In a 1:1 AI implementation consultation, you walk through how your business actually runs, the consultant maps the processes that take the most time, checks what your current tools can already do, and picks the few jobs where AI pays off. You should leave with a written plan your team can run, not a list of apps to try.
It isn't a demo, a sales pitch, a lesson in how AI works or a build session; the point is to find out what AI you can actually use in your business. Nothing gets automated on the call itself; the value is in the decisions: which job first, on which tool, owned by whom, and judged by what number. What you bring shapes what you get, so the preparation matters almost as much as the conversation, and the sections below follow the order a good consultation tends to run in.
Before the call: what to send, and why each item matters
A consultant who arrives knowing the basics can spend the call on your problems instead of on background. Send a short pack a few days ahead. Here's what an illustrative wine merchant with two shops, an online store and a trade arm (18 staff) might send:
- Your main processes, with rough weekly hours. Trade orders by email (office, about 7 hours a week); online orders (packing team, about 10); shelf notes for new wines (shop staff, about 3); supplier price-list updates (owner, about 2); customer emails about delivery and stock (office, about 4). Rough is fine; the consultant will test the numbers.
- The tools you use, with the plan names. Google Workspace Business Standard, an online shop platform, Xero on the Growing plan, a shared stock spreadsheet, a till system in each shop.
- Three pain points in numbers. "Four wrong trade deliveries last month, all vintage mix-ups." "New wines wait up to a week for shelf notes." "Price-list updates eat Sunday evenings."
- Two or three real examples, anonymised. A typical trade order email (some arrive as photos of handwritten lists), a supplier price list, a shelf note you're happy with.
- What you've tried. "Staff use ChatGPT on personal accounts for shelf notes; results vary and nobody checks the facts."
- Constraints. Staff time available, anything you won't change, and data you're wary about sharing.
It's worth seeing what happens without the pack. Imagine a butcher who books a call and arrives with no examples and no numbers, just "we'd like to use AI for our emails". The conversation stays general because it has to: which emails, how many, what goes wrong? The plan at the end says "use an AI assistant to draft email replies", which the owner could have written without the call. Three weeks later nothing has changed. The same butcher, sending five real customer emails and a count of how many arrive each week, would have learned on the call that most of them ask about click-and-collect times, and that the fix starts with the order confirmation email.
Each item in the pack saves time on the call and sharpens the advice. The examples matter most: a consultant who has seen your actual trade order emails will spot the photo-of-a-list problem in seconds, while one working from a description may miss it. Preparing for an AI consultation has a fuller checklist, and what an AI consultant needs from you covers access and data if the work goes further.
Mapping how the work really moves
The first part of the call is usually a walk through one or two processes, step by step, as they happen today rather than as they're meant to. A consultant is listening for six things: what starts the job, each step, which system each step happens in, who does it, how often, and where it goes wrong.
Here's the wine merchant's trade order process, mapped the way it might come out of the conversation:
| Step | Where | Who | Minutes per order | What goes wrong |
|---|---|---|---|---|
| Order arrives: free-text email, sometimes a photo of a handwritten list | Shared Gmail inbox | Trade customer | n/a | Product names abbreviated or misspelt; vintage often missing |
| Read the order and check stock | Stock spreadsheet | Office | 4 | Stock sheet sometimes a day behind the tills |
| Enter as a draft invoice | Xero | Office | 5 | Wrong vintage chosen when two are listed |
| Confirm to the customer with the delivery day | Gmail | Office | 2 | Forgotten on busy days |
| Add to the delivery run sheet | Google Sheets | Office | 2 | Occasionally missed, found at loading |
At about 35 trade orders a week and 13 minutes each, that's 455 minutes, or roughly 7.6 hours a week. The map also shows something the owner hadn't said out loud: the four wrong deliveries last month weren't typing errors, they were vintage mix-ups at the "read the order" step, made worse by a stock sheet that lags the tills. Mapping often moves the problem. An AI step that reads orders could still pick the wrong vintage unless the product list tells it which vintage is current, so part of the fix is a tidier stock list, not a cleverer model.
Finding where AI pays off, and where it doesn't
With the main jobs mapped, the consultant scores them. The criteria are plain: how often the job happens, how long each one takes, whether it follows rules or needs judgement, what a mistake costs, whether the information AI would need exists in usable form, and who would own it afterwards. An illustrative scoring for the wine merchant, each criterion from 1 (poor fit) to 3 (good fit):
| Job | Volume | Time each | Rule-based | Error cost (3 = low) | Data ready | Owner | Total |
|---|---|---|---|---|---|---|---|
| Trade order entry | 3 | 3 | 2 | 2 | 2 | 3 | 15 |
| Shelf notes from producer sheets | 2 | 3 | 3 | 2 | 3 | 3 | 16 |
| Customer emails about delivery and stock | 3 | 2 | 2 | 2 | 2 | 2 | 13 |
| Supplier price-list updates | 1 | 3 | 3 | 1 | 2 | 3 | 13 |
| Social media posts | 2 | 2 | 1 | 3 | 2 | 1 | 11 |
Shelf notes and trade orders come out on top for different reasons. Shelf notes are quick to fix: the staff already use AI for them, they just need a business account, a standard prompt and a rule that every fact is checked against the producer's sheet. Trade orders save the most time but need the stock list tidied first. Price-list updates score well on rules but badly on error cost, since a wrong price on the website is expensive, so they wait. Social posts score lowest because nobody owns them.
A good consultant also names jobs that AI shouldn't do at all. For a wine merchant, any decision about whether a sale can go ahead, such as checking a customer's age, stays with a trained person. Saying so plainly is part of the job.
Sometimes the scoring shows that the best fix involves no AI at all. Picture an illustrative farm shop whose biggest time sink is veg box changes: customers email to skip a week, swap a box size or change an address, about 300 emails a month, each copied into a spreadsheet by hand. An AI step could read those emails. But the mapping shows that nine in ten changes are one of three standard requests. A simple change form on the website, feeding the spreadsheet directly, removes most of the emails altogether, and the AI question only applies to the few free-text messages left over. A consultation that recommends the form over the AI is working properly, even though it sells nothing clever.
Checking what your current tools can already do
Before recommending anything new, the consultant goes through what you run today. Many business tools now include AI features that owners haven't switched on. Some examples worth checking, as of September 2026:
- Google Workspace Business Standard and above include Gemini across Gmail, Docs, Sheets and Drive, and Gmail's "Help me write" can match the tone of your past emails.
- Microsoft 365 business plans include Copilot Chat at no extra cost, which Microsoft now calls Microsoft Copilot Chat and which works on Outlook mail and open files.
- Xero has announced Smart Document Capture, built-in capture of bills and receipts, free on all plans.
- HubSpot's free and Starter tiers include the Breeze assistant for writing and summaries.
- Shopify Inbox includes a free AI agent that answers customers on the Basic plan or above, with new customer accounts switched on.
For the wine merchant, that check changes the plan. Gemini in Gmail, already included in their Workspace plan, can draft the trade order confirmations, so the first build needs an automation platform but no new AI subscription. The consultant might test it live on the call with a real order. An illustrative exchange:
Prompt: "Draft a confirmation to this trade customer for the order below. Our delivery day for their area is Thursday. List each wine with vintage and quantity."
Illustrative draft: "Hi [first name], thanks for your order. We'll deliver on Wednesday [wrong: the prompt said Thursday; the model picked up 'Wednesday' from the customer's own email, which asked for Wednesday if possible]: 12 x house red 2023, 6 x old-vine Garnacha [vintage missing: two are in stock]..."
Two useful lessons in one test: the AI will follow the customer's wishes over your rules unless told otherwise, and it can't choose a vintage the order didn't specify. Both become rules in the plan: delivery days come from the delivery sheet only, and any wine with more than one vintage in stock is flagged for a person. AI features already in your software lists more of these built-in features by tool.
The plan you should leave with
The call should end with something written down, ideally on one page. For the wine merchant it might read:
| Item | Decision |
|---|---|
| First job | Shelf notes: business AI account, standard prompt, facts checked against producer sheets |
| Second job | Trade order entry, once the stock list shows current vintages |
| Why these | Shelf notes are quick to fix and already half-done; trade orders save about 7.6 hours a week and cause the delivery errors |
| Tools | Gemini in Gmail and Sheets (already on the plan) plus an automation platform; no new AI subscription |
| Owner | Office manager for trade orders; shop manager for shelf notes |
| Before building | Tidy the stock list; collect 30 past trade orders as test cases |
| How we'll judge it | Minutes per trade order; wrong deliveries per month; days from arrival to shelf note |
| Check-in | Day 30 |
| Not now | Website chat assistant; automated social posts; price-list updates |
| Who builds | Decide after the stock list is tidy: in-house with guidance, or a fixed-price project |
The "not now" line is as valuable as the rest. It stops the next enthusiastic idea from pulling attention before the first job is working.
Questions worth asking during the call
- What would you not automate here, and why?
- What can we do ourselves this month without buying anything?
- What would this cost to run each month at our volumes?
- What needs to change in our process or data before AI can help?
- What's the most likely way this fails in the first three months?
- How will we know at day 30 whether it's working?
For a sense of what a useful answer sounds like, here's how the fifth question might be answered for the wine merchant's trade orders: "Three likely failures. Vintage mix-ups if the stock list falls behind the tills again, which is why tidying it comes before any build. Orders sent as photos of handwritten lists, which the AI step will sometimes misread, so those always go to a person. And the whole flow stopping if someone renames a column in the order sheet, so the column names get locked and a failure alert goes to the office manager." Specific, tied to the map, and each failure already has a counter-measure in the plan.
If the answers keep coming back to your own numbers and your own tools, the consultation is doing its job.
Three things a consultation won't do
- It won't build anything. The call produces decisions and a plan. Building happens afterwards, by your team, a freelancer or the consultant, as a separate piece of work.
- It won't guarantee a result. A consultant can estimate hours saved from your own figures, but anyone promising a specific saving before seeing your work is guessing.
- It won't make the decisions for you. You know your customers, your staff and your appetite for risk. The consultant's job is to lay out the options clearly enough that choosing is easy.
How to tell it was a good consultation
Judge it against the same standards you'd apply to anyone: did the consultant use your numbers rather than generic ones, name your actual tools and plans, tell you at least one thing not to do, leave you something written, avoid pressing you to buy more, and give you a first step you could take without them? If most of those are true, the call has earned its place. What an AI implementation consultant does describes the wider engagement if you go on to build with the same person, and free AI consultations explains how a free introductory call differs from a paid working session.
If you'd rather map this with someone, that's what my AI implementation consultation covers: your processes, the jobs where AI pays off, what your current tools already do, and a plan your team can run. Apply the same test to it as to any other.
The first fortnight after the call
Plans go cold fast. Within two days, share the one-page plan with whoever owns the first job and book the day-30 check-in. In the first week, do the "before building" items: for the wine merchant, tidying the stock list and collecting 30 past orders. By the end of the second week, the first job should be running or have a date to start. Turning consultation advice into a 30-day plan takes it from there, week by week.
Questions people ask before booking a consultation
Do I need to understand AI before the call?
No. You need to understand your business, which you do. A good consultant explains any term they use and works from how your jobs are done today, not from AI jargon. If you've never used a business AI assistant, twenty minutes trying one on a real email beforehand will help you picture the options, but it isn't required.
Should my staff join the call?
Often one person should: whoever does the job you most want to change. They know where the time really goes and where mistakes creep in, which owners sometimes don't. Keep it to one or two people so the conversation stays on decisions. If staff can't join, ask them to note their three biggest time sinks beforehand.
What if the answer is that AI isn't worth it for us yet?
That's a legitimate outcome and worth knowing before you spend on tools. The plan might be to tidy a spreadsheet, write down a process, or fix a web page first. Ask what would change the answer, such as a higher volume of orders or cleaner data, so you know when to revisit the question.
Will the consultant need logins to my systems?
Not for a consultation. Screen-sharing, screenshots or anonymised examples are enough to understand how the work flows. Never read out or send passwords on a call. If you later agree a build, access should come through user accounts you create and can remove, with two-factor authentication left on.
Further reads
- Is a 1:1 AI Consultation Worth It for a Small Business? — The break-even sum for a single paid call.
- How to Audit Your Workflows for AI Opportunities Yourself — Map your own workflows before or instead of a call.
- How to Choose Your First AI Project: 7 Tests Before You Commit — Seven tests for the first job you pick.
- Course, 1:1 Call, Audit, or Project: Which AI Help Do You Need? — Check a call is the right kind of help for you.
- How to Choose an AI Consultant: 20 Questions to Ask First — Twenty questions to put to anyone offering a consultation.
- AI Implementation Roadmap for Small Businesses: 5 Phases in 90 Days — Where the first job fits in a longer plan.
- What a Typical AI Consulting Engagement Looks Like, Week by Week — A six-week AI consulting engagement explained stage by stage: what the consultant does, what you do, what you should have each Friday, and why weeks slip.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Microsoft Learn (Copilot Chat naming and Microsoft 365 plans), Google Workspace plan pages (Gemini features, Gmail Help me write), Xero announcement of Smart Document Capture, HubSpot Breeze tiers, Shopify Inbox help pages (checked September 2026). Business examples are illustrative.