Ting Guides · For the COO
Where should a Saudi manufacturer start with AI, and where should it wait?
The short answer
Start where engineers, planners and buyers spend hours reading and retyping records, not on the machines. Good first candidates are drafting nonconformance reports from inspection results, checking supplier certificates against purchase orders, and summarising maintenance history before a breakdown review. Wait on predictive models and automated quality decisions until sensor history, clean master data and an accountable owner exist.
In this guide11 sections
- What it depends on
- Which factory tasks are worth AI first, and which should wait?
- When is AI the wrong tool on the shop floor?
- What has to be in place for each first use case?
- What should the first 90 days cover?
- Why do factory AI projects stall?
- What should you ask a vendor selling AI to a factory?
- What the Saudi industrial context changes
- One first use case, three sources: the nonconformance report
- Questions people ask
- What Ting does and doesn't do here
- Sources
What it depends on
How much of the plant's record is still on paper
Inspection sheets, shift logs and work orders on clipboards can be scanned and read, but only if they were filled in. AI can't recover a reading nobody wrote down, so check what the paper actually holds before you plan around it.
Whether master data agrees across systems
If an item, a lot or a machine carries one code in the ERP, another in the quality records and a third in maintenance, any answer that joins them will be wrong some of the time. Mapping the codes is usually the first piece of work, and it pays off with or without AI.
Whether the machines are connected
Predictive maintenance and real-time quality need signal history from controllers or a historian, with past failures labelled against it. Without that history, start with the text records your teams already keep.
Who signs off quality and safety decisions
Release, rework and scrap stay with the quality manager, and incident findings with the safety lead. AI can draft and flag for them. In food, pharmaceuticals and medical devices, where a sector regulator inspects the records, the bar for any automated step is higher still.
Which factory tasks are worth AI first, and which should wait?
Six areas where a manufacturer with an ERP, several lines and paper-heavy records can look for a first use case.
| Area | A good first use | Records it reads | Wait if |
|---|---|---|---|
| Quality inspection | Drafts the nonconformance report from the inspection result, the defect photos and the lot record, for the quality engineer to complete | Inspection sheets, defect photos, lot and batch records, past reports with their dispositions | The plan is to automate accept and reject with no labelled defect images, or to replace a trained inspector on day one |
| Maintenance | Summarises a machine's work orders, fault notes and spare parts used before a breakdown review or a shutdown plan | Work orders, fault notes, parts issued per asset | You want failure prediction but the historian holds no failures labelled against the signals |
| Production planning across the ERP | Answers planners' questions that span orders, stock and capacity, and explains why the plan changed since yesterday | Open orders, stock, routings, capacity per line | Bills of materials and routings in the ERP are known to be out of date |
| Procurement and supplier documents | Checks certificates, delivery notes and supplier invoices against the purchase order and lists every mismatch | Purchase orders, the certificate each item requires, documents as received | Nobody has agreed what a complete document set is for each item |
| Safety reporting | Turns shift-end incident and near-miss notes into structured reports and points out causes that repeat across lines | Incident and near-miss notes, agreed categories, line and shift | Reports would be used to blame individuals, or there are no agreed incident categories yet |
| Plant reporting | Assembles the daily production pack from ERP and line records and explains variances in output, scrap and downtime | ERP output, line logs, downtime records | Each line still records downtime its own way; agree the definitions first |
When is AI the wrong tool on the shop floor?
What has to be in place for each first use case?
For each candidate, name the owner, confirm the data can be read, and find the interface before anyone writes a line of code.
| Use case | Process owner | Data access | System interface |
|---|---|---|---|
| Nonconformance drafts | Quality manager | Inspection results, lot records and past reports with the disposition each one received | Read access to the quality records or inspection spreadsheets, plus a lot lookup in the ERP |
| Maintenance history summaries | Maintenance manager | Work orders, fault notes and parts issued, against one asset register | An export or read-only view of the maintenance system, with asset codes that match the ERP |
| Planner questions | Planning or supply chain manager | Open orders, stock, routings and capacity | Read-only ERP views or a reporting database refreshed at least once a shift |
| Supplier document checks | Procurement manager, with quality for certificates | Purchase orders, the required certificate list per item and the documents received | Purchase order lookup in the ERP, and the inbox or portal where supplier documents arrive |
| Safety reports | Health, safety and environment manager | Incident and near-miss notes, agreed categories, the line and the shift | The reporting form or log, with access limited to the safety team |
| Plant reporting | Plant manager or the COO's office | Agreed definitions of output, scrap and downtime for every line | Scheduled exports from the ERP and the line records |
What should the first 90 days cover?
Weeks 1-2: walk one product family through the plant
Follow one recent order from the supplier's documents to finished-goods release. Note each point where someone reads a document and types its contents somewhere else, and where work waits for a signature.
Weeks 3-4: choose one use case and record the starting point
Pick the task with the most reading and retyping and the lowest consequence if a draft is wrong. Record the current figure: hours per nonconformance report, or days from goods receipt to certificate check.
Weeks 5-6: open the data under a named service account
Agree read-only access with IT, map the item, lot and asset codes the use case joins, and collect past cases to test against, including the awkward ones.
Weeks 7-11: run it beside the people who do the work today
On one line or one supplier group, the AI drafts and the engineer or buyer edits and signs. Keep every edit: it is the test set that shows where the drafts go wrong.
Weeks 12-13: decide with the plant manager
Compare the result with the starting point. Extend to more lines, fix and run again, or stop, and record the reason with the figures.
Why do factory AI projects stall?
| Failure | How it shows up | What prevents it |
|---|---|---|
| Codes that don't match | The assistant ties a defect to the wrong batch because the quality records and the ERP use different item codes | Map item, lot and asset codes across systems before the build, and test lookups in both directions |
| A draft quietly becomes the decision | Engineers start approving AI-proposed dispositions without reading them | Leave the disposition field for the engineer to fill, and sample-check signed reports every week |
| Prediction with no failures to learn from | A predictive maintenance pilot finds patterns but has no recorded failures to check them against | Check the failure history first; start with work-order summaries until enough labelled failures exist |
| The pilot line isn't the plant | It works on the newest line with digital records and stalls on the lines that still run on paper | Pick a pilot line that records work the way most lines do, or fix capture there first |
| Safety data handled like any other data | Injury details reach a wide audience through AI summaries | Leave out names that aren't needed, restrict access, and treat injury details as health data |
What should you ask a vendor selling AI to a factory?
| Question | What a sound answer includes | Treat it as a warning if |
|---|---|---|
| Which of our records will it read, and from which system? | The ERP, quality, maintenance and document sources named, with the interface and access each one needs | The vendor wants a one-off data dump and offers no live connection |
| Can it run on our own inspection sheets and certificates? | A test on your scans, in Arabic and English, with the errors counted | Only clean sample documents are shown |
| What does it decide on its own? | Nothing that releases, scraps or ships product; it drafts and flags | It sets dispositions or changes ERP records without an approval |
| How will we know when its output gets worse? | A review rhythm, the measures watched and the person who acts on them | The answer is that the model improves itself |
| For predictive maintenance: what failure history is the model built on? | Named assets, the period covered and the failures behind it | A dashboard demo with no validation on your own machines |
| What do we keep if we stop? | Your data, configuration, prompts and test cases in a usable format | The configuration only exists on the vendor's platform |
What the Saudi industrial context changes
General information, not legal advice. Each regulatory point summarises the official source cited beside it, as checked on the date shown under Sources; confirm how it applies to your company with your legal adviser.
The National Industrial Strategy
The Ministry of Industry and Mineral Resources describes the National Industrial Strategy, launched by HRH the Crown Prince in October 2022, as a roadmap to accelerate and diversify industrial development, aiming for a resilient, competitive and sustainable industrial economy led by the private sector. It focuses on 125 industrial product groups across 12 industrial sectors.[1]
Future Factories: automation with a defined path
The ministry's Future Factories Program aims to shift 4,000 factories from dependence on low-wage, low-skilled labour towards operational efficiency, automation and advanced industrial solutions. It has two paths: new factories built to high manufacturing-efficiency standards, and existing factories converted to apply operational excellence and advanced technologies.[2]
If your plant has been assessed, start from that plan
The ministry describes the journey as a self-assessment, then an audited evaluation by a certified assessor, then a transformation plan; grants cover the audited evaluation and the plan. As of 10 May 2024 it reported over 700 factories evaluated, 370 with transformation plans and about 150 at the target level. If your plant has a plan, look for AI candidates in the gaps it already lists.[3]
Industry and logistics under one program
The National Industrial Development and Logistics Program, launched in 2019, covers industry, logistics, energy and mining, and describes its work as paving the way for Saudi Arabia's Fourth Industrial Revolution.[4]
Injury details are sensitive data
The Personal Data Protection Law (Article 1) defines sensitive data to include health data. An incident report that records an injury contains it, so decide which tools may process those reports and who may read their summaries before any AI touches them.[5]
One first use case, three sources: the nonconformance report
ExhibitWorked example
Rasikh Industries, the HVAC plant of the fictional Rasikh Holding, writes nonconformance reports by hand after incoming inspection. This is how a read-only first version splits the work between three record sources and the quality engineer.
- Quality recordsReads the inspection result and defect photosNo write
- ERPReads the lot, its supplier and where it was usedNo write
- Supplier fileReads the certificate that came with the lotNo write
Assistant
Drafts the nonconformance report
Quality engineerSets the disposition and signs
Report issuedOnly after the engineer signs
| Step | Today, by hand | With the assistant, read-only |
|---|---|---|
| Collect the inspection result | The inspector's sheet is typed into a spreadsheet at the end of the shift | The assistant reads the result and the photos from the quality records |
| Trace the lot | An engineer searches the ERP for the lot, its supplier and the orders that used it | It reads the lot's supplier and usage from the ERP |
| Check the certificate | Someone looks for the supplier's certificate in an email folder | It reads the certificate filed at goods receipt and notes any value outside the specification |
| Write the report | The engineer writes the report from a blank template | It drafts the report; the engineer sets the disposition, edits and signs |
The decisionNothing here writes to the ERP or decides what happens to the lot: the engineer still sets the disposition and signs. Measure hours per report against the starting point, then decide whether to extend to a second line.
Questions people ask
Do we need sensors and connected machines before using AI?
Can AI decide whether a batch passes inspection?
Should AI replace the planning module in our ERP?
We are in the Future Factories Program. Where does AI fit?
What Ting does and doesn't do here
What we do
- We walk a product family from supplier documents to dispatch across your plants, systems and shifts, and map where records wait or get retyped.
- We rank candidates in quality, maintenance, planning, procurement and safety by impact and ease, naming the owner, data and system interface each one needs.
- We state the expected return with its assumptions written down, and say which candidates should not become AI projects at all.
Where Ting helps: AI Transformational Audit
What we don't do
- We don't connect AI to control systems, interlocks or anything that can start or stop equipment.
- We don't make quality or safety decisions; they stay with the people accountable for them.
- We are not a Future Factories assessor, and our Audit is not a ministry evaluation.
Sources
Primary sources only, each read in full on the date shown.
- Ministry Achievements: National Industrial Strategy (October 2022)Ministry of Industry and Mineral Resources · checked
- Future Factories Program InitiativeMinistry of Industry and Mineral Resources · checked
- Factories of the Future program at Riyadh Industry Week 2024 (news, 10 May 2024)Ministry of Industry and Mineral Resources · checked
- National Industrial Development and Logistics ProgramSaudi Vision 2030 · checked
- Personal Data Protection Law (Royal Decree M/19 of 9/2/1443H, amended by Royal Decree M/148 of 5/9/1444H)SDAIA · checked
