Generative AI Consulting

Generative AI consulting for the writing, reading and searching your team does every day

Tools like ChatGPT, Claude and Gemini are good at drafting, summarising and finding information, and poor at knowing when they are wrong. This page covers where generative AI fits in an owner-managed business, where a person still has to check it, what staff should never paste into it, and how a free 3-minute review picks the first job to hand it.

Take the free AI Readiness Review See where it fits

What does generative AI consulting include?

Choosing which writing, reading and searching jobs to hand to generative AI, setting the checks and data rules, then building, testing and maintaining the workflow.

Generative AI means the models that produce text from an instruction. New Zealand's government business site describes the everyday version well: tools such as ChatGPT, Claude or Google Gemini act as assistants that help draft or summarise information. Some of your staff may already be using one of them. The consulting part is turning scattered personal use into a small number of jobs that are done the same way every time, with a person checking the parts that matter.

This page is deliberately narrower than our others. AI consulting services covers the whole catalogue, and AI implementation services explains how any workflow gets built, whether or not a language model is involved. Here the subject is one kind of AI: the models that read and write. Where the AI also takes actions across your systems, such as updating records or sending emails, read AI agents for business; for a chatbot that answers customers on your website, see AI chatbots for business.

Not to be confused with generative engine optimisation, which is about getting your business cited by AI search tools and is covered on our generative engine optimization agency page.

Where does generative AI actually help an owner-managed business?

In five everyday jobs: drafting quotes and proposals, summarising documents and calls, searching internal knowledge, replying to customers, and assembling routine reports.

Job	What the model does	Where it goes wrong	The check that stays with a person
Quotes and proposals	Drafts a quote or proposal from your price list, past jobs and site or meeting notes.	Invents a line item, uses an old price, or promises a date nobody agreed.	The person who prices the work reads every figure before it is sent.
Summarising documents and calls	Turns contracts, tenders, long email threads or call transcripts into a short summary with actions.	Drops the one clause or caveat that mattered, or attributes a comment to the wrong person.	Anything that drives a decision is checked against the original.
Internal knowledge search	Answers staff questions from your own procedures, manuals and records of past jobs.	Fills gaps from general knowledge when your documents are silent, or answers from an out-of-date file.	Answers name the document they came from, and one person keeps the documents current.
Customer replies	Drafts replies to enquiries and routine questions from information you have approved.	Sounds confident about stock, prices or policies it has never been given.	Routine replies follow approved wording; anything unusual goes to a person.
Reporting	Writes the commentary around weekly numbers pulled from your own systems.	Explains a trend that is really a data error, or does arithmetic it should not be trusted with.	Figures come from the systems, not the model, and a person signs off the narrative.

None of these jobs is new. They are the reading and writing that already fills an office week, and each maps to a workflow we build: AI-assisted quoting, document collection and processing, internal knowledge systems, AI customer service and weekly management reporting, all listed on the implementation page. Phone calls are a separate case, because the call has to be answered before anything can be summarised; that is what an AI receptionist is for.

Where is generative AI reliable, and where is it not?

It is reliable at drafting, rewording and summarising text you supply. It is unreliable at facts, figures and anything it has not been given.

The failure has a name. OpenAI describes hallucinations as plausible but false statements, says ChatGPT also produces them, and calls them a fundamental challenge for all large language models. The European Commission makes the same point for businesses in the EU: in its AI literacy questions and answers, it says staff using ChatGPT for tasks such as writing advertisement text or translating should be informed about specific risks, with hallucination given as the example.

Official small business guidance lands on the same answer: keep a person in charge. New Zealand's business.govt.nz recommends a "human in the loop" model, where a person operates the AI tool, checks its outputs and makes the final decisions. The US Small Business Administration advises that if you use free AI tools, another person should review everything they produce.

In practice that becomes four design rules for every generative AI workflow we build:

Give it the facts. The model drafts from your price list, your documents and your approved answers, not from its general knowledge.
Keep numbers out of its head. Totals, margins and dates come from your systems. The model writes the words around them.
Approve before it leaves the building. Anything a customer, supplier or regulator will read is approved by a named person.
Show the source. Summaries and answers point back to the document they came from, so checking takes seconds.

What should staff never paste into a public AI tool?

Passwords and bank details never. Customer and staff personal data or confidential commercial information only in a business tool your organisation controls and has checked.

Customer personal data. Names with addresses or phone numbers, full email threads, and anything about a customer's health, finances or complaints.
Staff records. Payroll, sick notes, disciplinary notes and performance reviews.
Access and payment details. Passwords, system logins, API keys, card numbers and bank details, in any tool, on any plan.
Confidential commercial material. Tender prices, margins, supplier terms and contracts covered by a confidentiality clause.

The reason is not that AI tools are uniquely dangerous; it is that data protection law still applies when the tool is new. Ireland's Data Protection Commission, in its guidance on AI, large language models and data protection, tells organisations using AI products to understand what personal data the tool uses, where that data goes and whether the provider keeps or reuses it, and to be able to handle access and deletion requests for anything staff put in. In the US, the SBA's guidance puts it more bluntly: try not to feed AI tools sensitive data or proprietary information.

Which account staff use matters as much as what they type. OpenAI states that by default it does not train its models on data from ChatGPT Business, ChatGPT Enterprise or its API platform. Personal accounts work differently: OpenAI's Data Controls FAQ explains how an individual user switches off the "Improve the model for everyone" setting so their conversations are not used for training. A business cannot see or enforce that setting on an employee's own account.

The practical fix is short: give staff business accounts for work, write the list above on one page, and make sure every built workflow only sees the data its job needs. None of this is legal advice; if your business handles health, financial or children's data, talk to your solicitor or data protection adviser as well.

Should you use off-the-shelf AI tools or a built workflow?

Use off-the-shelf tools for one person drafting or summarising. Build a workflow when the job repeats, crosses several systems, or needs approval steps.

Off-the-shelf tools

A business plan for ChatGPT, Claude or Gemini, or the AI features already inside software you pay for. As business.govt.nz points out, many businesses already have AI through Microsoft Copilot, Google Workspace or Canva.

Right for rewording an email, summarising a document someone has in front of them, or a first draft of a job advert. The limits show when the job repeats: each person prompts differently, information is copied in and out by hand, and there is no record of who checked what.

A built workflow

The same kind of model, connected to your inbox, job system or document store, given the same instructions and the same source documents every time, with its output landing where your team already works.

Right for the quote that goes out twenty times a week, the enquiry replies that should all sound like your business, or the Friday report. It costs more to set up, carries ongoing usage fees and needs looking after when the tools around it change.

Plenty of businesses should start with the first column and never need the second. If staff are not yet using AI tools with any confidence, our guide to AI training for small business covers the free courses worth taking before anyone spends money on a build.

Which AI is best for small business?

No single AI is best. The right one depends on the job and on the software you already use: Microsoft 365, Google Workspace or neither.

We do not rank them. The models are updated often, and a benchmark score says little about whether a tool drafts your quotes or summarises your tenders well. What each provider's own page says about its business offer is a fairer starting point:

ChatGPT. OpenAI describes ChatGPT Business as a secure workspace with shared company context, plugins and admin controls, says it connects to the tools a team already uses, and states that business data is never used to train its models.
Claude. Anthropic lists the Claude Team plan for teams of 2 to 150, with a Microsoft 365 connection, central billing and administration, single sign-on and no model training on your content by default.
Gemini. Google includes Gemini in its Google Workspace plans: in Gmail on Business Starter, and in Docs, Sheets, Drive and Meet from Business Standard upwards.
Microsoft Copilot. Microsoft says Microsoft 365 Copilot is built into Word, Excel, PowerPoint, Outlook and Teams, and sells it to businesses as an add-on to a Microsoft 365 plan or bundled with one.

For most owner-managed firms the practical route is simple. If the team already works in Outlook and Word, try Copilot first; if it works in Gmail and Google Docs, try Gemini; if neither, give ChatGPT and Claude the same real task for a week and keep the one your staff actually use. All four still need the checks set out above.

Choosing the tool is the smaller decision. Choosing which job to hand it matters more, and that is what the free AI Readiness Review does: six questions, about three minutes, and an instant report naming the two or three workflows worth starting with in your industry.

How does a generative AI consulting engagement work?

It begins with the free AI Readiness Review, then scoping one job, building it with checks, testing on real work, handover and maintenance.

Review. Six questions, about three minutes, and an instant report naming the two or three highest-impact AI workflows for your industry with estimated hours saved per week. Ger follows up personally within 24 hours.
Scoping one job. We pick the first job, gather a handful of real examples of it, and write down three things: what the model may read, what it must never see, and who approves its output.
Building it with checks. The workflow is built inside the tools you already use, fed with your own documents and figures, with the approval step designed in rather than added later.
Testing on real work. Drafts run alongside the current process, so the person who does the job today can compare the model's version with their own before anyone relies on it.
Handover. The people who use it see how it works, where it tends to go wrong and what to do when it does.
Maintenance. The tools a workflow depends on keep changing, and so do your prices and procedures. We keep the workflow working through both.

Typical timings for each stage, and what you need to provide, are set out in the stage table on AI implementation services. The difference with generative AI work is where the time goes: more of it is spent on the source documents and the approval step, and less on the model itself.

First-hand example

What my own business taught me about how AI models use information

I own Shamrock Electrical, an electrical wholesaler in Rathcoole, Dublin, trading for over a decade with a real Shopify store. Work is tried there before it is offered to anyone else.

The clearest result so far came from AI search rather than an internal workflow. We put structured data and question-and-answer content live on the Shamrock site, and ChatGPT named the business 72 hours later, with Perplexity and Gemini following. The lesson carries straight into generative AI inside a business: the models did well with information that was clear and well organised. A model given a clean price list and approved answers drafts well; a model left to fill the gaps will fill them.

Read the Shamrock Electrical case study

How much does generative AI consulting cost?

There is no single figure. The AI Readiness Review is free, and any workflow is scoped and priced after it.

Why there is no single figure

A summarising assistant for one shared inbox is a different job from a quoting workflow that reads a price list, job history and an accounts package. Price depends on which jobs the AI handles, which systems it connects to, how tidy the source documents are and how many approval steps are needed. Model usage fees paid to the AI provider sit on top.

What you get free first

The review, the instant report naming your two or three highest-impact workflows with estimated hours saved per week, and a follow-up from Ger within 24 hours. If an off-the-shelf tool would do the job, you will hear that before any money changes hands.

Who it is for

Owner-managed businesses with roughly 5 to 50 staff and a lot of repeated reading and writing: trades and construction, accounting, legal and recruitment firms, ecommerce, distribution and agencies. Not tech companies, and not anyone looking to train their own model from scratch.

What is free, what is fixed and how builds are scoped is set out on the pricing page.

Where we work

Clever Merchants is based in Dublin and works remotely, with the review online, calls by video and workflows built inside your own tools, for businesses in Ireland, the UK, Australia, New Zealand, Canada and the United States.

Questions people ask about generative AI consulting

What is generative AI consulting?

It is advice and hands-on work on using models such as ChatGPT, Claude and Gemini for real business tasks like drafting, summarising and searching. Good consulting also sets the checks and data rules that keep those tasks safe.

Can generative AI write our quotes without anyone checking them?

No, and it should not be set up that way. The model can draft a quote from your price list and notes in seconds, but the person who prices the work checks every figure before it goes out.

Is it safe for staff to paste customer information into ChatGPT?

Not into a personal or free account. Customer personal data belongs only in a business tool your organisation controls, with its data settings checked, and passwords or bank details should never go into any AI tool.

Do we need a custom-built AI system, or will ChatGPT or Copilot do?

For one person drafting or summarising, an off-the-shelf business plan is usually enough. A built workflow earns its cost when the job repeats every day, crosses several systems or needs an approval step.

Which AI is best for business consulting?

No single tool is best: ChatGPT, Claude, Gemini and Microsoft Copilot all draft and summarise, and the right one is usually the one that works inside the documents and email you already use. Whichever you choose, a person should check every figure and claim before it reaches a client.

How much does generative AI consulting cost?

There is no single figure, because the price depends on which jobs the AI handles and which systems it connects to. The AI Readiness Review is free, and any build is scoped and priced after it.

Find the first job to hand to generative AI

Six questions, about three minutes, an instant report naming the workflows worth building in your industry, and a personal follow-up from Ger within 24 hours. Free, no card, no obligation.

Take the free AI Readiness Review