Four kinds of AI, one input
AI with something real to work from.
Ask most manufacturing software what its AI runs on and the honest answer is: numbers somebody typed in. Ours runs on what your machines actually did and what your paperwork actually said.
Every other manufacturing AI is prompted with numbers a human typed.
Blitz's AI is prompted with the factory itself.
Read this first
Where do I start with AI in my factory?
Search that question and you get the same list every time. Assemble a cross-functional team. Hire data scientists, or a consultant who has them. Pick a pilot project. Start with predictive maintenance, or computer-vision quality inspection.
That advice is written for plants a hundred times your size, by companies who sell to plants a hundred times your size. Read it as an owner with twenty or fifty machines and the sane conclusion is that artificial intelligence is something to look at in a few years.
It isn't. The list is just wrong for you.
You do not need a data science team. You do not need a machine learning model of your own — nobody at this size is training anything. You do not need a server in the corner, a new MES, or eighteen months. What you need is a record worth reasoning over, and that is the part every guide skips.
Fix the input, not the model
AI adoption fails in a small factory for a boring reason. The numbers it would reason over were written on a sheet at the end of a shift and keyed in on Saturday. Ask the smartest model in the world what your cost per unit is and it will confidently repeat that guess back to you.
So the first move has nothing to do with AI. Get output measured off the machine itself. That is a sensor and an afternoon, not a project.
Point it at the paper you already handle
Every factory has a pile of purchase orders and supplier bills moving slowly toward a keyboard. That backlog is why your costing runs on last month's rates.
This is the first place AI pays for itself, because the job is narrow, it happens daily, and you can see whether it got it right. Photograph the bill, check the filled-in form, save.
Then start asking it things
Only now is there any point. With a measured record and current rates underneath it, you can ask which product actually made money last month, which customer has quietly stopped ordering, and who else might buy what you make.
Do it in the other order and you have an expensive way to be wrong faster.
None of that asks you to change how you make anything. No new routine on the floor, no operator keying figures into a tablet, no rip-and-replace of the software you run the office on. The longer version of this argument is here, and what Blitz does with it is below.
Why this is the whole argument
An AI is only as good as what you give it.
Put Blitz next to a mid-market ERP with an AI module and a dozen rows come back green on both sides. The grid is measuring the wrong thing: it scores whether a feature exists and is blind to the data underneath it.
A production figure somebody keyed in at the end of a shift and a production figure measured off the machine look identical in a database. They are not the same number, and everything computed from them inherits the difference. That is true of a costing report, and it is doubly true of an AI asked to draw a conclusion.
Blitz's answer is the same one it gives everywhere else: measured, not typed — including what the AI is told.
The four
Reads in. Reads back. Reaches out. Answers.
Four jobs, one shared foundation: a record of what your factory actually made, what it cost, what it sold and to whom, on real dates.
It reads your paper in
A vision model reads the purchase orders and GST invoices that arrive as photos and PDFs, and hands you a filled-in form to check. That is what keeps the rates behind your cost per unit current instead of three weeks stale.
Nothing is saved until you have looked at it. The form is a draft until you approve it, so a misread costs you a correction on screen rather than a wrong figure in your accounts.
See how it reads a billIt reads your own record back to you
Open a customer and Blitz writes you a profile: what kind of business they probably are, what they likely manufacture and for whom, why their peak season falls when it does, and what could help or hurt them.
What makes it different: the order cadence, the three-month growth trend, the peak quarter and the top products it reasons from are computed from closed machine sessions — sensor-measured production, not notes somebody typed into a CRM. Nothing is written back to your database.
It takes your record outward
This is the part people don't expect from factory software. Blitz finds you new customers — companies in the same industries as the ones you already supply, who would consume what you make.
Each prospect arrives with a plain-English reason naming which of your existing customers it resembles, so you can judge the fit before you pick up the phone. Point it at raw material instead and it shortlists suppliers for the grade you actually buy, and drafts the enquiry.
It answers you, and reaches you
Jarvis sits on all of it. Set the line that matters and it watches, staying silent until that line is crossed — a material below your limit, a machine down too long, a payment overdue. Ask it anything in plain language, by text or voice. Get a morning briefing without opening a thing.
On Blitz Plus. Jarvis reads the same live data your reports are built from — not a second set of numbers.
Meet JarvisAI in a factory, answered.
How does AI actually help a manufacturing business?
In a small or mid-sized factory the useful answer is narrower than the headlines suggest. AI earns its place in four jobs: reading the paperwork that arrives, so your costs are current instead of three weeks stale; reading your own production record back to you, so you can see which customers are growing and which have gone quiet; taking that record outward to find customers and suppliers who match the ones you already serve; and answering questions in plain language so you don't have to open a report. Blitz does those four.
What makes this different from other AI tools?
What it is working from. Most manufacturing AI reasons over figures somebody keyed in at the end of a shift. Blitz reasons over what the machines actually did — output measured at the machine, and the real rates off your own paperwork. A production figure that was typed and one that was measured look identical in a database, and everything computed from them inherits the difference.
Is the AI included, or is it an add-on?
Reading your paperwork, customer analysis, and finding customers and suppliers are all part of the Blitz platform. Jarvis — the agent that watches the lines you set, answers you in plain language and briefs you daily — is Blitz Plus.
Do I have to change how my factory works to use it?
No. Sensors fit onto the machines you already run, with no PLC and no integration project, and your paperwork keeps arriving the way it always has — as a photo, a PDF or a printout. Nobody on the floor has to learn a new routine for the AI to have something to work from.
Where do I start with AI in my factory?
Not with the AI. Start by fixing what it would read. In most small factories the production numbers were written on a sheet at the end of a shift and keyed in days later, so anything reasoning over them is reasoning over a guess. Get output measured off the machine itself first — that is a sensor, not a project. Then point AI at the purchase orders and supplier bills you already handle, because the job is narrow, it happens daily and you can see whether it got it right. Only then does asking it questions about the business return anything you can act on.
Do I need a data scientist or an IT team to use AI in manufacturing?
No. Nobody at this size is training a machine learning model of their own, and you do not need a server, an in-house IT department or a consultant to run a pilot. That advice comes from guidance written for large plants. Blitz is hosted and managed for you; the setup work is fitting sensors to machines you already own, which needs no PLC and no integration project.
Will AI replace my workers?
Not in the way the headlines suggest, and not here. The AI in Blitz does not run a machine, weave, mould or cut anything. What it removes is the copying — bills and purchase orders retyped into a computer, production figures transcribed off a sheet, reports assembled by hand on a Saturday. Your operators keep doing exactly what they do now; the people in your office stop spending their week on data entry.
Is AI worth it for a factory my size?
It depends entirely on what you point it at. Predictive maintenance and computer-vision inspection — the two use cases every article leads with — need volumes of failure and defect data that a 20 to 50 machine plant will not have for years, and that is why AI feels out of reach at this size. The jobs that pay back immediately are the unglamorous ones: reading the paperwork so your costs are current, reading your own record back so you can see which customers are drifting away, and answering questions without you opening a report. Those work on day one because they need your data, not a training set.
Can AI reduce my manufacturing costs?
Indirectly, and the honest version matters. AI does not lower your electricity tariff or your yarn rate. What it does is make the true cost of each product visible while you can still act on it — which is usually the first time an owner learns that a product they have been quoting for years loses money at the current rate. The saving comes from what you do with that: re-quoting, dropping a loss-making line, or moving a job to a cheaper machine. Blitz shows you the number; the decision stays yours.
See it run on your factory.
Book a 20-minute demo. Bring one of your invoices and one of your customers, and we'll show you both ends of it on real paper and a real name.