AI readiness assessment for a manufacturing plant

An AI readiness assessment tells you whether AI can work on your company’s data before you pay for a build. This one is written for a manufacturing plant. Its twelve questions check four things you can confirm this week: whether your records come out of the ERP, whether your systems agree with each other, who owns each answer, and who approves what AI drafts. The questions are scored in your browser, and you do not need to give an email to see the result.

Derik Lawlis Founder, ThriveAI

The AI readiness assessment

Twelve questions sit below, in four sections of three. Each question has three answers, worth 0, 1 or 2 points, and you can answer every one from what you already know about your plant.

Your answers are scored in this browser tab, and they are not sent to our server unless you choose to send your result. This site uses PostHog analytics on every page. It records page visits and where visitors click and scroll, and on this page it also records that the assessment was started and completed. The text of the questions, your answers and your result is kept out of those records. If you want a written answer, a form under your result sends the result to Derik with the name and email address you enter.

What the assessment measures

The assessment scores four areas, in the order they block each other: the record, agreement between systems, ownership of the answer, and permission. A quoting or costing tool cannot read hours that were never written against a job. A part number matched across three systems is worth little if nobody may decide which number is right. Each area carries three of the twelve questions, and each question points at something you can check this week. For what gets built once the four areas hold, see AI for manufacturing.

The record: can you get the data out

The first area asks whether someone inside can export a year of closed jobs from your ERP or system of record without calling the vendor. It asks where hours land, because hours on a payroll week say nothing about a job. It asks whether you hold the price you paid for material, or only a standard cost set once and left alone.

Cost per part depends on all three, and every quoted price starts from cost per part. How AI reads an existing ERP, including older ones with no API, is covered in AI for ERP.

Agreement: do your systems say the same thing

The second area asks whether a part number means the same thing in the ERP, the drawing vault and the price list. It also asks where customer requests arrive and where drawings live. Where two systems disagree, anything built on either one produces a number someone corrects by hand, and that correction is the work the tool was meant to remove.

Two systems that disagree can be reconciled. The question is whether a written rule can map one to the other, or whether the mapping lives in one person’s memory.

Ownership: does one named person own the answer

The third area asks who decides the right number when a system produces a wrong one, how many people can price work the shop has not made before, and who can give a day a week to a build. A shop where one person holds the pricing method can still build, and the build then moves at that person’s pace.

Permission: can the data leave the building, and who approves

The fourth area asks whether a written rule covers what staff may paste into a public chat tool, whether customer contracts say where their data may be stored or processed, and who signs off on a drafted quote before it reaches a customer. The AI policy template helps with the written rule. Data sovereignty in Canada covers what the law and your contracts ask about where data sits, and human in the loop covers who approves what AI drafts.

How the score is built

Each of the twelve questions scores 0, 1 or 2, for a maximum of 24. Nothing is weighted, because no weighting has been measured here. Read the number as a way to sort your plant into one of four situations, and act on the gaps listed under your result. There is no percentile or peer comparison, because no group of plants has been measured to compare against.

What each result means next

The record is not there yet. Get one year of closed jobs out of the system of record and into a file, and start putting hours against job numbers. The first is an operations task and the second is a habit on the floor, and neither needs new software. AI for ERP covers what AI can read from the ERP you already run.

One good record, the rest disagree. Take the two systems that disagree most and write the rule that maps one to the other, starting with the part number. AI for manufacturing shows how each module reads one shared record.

Ready for one module. Pick the job with the strongest record behind it and build that one first. AI for small business covers where AI helps a smaller plant first, and the cost of AI explains how each part of a build is charged.

Ready to build, and the open question is legal. Settle where the data sits and who approves a draft before it leaves the building, and put both answers in writing before the first tool goes in. Data sovereignty in Canada covers the first question and human in the loop covers the second.

Why a general AI readiness framework misses a plant

The published frameworks are real instruments. Cisco’s AI Readiness Index, according to its published methodology, scores six pillars across 49 indicators: strategy, infrastructure, data, governance, talent and culture. It weights each pillar by its importance to overall readiness. Microsoft’s AI Readiness Wizard asks 10 questions across five areas: business strategy, technology and data strategy, AI strategy and experience, organization and culture, and AI governance and security.

Their questions ask about AI objectives, use cases, AI teams, skills and governance controls. None of the Microsoft wizard’s ten questions asks where hours are recorded or whether the ERP exports, and those two answers decide whether a costing tool has anything to read. A plant with no AI strategy document can still be a good place to build, if its time records are clean.

Cisco’s own results point at the record. In its latest index, 76% of Pacesetters, the group Cisco scores as fully prepared, have fully centralized data, against 19% of organizations overall.

Where your data sits if you build

ThriveAI is an AI engineering company in Ottawa that builds private AI systems for manufacturers and distributors in Ontario and Quebec, on their own data. Derik Lawlis, the founder, leads every project and stays close to the build.

The platform ThriveAI builds on is designed to keep each client’s data on its own server in Canada. You choose the model that reads it: one that runs on that server, or a hosted model under a written zero data retention agreement. A hosted model may process requests outside Canada, so the contract names the model and its service tier. The systems read your ERP as it is, including older versions installed on your own server, and nothing is sent or saved in your systems until the person responsible approves it.

The rules on where data may sit are in data sovereignty in Canada, and the ways to set up private AI are compared in private AI for business. For the company and how a project runs, see About ThriveAI.