You’ve Given Your Employees ChatGPT. But Have You Actually Changed How Your Business Works?

Imagine you've introduced ChatGPT across your business.

You've purchased licences for your employees. You've organised some training. Staff have been shown how to write effective prompts, analyse information, prepare documents and use AI to help with everyday tasks.

Some employees have embraced it enthusiastically. Others use it occasionally. A few have become particularly proficient.

From a management perspective, it feels like progress. The business is adopting AI, staff are developing new skills and the investment appears to be moving in the right direction.

But there's a question I think business owners need to ask.

What has actually changed in the way your business works?

Are client proposals being prepared differently? Has the process for producing management reports improved? Are recurring administrative tasks being handled more consistently? Have employees agreed better ways of working, or are they simply using ChatGPT to carry out parts of their existing jobs?

There's an important difference.

Giving employees access to AI tools can be valuable. Helping them become proficient with those tools is also worthwhile.

But neither necessarily means you've implemented AI into your business processes.

In my view, understanding that distinction is one of the most important steps Irish businesses can take as they move from experimenting with AI towards putting it to practical use.

Using AI is not the same as implementing an AI workflow

It's useful to distinguish three different stages of business AI adoption.

The first is access to the technology.

An organisation provides employees with ChatGPT, Microsoft Copilot or another AI tool. People have somewhere to experiment, ask questions and begin developing their skills.

The second is individual productivity.

Employees start applying AI to their daily work. Someone uses ChatGPT to draft a customer email. Another uses it to summarise a document. A manager uses it to prepare a report or examine an unfamiliar topic.

These activities can be genuinely useful. They may help individual employees work more effectively.

The third stage is organisational implementation.

This is where a business examines a particular process and deliberately establishes how AI should contribute to completing the work.

The organisation agrees which information should be used, what the output needs to contain, which steps AI can assist with and where human judgement and review remain necessary.

The method is tested, documented and introduced to the relevant employees.

That's a rather different proposition from simply giving everyone a ChatGPT licence.

Importantly, an AI workflow doesn't have to be fully automated.

A process that uses AI to prepare a first draft, followed by structured human review and approval, can be a perfectly sensible implementation.

The objective isn't to remove people from the process. It's to identify where AI can make a useful contribution and establish a reliable way of working around it.

Why businesses can get stuck at the experimentation stage

Irish research provides some useful context.

According to the Central Statistics Office's Information Society Statistics for 2025, 20.2% of enterprises surveyed reported using AI technologies.

However, only 6.2% reported using AI specifically for automating workflows or assisting decision-making.

These are different measures of technology usage, and they don't establish whether individual organisations have well-designed working processes. Nevertheless, they help illustrate that using AI and applying it to particular business functions are not identical activities.

The Small Firms Association's 2026 AI Adoption Survey also highlighted widespread use of generic AI tools alongside lower reported adoption of customised AI technology.

The SFA survey covered 404 small-business decision-makers. Its findings should not be directly compared with the CSO figures because the populations and questions differ.

And there's another important distinction: a company doesn't need to develop customised AI software to create a worthwhile AI-assisted workflow.

The broader challenge is knowing where and how to apply the technology.

Research published by the Economic and Social Research Institute in April 2026 found that knowledge and awareness barriers help explain differences in AI adoption among Irish SMEs. Finance, security concerns and skills also influence intentions to invest.

For a busy business owner, these findings aren't particularly surprising.

It's one thing to recognise that AI is useful. It's another to find the time, expertise and internal resources to examine how a recurring business process could be improved.

Often, the difficulty isn't choosing another tool. It's deciding precisely what needs to change.

Let's look at a practical example: preparing a client proposal

Consider a fictional professional services company with 30 employees.

The company prepares proposals for prospective customers on a regular basis.

The existing process is reasonably familiar.

An employee receives a customer enquiry and gathers the requirements. They look for relevant service descriptions, find previous proposals, adapt the wording and prepare a draft.

They then send it to a manager for review.

The manager checks that the proposal reflects the customer's requirements, that the commercial details are correct and that the company hasn't made commitments it cannot fulfil.

The document might go through several revisions before being approved.

Now imagine the company gives its employees ChatGPT.

An employee discovers that they can paste a brief into the tool and ask it to generate a proposal.

They may produce a first draft more quickly.

That's useful, potentially.

But look at what hasn't changed.

The employee still needs to locate the company's information. They still have to decide which previous proposals are relevant. They still need to explain the document structure and check whether the AI has introduced inaccurate or unsupported claims.

The manager may still need to correct the same problems that existed before.

The employee is now using AI, but the business hasn't necessarily improved the proposal preparation process.

What would happen if we approached this as a workflow implementation instead?

We could start by examining the entire process, not just the writing stage.

The company could agree a standard method for gathering customer requirements, identify approved service information, establish a proposal template and define which parts of the drafting process AI is allowed to assist with.

The AI could be instructed to prepare a structured first draft, using only the approved information provided, and to flag missing customer requirements rather than inventing details.

A named employee would then review the draft, while the authorised manager would retain responsibility for the final commercial approval.

The method could be tested using representative examples before being introduced into routine work.

The difference becomes clearer when we compare the two approaches.

This is an illustrative example, not a claim about a completed customer implementation or measured savings.

Nor does it mean the proposed process would automatically perform better.

We would still need to test it.

But it shows why I believe businesses should examine the work before deciding what role AI should play.

The opportunity may be much broader than generating text more quickly. It may involve improving how information is gathered, how drafts are structured, how exceptions are handled and how employees review completed work.

Six decisions that turn an AI experiment into a working process

For a business considering its first serious AI workflow, I would begin with six practical decisions.

I think of this as the One-Workflow Implementation Test.

1. What task are we actually trying to improve?

Start with the business process, not the technology.

Choose a recurring task that employees understand and that has a clearly identifiable owner.

Map how the task works today, including the time-consuming steps, sources of error, hand-offs and approval requirements.

If the problem is unclear, it's difficult to know whether AI will improve anything.

2. What information does the task require?

Identify the documents, data, instructions and business knowledge needed to complete the work.

Are those sources accurate? Are they current? Is the organisation authorised to use them with the proposed AI system?

Poor information can undermine an otherwise well-designed workflow.

In some cases, preparing the company information will be more important than developing a sophisticated prompt.

3. What does an acceptable output look like?

Be specific about what the finished work should contain.

For a client proposal, that might mean an agreed structure, accurate service descriptions, properly identified customer requirements and clear treatment of missing information.

If nobody has defined what good looks like, employees will struggle to assess whether an AI-generated result is suitable.

4. Which parts should AI perform?

Not every step needs AI.

Some tasks may be better handled by existing business software, document templates or conventional automation.

Other activities require human judgement, interpretation or approval.

The objective is to identify a sensible division of responsibilities, rather than automatically attempting to automate the whole process.

5. Who checks the work and owns the process?

An AI-assisted workflow still needs accountable people.

Who reviews the output? Who resolves errors? Who approves the final document? Who updates the instructions when business information changes?

In a smaller company, several responsibilities may sit with one person.

What matters is that they are understood.

6. How will we know whether the new method is worthwhile?

Agree how the process will be tested before committing to wider use.

Compare representative examples of the current approach with the proposed AI-assisted method.

Look at quality, effort, corrections, employee experience and any new risks or support requirements.

If the results aren't sufficiently useful, improve the method or decide not to proceed.

This is an important point: a successful AI assessment doesn't always result in an AI implementation.

Sometimes the best outcome is discovering that a simpler process change would be more appropriate.

Training people is essential, but it needs to connect to the work

I place considerable value on AI training.

Employees need to understand the capabilities and limitations of the tools they're using. They need opportunities to experiment, ask questions and develop confidence.

Good training can help people identify opportunities that management may not have considered.

However, there's a difference between teaching employees how to use ChatGPT and teaching a team how to carry out an agreed business process with AI assistance.

For example, a workshop might show employees how to prepare a proposal draft.

That's useful learning.

But if the business subsequently establishes an approved AI-assisted proposal workflow, employees also need to understand the specific company information they should use, the required document structure, the limits of the AI's role and their responsibilities for reviewing the output.

They need to practise the method with realistic examples.

And they need to know what to do when something goes wrong.

This is why I see training and implementation as complementary activities.

Training develops employee capability. Workflow implementation establishes how that capability is applied within a particular business process.

Both can be valuable independently, and the appropriate sequence depends on the organisation's needs.

How do you know whether the workflow is actually better?

One of the easiest mistakes in an AI adoption programme is to measure activity rather than business improvement.

How many employees have activated their licences?

How many people attended training?

How frequently are they using ChatGPT?

These can be useful indicators of access and engagement, but they don't tell management whether a particular process has improved.

Return to the proposal example.

If we want to assess the proposed AI-assisted workflow, we might examine the total time needed to produce an approved proposal, the number and significance of corrections, the quality of the finished document and the effort required from reviewers.

We should also ask whether employees can follow the method consistently and whether it introduces additional work elsewhere.

Perhaps the first draft is prepared more quickly, but the manager now spends longer correcting inaccuracies.

Perhaps the workflow produces consistent structures, but employees find it difficult to handle unusual customer requirements.

Perhaps the process works extremely well for standard proposals and poorly for highly customised engagements.

All of these are useful findings.

The business needs a reasonable baseline, comparable work examples and clear acceptance criteria before concluding that the workflow has delivered an improvement.

I would be cautious about claims of dramatic time savings unless they have been properly measured.

An AI-assisted process should ultimately be judged on the quality and suitability of the completed work, not how impressive the initial output looks.

Not every business process needs AI

There's another point worth making.

Not every inefficiency should be solved with artificial intelligence.

Sometimes employees are struggling because they have inconsistent document templates, poorly organised information or unnecessarily complicated approval processes.

A better template, clearer responsibility or a straightforward change to existing software may solve the problem more effectively.

There are also activities where the information is too sensitive, the consequences of error are too serious or the proposed AI method cannot be reviewed adequately without disproportionate effort.

In those situations, it may be better to defer the implementation, narrow its scope or choose another approach.

I don't think businesses should measure progress by the number of AI workflows they introduce.

The objective is to make work better.

AI should be used where it contributes meaningfully to that objective.

Start with one worthwhile business task

For an Irish SME beginning to move beyond individual AI experimentation, my suggestion would be relatively straightforward.

Don't start by asking how to automate the entire business.

Identify one recurring process that employees understand and that management believes could be improved.

It might involve preparing proposals, generating recurring reports, researching information or producing internal documentation.

Speak to the employees who actually perform the task.

Understand the current process, identify its difficulties and agree what a better result would look like.

Then assess whether AI can make a useful contribution.

Define the inputs, the intended output and the human review responsibilities. Test the proposed approach with realistic examples.

If the method is genuinely worthwhile, document it, train the people involved and decide how it will be maintained.

If it isn't, you've still learned something useful about the process.

The important thing is to approach AI adoption as a business improvement exercise rather than simply a technology rollout.

Giving your employees ChatGPT is a useful starting point. The real opportunity comes when you can show what has changed in how the work gets done.

Have a business process that could benefit from AI?

AI Strategy & Training Ltd helps Irish organisations identify suitable AI opportunities, examine existing workflows, develop and test practical AI-assisted working methods, and equip employees to use them responsibly.

Whether your business needs an initial assessment, practical staff training or a defined implementation project, the starting point should be the work you want to improve.

Discuss an AI workflow opportunity with AIST.