# Where Should a Small Manufacturer Start With AI First?

Canonical URL: https://business.theaieducator.io/posts/ai-for-small-manufacturers
Publication: Dan Fitzpatrick Insights
Author: Dan Fitzpatrick
Topic: AI in Business Operations
Published: 2026-11-05T16:42:37.000Z
Modified: 2026-11-05T16:42:39.882Z

Small manufacturers lose most of their AI-ready hours between people, not on the shop floor. Use the Retype Rule and a one-page Job Trail Sheet to pick your first AI job and keep safety and quality sign-offs with people.

## In brief

A small manufacturer should start AI where people retype or re-explain the same information most often, usually the quote, work order and customer update, not the shop floor. Map one job on a Job Trail Sheet, start at the busiest handoff, and keep safety and quality sign-offs with a named person.

## Key takeaways

- The Retype Rule says to start AI where your people retype or re-explain the same information most often, and nowhere else first.
- For most small manufacturers that means the quote, the work order, the delivery note and the customer status reply, not the machines.
- The Job Trail Sheet follows one real job from inquiry to invoice and ranks each step by times a week multiplied by minutes each time.
- Any step with a safety, quality or legal sign-off stays with a named person until a human check has passed the Loop Test for a full quarter.
- UK government data from 29 October 2026 puts AI use at 32% of businesses, and the sources I read publish no manufacturing figure in their text, so be wary of anyone who quotes one.
- Start with one job type, one person and four weeks, then decide on a fixed date what to widen.
- Wait if nobody owns the job or the paperwork lives only in people's heads.

A small manufacturer should start AI where people retype or re-explain the same information most often, which is usually the quote, the work order and the customer update, not the shop floor. Pick one job, mark every place it changes hands, begin at the busiest handoff, and keep every safety and quality sign-off with a named person.

That answer comes from a pattern, not a product. In a business of 10 to 100 people that makes things, most of the lost hours sit between people: an inquiry becomes a quote, a quote becomes a work order, a work order becomes a delivery note, and each time someone keys the same details in again. This article gives you a rule for finding those handoffs and a one-page sheet to find yours. It is written for owners and operations managers of small and medium-sized manufacturers (SMEs in the UK and Europe), with no data team and no appetite for a big project.

## What is the best place for a small manufacturer to start with AI?

The best place to start is the handoff where your people retype or re-explain the same information most often. I call this the Retype Rule: it is my suggested way of thinking, not established research.

**The Retype Rule is one rule for choosing where AI starts in a business that makes things: begin where your people retype or re-explain the same information most often, and nowhere else first. Every retype is a handoff, every handoff is a chance for an error, and a handoff is the one place AI can help without touching the work itself.**

Two things follow. First, you do not need to understand machine learning, sensors or your factory data to get a first win; you need to know your own paperwork. Second, the rule keeps AI away from the places where a mistake hurts most. Cutting, welding, mixing, inspecting and releasing product stay with people. The quote, the confirmation email, the delivery note and the "where is my order" reply are where words, numbers and documents change hands, and that is work AI can draft while a person checks. A NIST Manufacturing Extension Partnership author put the attitude well in 2019, calling AI "just another tool in the lean toolbox"; the Retype Rule is one way of deciding which job in the box to pick up first.

My article on [AI for trades and construction](https://business.theaieducator.io/posts/ai-for-trades-and-construction) draws a line between work on the site and work at the desk. This one asks a different question: inside a business that makes a product, which single handoff is worth starting with?

## How common is AI among small manufacturers?

Use among small firms is rising fast but is still a minority, and the main official sources do not publish a manufacturing figure in the text I read, so be wary of anyone who quotes you one.

What the sources do say:

- In the UK, the Department for Business, Innovation, Science and Trade reported on 29 October 2026 that [one-third of UK businesses were using at least one AI technology in 2026 (32%, against 16% in 2025)](https://www.gov.uk/government/publications/ai-adoption-survey-2026/ai-adoption-survey-report-2026-executive-summary). The survey was a representative sample of UK businesses: 3,503 telephone interviews, fieldwork 7 January to 21 April 2026. It also found that a lack of a clear business case or proven return on investment was cited by many businesses, particularly non-adopters (65%), and that 81% of adopters apply human oversight to AI outputs.
- The Office for National Statistics, using the Business Insights and Conditions Survey with fieldwork in June 2026, reported on 20 July 2026 that [self-reported AI use in UK businesses with 10 or more employees rose from around 12% to around 35% since late 2023](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026). The same release shows 28% of firms with 0 to 9 employees against 49% of firms with 250 or more.
- In the US, a Census Bureau working paper using the 2026 AI supplement to the Business Trends and Outlook Survey (reference period November 2025 to January 2026) found that [18% of firms used AI in a business function, and 57% of adopting firms used it in three or fewer functions](https://www.census.gov/library/working-papers/2026/adrm/CES-WP-26-25.html). Its authors report that use is substantially higher in large firms and knowledge-intensive sectors, and that AI-related employment decreases occurred in 2% of firms.

What this means for you is my reading, not the sources'. Most firms that use AI use it narrowly, which is the right way to begin. Sectors that deal mostly in documents are ahead of sectors that make things, and that gap is a reason to start with the document-shaped part of your own business. If you are in a manufacturing firm and you have not started, you are not far behind. You are in a large group.

## How do you find the handoffs? Use the Job Trail Sheet

Pick one typical job and follow it from the first inquiry to the final invoice, listing every step and who touches it. This sheet is the artefact in this article, and you can copy it into a spreadsheet or onto a sheet of paper.

**The Job Trail Sheet, one row per step of one real job:**

| Column | What you write |
|---|---|
| Step | What happens, in five words or fewer (for example, "quote the inquiry") |
| Who touches it | The roles, not the names |
| What is retyped or re-explained | The information that moves from one place or person to another |
| Times a week | How often this step happens across all jobs |
| Minutes each time | A rough, honest figure |
| Sign-off | Mark Y if a person must approve it for safety, quality, legal or customer reasons |

Then multiply times by minutes for each row and rank the rows. The highest score with no Y in the sign-off column is where you start. If the top row has a Y, start at the highest row without one and come back later.

### What does a filled-in sheet look like?

Take a hypothetical 30-person sheet-metal fabricator that makes brackets and enclosures to customer drawings. The numbers below are invented for illustration, not data.

| Step | Retyped or re-explained | Times a week | Minutes each | Score | Sign-off |
|---|---|---|---|---|---|
| Quote the inquiry | Specs and quantities copied from the customer email into the quote template | 12 | 25 | 300 | N |
| "Where is my order" replies | Status looked up in the job list and typed into an email | 15 | 8 | 120 | N |
| Delivery notes | Details retyped from the work order | 20 | 5 | 100 | N |
| Drawing revision notices | Changes explained again to the floor | 8 | 10 | 80 | Y |

The quote step scores highest and has no Y, so the owner starts there. The first trial is small: an estimator pastes the inquiry into an AI assistant, asks it to list the specs, quantities and open questions in the quote template's order, and checks the list against the original before anything goes out. One estimator, one job type, four weeks. The drawing revision step scores lowest and carries a Y, so it stays with the person who owns quality until the first three have run clean.

That discipline matches the one piece of manufacturing advice from an official source that I found. In a January 2022 [NIST Manufacturing Extension Partnership blog post](https://www.nist.gov/blogs/manufacturing-innovation-blog/artificial-intelligence-manufacturing-real-world-success-stories), a vendor's founder recommended deploying AI "at a single site with a single line and then scale out to 2-3 lines before expanding to more sites." The post reports one vendor's account of about 90 projects since 2017 and is not an independent evaluation, so treat it as a practitioner's rule of thumb. Its logic applies equally to a desk job: start with one, prove it, then widen.

## Which steps should stay out of AI for now?

Keep AI away from any step where a wrong answer reaches product, a person's safety or a regulated record, until a named person has checked the AI's work for a full quarter. The rest of the sheet can wait its turn.

In practice, that means the Y rows on your sheet: release of a batch, inspection results, drawing revisions that change what is made, safety notices, and certificates of conformity. A human check only counts if it is a real one. Use the Loop Test I put to any leader who says a person checks the output:

**The Loop Test is three questions I ask before accepting "a human checks it" as an answer: Would that person notice a wrong result without being told to look for it? Would they notice in time to stop it mattering? And if they missed it, could the work be put back? Oversight that survives all three is a safeguard. Oversight that fails any of them is a signature.**

A quote check passes the Loop Test: the estimator knows the job and the customer sees nothing until a person has approved it. A certificate check usually does not, and nothing on a certificate can be put back once it has shipped. My piece on [training staff to check AI output](https://business.theaieducator.io/posts/train-staff-to-check-ai-output) shows how to teach the checking skill itself.

## What I See When I Train Business Teams

Across the business teams I train, the first AI idea is almost always the most exciting one and rarely the most useful. People reach for the clever job, such as predicting machine faults, because it sounds like what AI is for. The useful job is usually duller and closer to the desk: the email that takes twenty minutes because someone has to find three facts in three places.

There is a public case on [theaieducator.io/business](https://theaieducator.io/business) that shows how this tends to grow. In April 2023 I spoke about AI in marketing at a business event in Penrith. A year later the marketing team at theidol.com was using AI for content writing, and the managing director asked me back to show teams across the business what was possible. After the September 2024 session the IT Director wrote: "It was a great session and I've heard a lot of good feedback from the attendees." The pattern there was a small start with one team, a visible result and then a wider invitation, not a company-wide launch.

## What should a small manufacturer do this week?

Do these five things in the next five working days, with no budget meeting and no data project.

1. **Pick one job type.** Choose the one you quote or deliver most often, such as one family of parts or one customer type.
2. **Fill in the Job Trail Sheet for that job.** Do it with the estimator, the planner and whoever sends the invoice, in one 45-minute session. Use the people who retype, not only the people who manage.
3. **Score and rank the rows.** Mark every Y. Choose the top row with no Y.
4. **Write the trial on one line.** Who, which task, which AI assistant, what a good result looks like, who checks it, and the date you will decide. This is the Exit Question I put to any leadership team before a pilot: if this works, what will change, who has already agreed to it, and on what date will you decide? A trial with no answer is an experiment with no exit.
5. **Check the data before you paste it.** Decide what customer drawings, supplier terms and customer lists must never go into a general assistant, and say so in writing. [Is your business data ready for AI](https://business.theaieducator.io/posts/is-your-business-data-ready-for-ai) helps with this, and [questions to ask suppliers who use AI](https://business.theaieducator.io/posts/questions-to-ask-suppliers-who-use-ai) covers the other side of the same coin.

At the end of four weeks, ask whether the minutes on your sheet went down, whether anyone had to fix an error, and what the freed time was used for. My article on [measuring whether AI is working](https://business.theaieducator.io/posts/measure-whether-ai-is-working) shows how to take the before number.

## When should a small manufacturer wait?

Wait if no one owns the job, if the paperwork lives in people's heads, or if the only reason to start is that a competitor has. The non-adopters in the UK survey who cited no clear business case (65%) are not wrong to hold back. They are wrong only if they never look. An afternoon with the Job Trail Sheet is the quickest way to find out whether you have a business case.

If you are weighing AI agents that act for you, such as sending the confirmation or booking the delivery, read [what AI agents mean for small businesses](https://business.theaieducator.io/posts/ai-agents-for-small-businesses) first. Start with advice and drafts, then move to action. If your customers' purchasing teams are now asking about your AI use, [AI questions in tenders](https://business.theaieducator.io/posts/ai-questions-in-tenders-small-business) shows how to answer.

## Where to go from here

If your team keeps retyping the same details and nobody has shown them a better way to do the first draft, that is what my [AI training for teams](https://theaieducator.io/ai-training-for-business?utm_source=business.theaieducator.io&utm_medium=referral&utm_campaign=ai-for-small-manufacturers) is built for. If you want to know where your business stands before you pick a first job, take the [Business AI Readiness Scorecard](https://theaieducator.io/business-ai-readiness?utm_source=business.theaieducator.io&utm_medium=referral&utm_campaign=ai-for-small-manufacturers). And if you run a trade association, chamber or industry event for manufacturers, you can [book me to speak](https://theaieducator.io/business-keynotes?utm_source=business.theaieducator.io&utm_medium=referral&utm_campaign=ai-for-small-manufacturers). This article is general information and not legal, safety or financial advice.

## Sources and further reading

- Department for Business, Innovation, Science and Trade, "AI adoption survey report 2026: executive summary", GOV.UK, 29 October 2026 (updated 2 November 2026). [gov.uk](https://www.gov.uk/government/publications/ai-adoption-survey-2026/ai-adoption-survey-report-2026-executive-summary)
- Office for National Statistics, "Artificial intelligence in UK businesses: 2023 to 2026", 20 July 2026. [ons.gov.uk](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026)
- Bonney, Breaux, Dinlersoz, Foster, Haltiwanger and Pande, "The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks", US Census Bureau Center for Economic Studies working paper CES-26-25, April 2026. [census.gov](https://www.census.gov/library/working-papers/2026/adrm/CES-WP-26-25.html)
- Katie Rapp, "Artificial Intelligence in Manufacturing: Real World Success Stories and Lessons Learned", NIST Manufacturing Innovation Blog, 7 January 2022. [nist.gov](https://www.nist.gov/blogs/manufacturing-innovation-blog/artificial-intelligence-manufacturing-real-world-success-stories)
- Doug Devereaux, "Smaller Manufacturers Get Lean with Artificial Intelligence", NIST Manufacturing Innovation Blog, 6 March 2019. [nist.gov](https://www.nist.gov/blogs/manufacturing-innovation-blog/smaller-manufacturers-get-lean-artificial-intelligence)

*[Dan Fitzpatrick](https://theaieducator.io/about) is a keynote speaker, Forbes contributor and five-times bestselling author who has trained more than 150,000 educators in 30+ countries.*

## Frequently asked questions

### Where should a small manufacturer start with AI?

Start where your people retype or re-explain the same information most often, usually the quote, the work order, the delivery note or the customer status reply. These are handoffs between people, which AI can draft while a person checks. Leave machines, inspection and safety decisions with people at first.

### Can a small manufacturer use AI without a data team?

Yes. The first useful jobs are document-shaped, such as turning an inquiry email into a quote checklist, and need no sensors, models or factory data. Use a general assistant, keep customer drawings and supplier terms out of it unless your policy allows them, and have the person who owns the job check every result.

### Which manufacturing tasks should stay with people?

Keep any step where a mistake reaches product, safety or a regulated record with a named person: batch release, inspection results, drawing revisions that change what is made, safety notices and certificates. A human check only counts if the person would notice a wrong result, in time, and the work could be put right.

### How do I find where my team wastes time on paperwork?

Follow one real job from inquiry to final invoice and list every step, who touches it, what is retyped or re-explained, how often it happens and how many minutes it takes. Multiply times by minutes for each step. The highest score without a sign-off requirement is your first candidate.

### How many manufacturers are using AI?

The sources I read give no manufacturing figure in their text. The UK Department for Business, Innovation, Science and Trade found 32% of UK businesses used at least one AI technology in 2026, and a US Census Bureau paper found 18% of firms used AI in a business function.

### How long should a first AI trial take in a small factory?

Four weeks is enough for a first trial: one job type, one person, one task, a named checker and a decision date written down in advance. At the end, compare the minutes on your sheet, count the corrections, and decide what to widen, keep or stop.

---
Source: [Where Should a Small Manufacturer Start With AI First?](https://business.theaieducator.io/posts/ai-for-small-manufacturers)
Publisher: [The AI Educator](https://theaieducator.io)