even a short training session changes the impression, if it does nothing more than put a good prompt and a bad prompt side by side. Set the purpose of the training as creating a first success, not as covering every feature.
Nobody is looking at usage data
Most companies that hand out seats and stop there have no idea who is using how much. Without visibility, you cannot copy the practices of the departments that are growing, and you cannot intervene in the departments that have stalled.
Usage data is available in the administrator reports. Look at it by department rather than as a company-wide average and you will almost always find wide variation. That variation itself tells you what to do next.
If you want to build a systematic adoption programme
All three of these points are common to generative AI tools in general, not just to Copilot. If you want to design the training, decide how to staff internal champions and set the metrics for measuring effect, our article on supporting AI adoption and making it stick covers that ground in detail. Here we stay with the issues specific to Copilot.
Questions to Settle Before Rolling Out Copilot at a Thai Site
Bring the head office playbook to Thailand unchanged and you can run into walls you did not plan for. Three points are worth settling in advance.
How to position Thai-language use
Copilot handles input and output in Thai, but the output quality differs from Japanese or English. How far Thai staff can rely on it in daily work depends on the nature of that work.
The split that works in practice is to use Thai for internal drafts and summaries, and to generate anything that leaves the company in English for a person to review. Separately, using Copilot to take a document a Japanese expatriate wrote in Japanese and roll it out in Thai delivers high value as a bridge between languages. The answer is neither “Thai does not work” nor “everything can be done in Thai”. It is to draw the line by use case.
Choosing a plan and thinking about cost
On the Microsoft Thailand pricing page for Microsoft 365 Copilot, the pricing published is for two add-ons that sit on top of an existing Microsoft 365 licence. There is a Copilot Business add-on aimed at small and medium-sized businesses, and a Copilot Enterprise add-on for large organisations with no cap on the number of seats.
The condition to watch is that these two are not priced the same way. The Copilot Business add-on carries a limited-time discount running from 1 July 2026 to 30 September 2026, during which it is listed at USD 18 per user per month on an annual commitment, against a standard price of USD 21. The Copilot Enterprise add-on, the one with no seat cap, is listed at USD 30 per user per month on an annual commitment and is excluded from that limited-time discount. For both, paying monthly rather than annually raises the unit price.
The practical consequence is that if you are contemplating a rollout of several thousand seats along the lines of Nippon Steel, the number to plan against is the large-organisation rate, not the discounted small-business rate. If a renewal or a seat-count review is coming up, decide first which add-on you are budgeting against and whether you are quoting annual or monthly, before arguing about the total.
Plan comparison, estimating the seats you actually need and the permission design that goes with them are deployment decisions, and they are covered in detail in our article on Microsoft Copilot cost and permission design.
Security and data handling
Two issues are particular to a Thai site. One is compliance with the local personal data protection law, the other is the design of what data is shared with the head office in Japan. Because Copilot generates its answers by referring to data inside your own tenant, your existing permission model, meaning who can already open which file, is reflected directly in what Copilot will tell people.
Put differently, roll Copilot out company-wide on top of a loose permission model and files that were previously invisible in practice become visible through search. If permissions were not reviewed at deployment time, review them before you widen usage. This is an operating-rules problem, not a technical one.
The other question that comes up quickly from inside the business is data residency. What Copilot refers to is data inside your own Microsoft 365 tenant, and which region that data is stored in, along with where processing takes place, is determined by your tenant configuration and your contract terms. If you are being asked to demonstrate alignment with Thai personal data protection law or with a data management policy set by the head office, confirm your own tenant’s storage location before widening the scope of use and have the answer ready for the internal briefing. That conversation moves much faster with the answer already in hand.
How to Get Started With Generative AI Efficiency Gains
Assuming the licences already exist, here is the order in which to bring real usage to life. The Nippon Steel rollout, as it happens, followed the same sequence.
Start with small work scenarios
The first move is to narrow the target, not to go company-wide. Pick one or two departments, and within them choose about three tasks that are frequent and repetitive. Customer complaint reports in quality assurance, quotation comparison tables in purchasing, multilingual rollout of internal notices in general affairs. That is the right level of granularity.
Fix a window of one to three months and record what was shortened and by how much during that period. Those records become the material that gets the next expansion approved.
Grow the number of people who can use it
Once the trial has surfaced usage patterns that work, turn them into prompts and roll them out to the other departments. What matters most here is designating one person per department who others can ask. Whether the small questions that do not justify a ticket to the IT department get resolved on the spot is what determines whether usage continues.
Expand the rollout
Move to increasing the seat count only once usage has taken hold and you can show the effect in numbers. Reverse that order and the cost arrives first, and the renewal date comes around while the benefit is still invisible. Nippon Steel’s progression from a 300-seat pilot to 4,400 seats and then to 11,000 seats across the group can be read as an example of keeping that order intact.
Checklist for Whether Your Copilot Was Simply Handed Out
Use the following items to assess where you stand. The more of them that do not apply to you, the more room there is to work on usage design.
- Each department has at least three concrete Copilot tasks defined, each with an example prompt
- Company usage can be seen at department level, not just as an overall average
- You can state, as a number, the share of employees who open Copilot at least once a week
- Some form of training or guide on how to write instructions is available, even a short one
- Each department has a designated person others can ask about usage
- Time saved through Copilot is being recorded in some form
- A review of the permission model was carried out before or after the Copilot rollout
- There is an agreed policy on which of Thai, English and Japanese Copilot is used for
- You have the material in hand to decide the seat count at the next licence renewal
If more than half of these are a no, the return on investment is higher from investing in the usage design of the licences you already hold than from buying more of them.
Frequently Asked Questions
Does Copilot really go unused if you only deploy it?
Yes, that is the pattern. Microsoft’s announcement of 30 April 2026 took paid seats past 20 million, while independent surveys report that only around 20% to 30% of purchased licences are actually used on a weekly basis. The causes come down to three things, which are that no concrete work scenarios have been presented, that users never reach a first success, and that nobody is monitoring usage. Treat licence distribution and usage design as two separate pieces of work.
How much working time can the Copilot Teams meeting notes feature save?
As a reference point on volume, Nippon Steel logged approximately 20,000 uses of AI meeting notes for Teams meetings in the month following its expanded enterprise-wide rollout in February 2025, and the company expects efficiency gains on the order of tens of thousands of hours per year across its initiatives overall. The size of the saving depends on the nature of your meetings and the standard your minutes have to meet, so the practical approach is to measure the post-meeting processing time for a single meeting in your own company and estimate what share of it is displaced.
How much does Copilot cost?
The Microsoft Thailand pricing page for Microsoft 365 Copilot publishes two add-ons. The Copilot Business add-on for small and medium-sized businesses is listed at USD 18 per user per month on an annual commitment under a limited-time discount running from 1 July 2026 to 30 September 2026, against a standard price of USD 21. The Copilot Enterprise add-on, which has no cap on the number of seats, is listed at USD 30 per user per month on an annual commitment and is not covered by that discount. Paying monthly raises the unit price in both cases, so the real burden depends on the size of your organisation, the payment terms and the timing. How to select a plan and how to think about the number of seats you need are covered in the deployment article.
Can staff who do not know Excel formulas still use Copilot Excel?
They can. Select a table and give an instruction in ordinary English or Japanese, and aggregation, filtering, sorting and simple trend analysis all run. In fact, breaking the pattern where every reporting request lands on the handful of Excel-literate people in the department is the main value of Copilot Excel use. That said, processes where numerical accuracy is the top priority, cost calculation being the clearest example, should stay with validated formulas or existing systems, with Copilot writing the explanation of the results.
Can Thai staff use Copilot in Thai?
Copilot supports input and output in Thai. Output quality does differ from Japanese and English, however, so drawing a line by use case is the realistic approach. Using Thai for internal summaries and drafts, and generating anything that leaves the company in English for a person to check, is the split that works. Rolling out an internal document written in Japanese into Thai is a use case where the benefit is particularly easy to obtain.
We have already handed out Copilot. Can we still turn it around?
You can. The sequence is to check usage at department level first, then collect the specific practices of the departments that are using it. Turn those into prompts and spread them to the departments that are not. Circulating examples from inside your own company that are demonstrably working moves people further than a company-wide training session. Nippon Steel also accumulated usage patterns during its 300-seat pilot before expanding to 4,400 seats, and the same sequencing logic applies here.
Summary
Here are the points worth carrying away.
Microsoft’s announcement of 30 April 2026 took paid Copilot seats past 20 million, and customers holding 50,000 seats or more grew 4 times compared with a year earlier. The decision to buy has become ordinary. At the same time, independent surveys report that only around 20% to 30% of purchased licences are actually used weekly, which means a gap persists between seats distributed and seats used.
That gap is not a failed deployment. It is what happens when no usage design exists. The four entry points for closing it are drafting, summarising and translating in Word, natural-language aggregation and cleaning in Excel, minutes and decision extraction in Teams, and thread summaries and reply drafts in Outlook. In manufacturing, further uses hold up beyond administrative work, including programming support for equipment troubleshooting and comparing differences between specification documents.
Nippon Steel started with a 300-seat pilot in April 2024, moved to 4,400 seats in January 2025, and later expanded to 11,000 seats across the group. In the month after the expanded rollout, the company recorded approximately 20,000 AI meeting notes for Teams meetings, approximately 4,500 email thread summaries, and more than 50,000 prompts submitted to Copilot Chat, largely for internal file search. Trying small, accumulating usage patterns and only then widening is the sequence behind those results.
When rolling out at a Thai site, settle three things first, namely where the line falls on Thai-language use, which plan and cost basis you are working from, and a review of the permission model. And before buying additional licences, check department by department whether the licences you have already distributed are actually being used.
Perhaps your Copilot licences have already been distributed but the floor is not getting value from them. Perhaps you want to organise usage around the work scenarios that are specific to a Thai site. Either is a fine place to start a conversation. TOMAS TECH works from the reality of Japanese manufacturing sites in Thailand, and we are happy to begin with the simple question of which tasks it is realistic to tackle first. If you would like to talk it through, please get in touch through our contact page.
References
- CIO Dive, “Microsoft touts Copilot growth, boosts spending as revenues soar”, 30 April 2026
- Microsoft Investor Relations, “FY26 Q3 Productivity and Business Processes Performance”
- Microsoft customer story on Nippon Steel’s staged Microsoft 365 Copilot rollout and the productivity results it delivered, Japanese-language page
- Ricoh’s column introducing Microsoft 365 Copilot use cases for improving office productivity, Japanese-language page
- Microsoft Thailand, “Microsoft 365 Copilot Pricing”
