Home / Advisors guide

Advisor guide

What to tell your clients about AI in 2026.

A practical guide for accountants, bookkeepers, MSPs, web partners, and business advisors whose clients are asking, "Should we be using AI?"

Talk through a client situation

Who this is for

Your client has heard enough AI hype to feel behind, but not enough operational guidance to act safely. They may already have staff using ChatGPT, Microsoft Copilot, Gemini, Claude, or niche AI tools. They may have no policy, no approved-tool list, no review cadence, and no idea what customer data is being pasted where.

This guide gives you a plain-English way to talk about AI without pretending to be an AI vendor. It is for the client who asks:

  • Should we be using AI?
  • What are other businesses doing?
  • Is it safe to put client data into these tools?
  • Could AI help our admin, inbox, reporting, quoting, or onboarding work?
  • Who should own this internally?

The answer is not "buy a chatbot". The answer is: get control first, map one workflow, then test one bounded assistant if it earns its place.

Page 1

The five things every client should have before serious AI use.

If these are missing, the client does not have an AI strategy. They have staff improvising with software.

1. Data rules

What can and cannot go into AI tools. Client data, health data, employee data, financial records, credentials, contracts, and private keys need explicit rules.

2. Approved tools

A short list of tools the business accepts, including whether prompts and outputs are retained, used for training, or covered by paid business terms.

3. Named owners

Every AI-assisted workflow needs one accountable human. Not "the team". A person or role that owns review, exceptions, and the off-switch.

4. Human review

Customer-facing, money-moving, legal, HR, clinical, and compliance-touching outputs should be reviewed before anyone acts on them.

5. Kill criteria

Written triggers to pause use: errors, complaints, tone drift, vendor changes, unexplained outputs, or a workflow owner losing confidence.

Advisor line

"AI can be useful here, but only if we know what it can see, who checks it, and how to turn it off." That sentence saves a lot of nonsense.

Page 2

The risk is not usually the model. It is the unmanaged workflow around it.

A client rarely gets into trouble because someone used AI once. The risk grows when AI becomes quietly embedded in everyday work without rules.

  • Staff paste client information into free tools because it is faster.
  • AI drafts customer emails that no named person reviews.
  • Reports are summarised from messy spreadsheets without checking source data.
  • Multiple paid tools appear on cards and expense claims, all with different data terms.
  • The CEO asks "what is our AI policy?" and everyone looks at the Ops Manager.
What to say

A practical client script.

"Start with an AI hygiene pass. First we list the tools, write data rules, name owners, and choose one workflow worth improving. Then we test something small."

That framing keeps the client calm and gives them a useful next step.

Page 3

Where AI can help first.

Start with work that is repetitive, bounded, reviewable, and painful enough that someone already complains about it.

Good first workflows

  • Inbox triage and routing.
  • Quote follow-up drafts.
  • New client onboarding checklists.
  • Recurring report summaries.
  • Support request classification.
  • Meeting-note extraction into tasks.
  • Policy or knowledge-base retrieval.

Poor first workflows

  • Unsupervised customer chat on launch day.
  • Pricing decisions.
  • Legal or clinical recommendations.
  • HR decisions.
  • Anything nobody owns.
  • Anything with unclear data boundaries.
  • Anything the business cannot disable.
Page 4

The first 30 days: what a sensible client should do.

  1. List current AI tools. Include free accounts, paid accounts, browser extensions, meeting tools, CRM features, and anything reimbursed on expenses.
  2. Write the data rule. One page. What data can go in, what cannot, and which tools are approved.
  3. Run the governance self-check. Score data handling, vendor retention, human review, named ownership, review cadence, drift, and kill criteria.
  4. Pick one workflow. Not ten. One. The workflow people already complain about.
  5. Map it before automating it. Trigger, owner, systems, steps, decisions, handoffs, output, and failure points.
  6. Decide the first safe assist point. Draft, classify, summarise, route, retrieve, or alert. Do not start with autonomous action.
  7. Test the off-switch. If the team cannot operate when the assistant is off, the assistant is not ready.

None of this requires a giant programme. It requires operational hygiene and someone willing to write things down. Revolutionary stuff, apparently.

Page 5

When to refer a client.

A client is a good fit for Automation Nation if they have enough operational complexity that AI governance matters, but not enough internal capacity to build a safe programme alone.

  • 20 to 150 staff, especially 30 to 80.
  • Service-based, knowledge-work-heavy, or regulated-adjacent.
  • Uses Microsoft 365 or Google Workspace plus one line-of-business platform.
  • Has an Ops Manager, Practice Manager, GM, COO, Business Manager, or Service Delivery Manager.
  • Already has staff experimenting with AI but no clear policy or ownership.
  • Has one painful workflow they have discussed automating for months.
Offer path

The order we recommend.

  1. Governance Starter Pack: rules, tool list, owners, review cadence, and kill criteria.
  2. Process Digitiser: map one messy workflow and identify the first safe assist point.
  3. Assistant Pilot: build one bounded workflow assistant with human review and an off-switch.
  4. Managed AI Operations: maintain, review, document, and improve the system over time.
Page 6

How to use this guide with a client.

Send it when a client asks about AI and you do not want to give them either hype or fear. The useful move is to make the next conversation specific.

Three good client questions

  1. Where is AI already being used by your team, officially or unofficially?
  2. Which workflow creates the most admin, rework, waiting, or customer chasing?
  3. If AI helped with one thing safely this quarter, who would own it and who would review it?

Useful next links

Have a client asking about AI?

Send the messy version. We can help you decide whether they need governance, a process map, a bounded pilot, or nothing yet.

Email Automation Nation