# Real-Time AI Inventory Tracking for Metals Warehouses



> Missing coils, messy mill certs and a dispatch scramble every shift. See how to fix stock, offcuts and traceability with AI, and hit 98-99% accuracy.
> 
> **URL:** https://www.gosmarter.ai/blog/how-to-track-inventory-real-time-with-ai/

**Date:** 2026-08-20
**Author:** BlogSmarter AI

**Categories:** blog

**Tags:** artificial-intelligence, automation, data-strategy, inventory, manufacturing

## 



**Most metals firms do not track stock in real time. They guess, then pay for the gap.** A coil lands, the mill cert sits in an inbox, an offcut disappears into limbo, and dispatch turns into a last-hour scramble. If you want live inventory with [Artificial Intelligence (AI)](/hubs/metals-manufacturing-glossary/), start by fixing bad stock records. Then record each move at the point of work.

The pain is simple. **Missing [heat numbers](https://gosmarter.ai/docs/what-is-a-heat-number/), lost offcuts, late updates, and scrap nobody can pin to a job** burn cash and make audits harder than they need to be.

I would keep it simple. [GoSmarter](https://gosmarter.ai/), built by [Nightingale HQ](https://nightingalehq.ai/categories/offerings/), sits on top of your existing [Enterprise Resource Planning (ERP)](/hubs/metals-manufacturing-glossary/), Warehouse Management System ([WMS](/hubs/metals-manufacturing-glossary/)), Excel, or email setup and ties live stock events to mill certs, offcuts, and scrap. No rip-and-replace project needed. It reads mill certs, checks stock moves, spots bad locations, and helps your team pick and load with the right traceability in view.

What you get:

-   **A clear starting point** for barcodes, [Radio Frequency Identification (RFID)](/hubs/metals-manufacturing-glossary/) tags, scales, cameras, and machine signals
-   **Tighter traceability** between stock, heat numbers, and certs
-   **Less stock chasing** and fewer end-of-shift fixes
-   **Better offcut control** so usable metal does not vanish into scrap
-   **Cleaner dispatch**, with the right material and paperwork leaving together
-   **A pilot path** you can test in one bay before you roll it across the site

> Real-time stock tracking is not magic. It is just fewer blind spots and less daft admin.

**What is real-time inventory tracking?** Every stock event posts to your system the moment it happens. A coil arrives, a cut gets made, an offcut goes back on the shelf, and the record updates straight away. Nobody waits for an end-of-shift update or a weekend stocktake to find out what is actually in the yard.

Here's how to fix it.

## AI in Action: How AI Transforms Inventory Management

{{< youtube width="480" height="270" layout="responsive" id="bpisglNNrUY" >}}

## Find where your stock data breaks down before adding any AI

Before any Artificial Intelligence ([AI](/hubs/metals-manufacturing-glossary/)) can help, you need to know where your stock process goes off the rails. Start on the shop floor. Walk the route. Map every point where material changes state, not just the obvious bits.

### Trace every stock movement from gate-in to dispatch

In a UK metals plant, stock usually moves through the same chain: goods receiving, quality checks and [mill certificate capture](https://gosmarter.ai/blog/ai-tools-compliance-driven-document-operations/), tagging and yard put-away, cutting and processing, offcut handling, returns to stock, scrap collection, and final picking and dispatch. The trouble starts where data falls out. That usually happens at receiving, tagging, put-away, cutting, offcut return, scrap, and dispatch.

When that happens, you get the usual mess:

-   **late stock updates**
-   **lost heat numbers**
-   **stock sitting in the yard but missing from the system**

Run a paper-and-floor audit over a week that looks normal for your site. At each stage, write down:

-   what people physically do
-   what the system records
-   what nobody records at all

That gap between the second and third points is the real problem list. It also shows where AI will fall flat if you bolt it on now. Your live tracking setup needs to catch those exact events.

### Clean up SKUs, units, locations, and heat numbers

Before you connect live data, sort out four fields:

-   one Stock Keeping Unit (SKU) per item
-   one unit standard per material family
-   one fixed location list
-   mandatory heat numbers for all traceable stock

This sounds dull. It is dull. But messy master data wrecks everything downstream.

Even a small error rate snowballs when you have hundreds of stock moves each week. If those four basics are messy, your AI system will learn from bad inputs and give you polished nonsense back. A missing dimension field on a structural section is not just a reporting headache either. It can send an AI put-away suggestion to a rack that physically cannot take the profile.

### Choose the KPIs that show whether this is worth the cost

Baseline five Key Performance Indicators ([KPIs](/hubs/metals-manufacturing-glossary/)) before you change the process. These numbers show whether live tracking cuts waste, delays, and manual stock chasing.

| KPI | What it measures | Why it matters in GBP |
| --- | --- | --- |
| Inventory accuracy (%) | System quantity vs physical count for a sample of high-value SKUs | A small gap between the system and the yard soon turns into write-offs, panic buys, and wasted hours |
| Stockouts per month | Production stoppages caused by unavailable material | Each event can mean lost machine time, overtime, and urgent buying |
| [Scrap rate (%)](https://gosmarter.ai/blog/how-to-calculate-scrap-rate/) | Scrap as a percentage of input material by kg or tonnes | Lower scrap cuts material spend and disposal cost |
| Reconciliation time (hours/month) | Hours spent fixing stock discrepancies | Multiply hours saved by fully loaded labour cost per hour |
| Working capital in stock (£) | Total value locked in raw stock and Work In Progress ([WIP](/hubs/metals-manufacturing-glossary/)) | A 10% reduction on a GBP 2,000,000 inventory frees GBP 200,000 in working capital [\[8\]](https://www.pwc.co.uk/services/value-creation/insights/working-capital-study.html) |

Best-in-class inventory accuracy sits at 98-99% [\[9\]](https://cfoiquk.com/manufacturing-finance-kpis/)[\[10\]](https://www.cleverence.com/articles/for-business/inventory-management-performance-metrics-4827/). If you are sitting closer to 85-90%, that gap is already costing you. Manual counts, spreadsheet updates, and operator memory often sit far lower still, sometimes around 63% [\[4\]](https://ifactoryapp.com/ai-vision-camera/ai-warehouse-automation-and-inventory-management-using-computer-vision). That range is common in operations that still rely on periodic counts and shift-end updates rather than live capture.

Get these numbers now, before you roll out new tech. Then use them to judge whether the shop-floor signals in the next step actually work.

Once you know the weak points and the baseline, the next step is to capture each move as it happens.

## Set up the shop-floor signals that record what moved, cut, or went missing

{{< image src="6a86fc1bdc1e9c396e6c9004-1787233103034.jpg" alt="Barcode vs RFID Inventory Accuracy in Metals Manufacturing" >}}

Once you know where the data falls apart, fix the **capture layer**. That means the physical signals that tell your system what happened, when it happened, and where. Start at the point of capture. Not at the desk. Not at the end of the shift.

A live tracking setup uses barcodes, Radio Frequency Identification ([RFID](/hubs/metals-manufacturing-glossary/)), scales, cameras, and machine data. It posts each event to your Enterprise Resource Planning ([ERP](/hubs/metals-manufacturing-glossary/)), Warehouse Management System ([WMS](/hubs/metals-manufacturing-glossary/)), or Manufacturing Execution System ([MES](/hubs/metals-manufacturing-glossary/)) within minutes.[\[18\]](https://aws.amazon.com/marketplace/pp/prodview-mpjchfvch5umg)[\[17\]](https://www.parsec-corp.com/blog/rfid-for-inventory-management)[\[16\]](https://knarrtek.com/wiptracker-wip-tracking-system/)

### Start with barcodes, then add RFID where the pain justifies it

For most UK metals plants, barcodes are the right place to start. They are cheap, simple, and good enough for most movement tracking if your team actually uses them.

Print durable Code 128 or Quick Response (QR) labels at goods receipt. Use one label per bundle, coil, plate, or cut length. Put the Stock Keeping Unit (SKU), heat number, bundle ID, and quantity or length on the label. Place labels where drivers and crane operators can scan them at ground level. Then make scans part of the job at every movement point:

-   into and out of the rack
-   into the saw bay
-   onto the lorry

That is what keeps heat numbers tied to the right stock location. Following [digital traceability best practices](https://gosmarter.ai/blog/digital-traceability-metals-best-practices/) ensures these records remain audit-ready.

RFID costs more, so earn the spend. Use it where barcodes keep failing. If you move heavy coils by overhead crane, line-of-sight scanning can be a faff. Oil and mill scale often wreck printed labels. Passive Ultra-High Frequency (UHF) RFID tags with fixed portal readers can record those movements on their own instead. No manual scan needed. Industrial RFID tags built to survive dust, heat, and rough handling also help where line of sight stays poor no matter what you do.

RFID-enabled warehouses typically hit **99.0-99.8% inventory accuracy**[\[5\]](https://ifactoryapp.com/stock-management/warehouse-barcode-rfid-tracking-manufacturing-plant)[\[7\]](https://oxmaint.com/industries/delivery-operations-management/real-time-inventory-tracking-rfid)[\[15\]](https://cybra.com/rfid-faqs/how-accurate-is-rfid-compared-to-barcode-scanning/)[\[19\]](https://www.hashmicro.com/barcode-inventory-system). Barcode-led operations usually sit around **85-95%**[\[5\]](https://ifactoryapp.com/stock-management/warehouse-barcode-rfid-tracking-manufacturing-plant)[\[7\]](https://oxmaint.com/industries/delivery-operations-management/real-time-inventory-tracking-rfid)[\[14\]](https://omneelab.com/barcode-vs-rfid-warehouse-management-cost/). That gap matters when you track high-value or safety-critical stock. We've seen sites lose a single aerospace-grade coil worth tens of thousands of pounds. That one mistake wipes out years of savings from a cheaper label printer.

Start with barcodes. Add RFID only where missed scans happen a lot, or where mislocation burns cash.

### Add cameras and sensors to catch what scans alone miss

Physical checks break down in metals for very practical reasons. Grease and mill scale hide barcodes. Labels get torn off. Someone moves a bundle without scanning it, and now the stock record points to thin air.

Camera-based systems keep watching in the background. High-resolution cameras on forklifts, gantry cranes, or fixed points near staging areas grab images when stock moves. AI models read labels and codes, pull out heat numbers or stock IDs, and flag mismatches before you load the wrong material. Depth cameras and stereo vision can also estimate stack height and rough dimensions, catching miscounts that a barcode scan alone would miss.

The same cameras can spot condition issues too, such as rust streaks, edge damage, and torn labels, then send an exception to an inspector for review instead of letting suspect stock slip through. Vision-based systems that match records against scan data, camera feeds, and weight checks all the time, rather than only at a weekend stocktake, report continuous inventory accuracy above 99.5%[\[4\]](https://ifactoryapp.com/ai-vision-camera/ai-warehouse-automation-and-inventory-management-using-computer-vision) — well ahead of the manual baseline above. Human judgement stays in the loop throughout. AI just spots the faults before dispatch.

### Turn every receipt, move, cut, and issue into a live stock event

Once the capture points are in place, make every physical move post at once.

At goods receipt, the operator scans the delivery note or Advanced Shipping Notice ([ASN](/hubs/metals-manufacturing-glossary/)) number and the barcode on each bundle or coil as it comes off the lorry. The system logs a receipt event with supplier, weight in kilograms, date such as **20/08/2026**, and first location.

When a forklift driver puts a coil down in the destination bay, they scan the coil and the bay code. The move event updates the system location straight away.

At the saw line, the operator scans the source bar and the job number before cutting. As each piece is cut, the system cuts the bar's remaining length and issues material to the work order. Offcuts above **500 mm** get a new barcode and go back into stock on their own. Shorter pieces go to scrap. Each cut event links back to the parent stock record, so the offcut gets its own entry with length, grade, heat number, and rack location. That turns it into searchable stock instead of yard clutter, and lets automated cut planning check that offcut catalogue before it books out new full-length stock.

Scales matter just as much. Connect your weighbridge and floor scales so each weigh event posts straight to inventory. No manual keying. No fat-fingered numbers. No scrap totals worked out on the back of a delivery note.

When a scrap bin goes on the scale and the operator scans it, the system logs the weight, material type, and linked job there and then. Accuracy of **+/-0.5-1.0%** is enough for shop-floor control in most UK plants. Aerospace and nuclear sites need tighter tolerances and United Kingdom Accreditation Service ([UKAS](/hubs/metals-manufacturing-glossary/))-traceable calibration.[\[16\]](https://knarrtek.com/wiptracker-wip-tracking-system/)[\[18\]](https://aws.amazon.com/marketplace/pp/prodview-mpjchfvch5umg)[\[12\]](https://gosmarter.ai/solutions/finance/)

End-of-day batch entry creates the **"system says 5, rack has 0"** mess. Event-driven updates get rid of it.[\[18\]](https://aws.amazon.com/marketplace/pp/prodview-mpjchfvch5umg)[\[17\]](https://www.parsec-corp.com/blog/rfid-for-inventory-management)[\[20\]](https://intelligex.ai/case-study/warehouse-inventory-accuracy-ai-cycle-counting/)

### Use AI to flag stock problems your team usually catches too late

Barcodes, scales, and cameras give you data. GoSmarter, built by Nightingale HQ, uses Artificial Intelligence ([AI](/hubs/metals-manufacturing-glossary/)) to tell you when that data looks wrong. That is the useful part. Not the sales pitch. Not the shiny dashboard. The bit that stops bad stock records from sitting there until someone loses half a day sorting them out.

It helps in three direct ways: pattern recognition, anomaly detection, and data reconciliation.

The system learns normal flow by product family. It sees usual dwell times by location. It tracks expected [scrap rates](https://gosmarter.ai/docs/scrap-calculator/) by machine and material. Then it flags events that do not fit.

> A bundle that appears in Despatch without passing through any intermediate bay is a tagging mistake or a missing scan, not a miracle.

A coil with repeated move records but no linked scans usually means the barcode fell off. A jump in scrap kilograms per tonne processed on one saw needs checking before it turns into a write-off.[\[11\]](https://www.growexx.com/ai-powered-account-reconciliation/inventory-reconciliation-solution/)[\[13\]](https://www.nexusphere.ai/use-cases/inventory-management)[\[21\]](https://www.rfgen.com/industries/manufacturing/)

Keep the AI checks tight. Show supervisors the stock variances that matter, the scrap rates that look off, and the traceability chains that break. If you flood them with noise, they will ignore the lot.

That gives you live events. The next step is putting stock in the right place and getting orders out on time, then linking everything to certs and scrap.

## Put material in the right place and get orders out on time

Fixed slotting falls apart when demand shifts or rush orders land. Fast-moving stock ends up at the back of the warehouse, and the forklift burns extra trips just to reach it. [AI improves put-away](/blog/smart-warehousing/) by using demand, urgency, weight, and route distance instead of fixed rules. Research on AI-based order picking shows travel distance cuts of 20% to 69% against older methods.[\[6\]](https://link.springer.com/article/10.1007/s11846-025-00858-1)[\[7\]](https://www.tandfonline.com/doi/full/10.1080/17517575.2024.2448834) In a metals warehouse, where each extra move may need a forklift, a crane, or two operators, even the low end of that range cuts fuel, labour, and equipment wear.

Dispatch is the other end of the same problem. Picking orders first-come, first-served, with urgent jobs surfacing late, turns the last two hours of a shift into a fire drill. AI sequences picks by due date, carrier cut-off, material readiness, and load plan instead. It groups orders that share the same heat number, grade, or destination, which cuts re-handling, and it flags at-risk orders early if two jobs need the same coil or bundle. Studies on AI-driven order batching and routing report a 27% cut in travel distance through smarter grouping alone.[\[6\]](https://link.springer.com/article/10.1007/s11846-025-00858-1) Machine learning applied to warehouse logistics in metallurgical operations cuts order processing time by 25% to 30%.[\[5\]](https://s-lib.com/en/issues/eiu_2026_05_v1_a19/) That means fewer missed cut-offs and fewer awkward calls explaining why the load still hasn't left.

## Use [GoSmarter](https://gosmarter.ai/) to connect live stock, mill certs, and scrap data

{{< image src="8605749f8fe5d3f83238578ba24726c4.jpg" alt="GoSmarter" >}}

Barcodes, scales, and cameras throw off live events all day. Fine. Raw data on its own does not help much. **GoSmarter, built by Nightingale HQ, turns that feed into something you can use**. It sits on top of your existing [Enterprise Resource Planning (ERP)](/hubs/metals-manufacturing-glossary/) system instead of forcing you into a long, painful replacement project.

### Start with MillCert Reader so heat numbers stop getting lost

**MillCert Reader** reads mill cert PDF files on its own and links each heat number to stock. It pulls out the heat number, grade, chemical composition, mechanical properties, section type, batch weight, and test dates in seconds. No retyping. No one squinting at a PDF and keying numbers into a tired old screen.[\[4\]](https://gosmarter.ai/hubs/mill-cert-automation/)[\[3\]](https://gosmarter.ai/llms-full.txt) It trains on certificate templates from hundreds of mills, so it standardises data from different suppliers and layouts into one structured record automatically, whichever mill sent the cert.

Each record then links straight to the matching stock item through the heat number. Offcuts stay tied to the original heat, which matters when someone asks where that bit came from three weeks later.[\[4\]](https://gosmarter.ai/hubs/mill-cert-automation/)[\[22\]](https://gosmarter.ai/docs/what-is-a-mill-test-certificate/) The module supports **[EN 10204](https://en.wikipedia.org/wiki/Mill_test_report)** types 2.1, 2.2, 3.1, and 3.2, plus [ASTM](https://www.astm.org/) and BS specifications. That keeps the audit trail in line with UK metals supply chain compliance needs.[\[24\]](https://gosmarter.ai/solutions/compliance/)[\[12\]](https://gosmarter.ai/solutions/finance/)

Quality teams can download certificates by page or by heat code, so they can pull order packs together without the usual faff. Automating this extraction saves production teams over **120 hours a year** of manual data entry, roughly three full working weeks.[\[1\]](https://gosmarter.ai/hubs/gosmarter-for-metals-operations/)[\[23\]](https://gosmarter.ai/products/millcert-reader/) MillCert Reader runs **£295 a month**, and high-volume teams typically recover that cost inside the first month at standard loaded labour rates.

Once you link the certs, you can use those same live events to track stock movement, offcuts, and scrap.

### Use Metals Manager to keep live stock, orders, and offcuts in a single live view

With certs tied to heat numbers, Metals Manager gives you one [live view of stock, offcuts, and scrap](https://gosmarter.ai/docs/what-is-metals-inventory-management/). It syncs master data such as item codes, customers, suppliers, and orders. Then it adds the bits most [Enterprise Resource Planning (ERP)](/hubs/metals-manufacturing-glossary/) systems handle badly in metals: a coil is not just a Stock Keeping Unit (SKU), it is a piece of material with a grade, heat number, dimensions, finish, and handling limits that matter on the floor. Each piece gets tracked by size, grade, length, section type, and exact location. Offcuts and scrap get logged in both tonnes and £.

Metals Manager adds three things:

-   **Live location** by bay and rack, updated as events happen
-   **Offcut records** for every cut, with size, grade, location, and £ value
-   **Scrap logging** by job, heat, and shift, valued in tonnes and £

The same heat-number spine that links certs to stock also feeds the Cutting Optimiser and the wider Metals Manager view. It's one record doing the work across every tool, not a separate database for each one.

[Midland Steel](https://midlandsteelreinforcement.com/), a UK-based rebar and long products manufacturer, ran a two-week production trial using [algorithm-optimised cutting plans](/products/cutting-optimiser/) across 193 jobs and 734 tonnes of steel. Scrap dropped from 5% to below 2.5%. For a site cutting 100 tonnes a week, that saves about **£5,700 a month**, or more than **£68,000 a year**.[\[2\]](/hubs/cutting-optimiser/)[\[3\]](https://gosmarter.ai/llms-full.txt) Read the [full Midland Steel case study](/casestudies/midland-steel/) for the trial methodology. You only see numbers like that when you log scrap by heat and job. If you wait until month end and guess, the money has already gone.

## Roll this out without disrupting a working warehouse

Artificial Intelligence (AI) only works if you feed it data you can trust. In metals warehouses, pilots usually fall over for a dull reason. **Bad master data. Patchy records. Missing links.** Not the model.[\[12\]](https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/clearing-data-quality-roadblocks-unlocking-ai-in-manufacturing)[\[17\]](https://www.mckinsey.com/capabilities/mckinsey-technology/our-insights/ai-data-readiness-the-key-to-scaling-impact)

That mess leads to bad suggestions. Then the team stops trusting the system. Fair enough. Keep the core system as the source of truth. Add an AI layer that reads and writes through Application Programming Interfaces ([APIs](/hubs/metals-manufacturing-glossary/)) or file interfaces instead of ripping anything out.[\[19\]](https://supplyaihub.com/implementation-guides/pilot-to-production-sequencing-warehouse-ai-implementation)[\[20\]](https://supplyaihub.com/comparisons/pilot-to-production-sequencing-warehouse-ai-implementation) GoSmarter, built by Nightingale HQ, fits that model: it automates document-heavy and planning-heavy work without forcing a full rip-and-replace, and you can tune it or switch it off without touching the core system.

The rollout sequence that works on the floor has three stages. First comes read-only mode, where AI looks at the data and gives recommendations but writes nothing back. Then come assisted workflows, where a planner or supervisor approves the action. After that, you can move to closed-loop automation for low-risk, high-volume tasks once the team trusts the output.[\[19\]](https://supplyaihub.com/implementation-guides/pilot-to-production-sequencing-warehouse-ai-implementation)[\[22\]](https://inferensys.com/integration/warehouse-management-platforms/custom-ai-development-for-warehouse-management-platforms) Skip straight to full automation and you usually lose the floor team. Once that trust goes, no dashboard is going to win it back.

### Pilot one product line before rolling out across the factory

Start small. That is the sane way to do it. Pick one product line where traceability and scrap losses hurt. Rebar, plate, or structural sections are common choices in UK plants. Then pick one physical area, such as a rebar yard or plate warehouse.

Connect GoSmarter to your [Enterprise Resource Planning (ERP)](/hubs/metals-manufacturing-glossary/) system by CSV or [Representational State Transfer Application Programming Interface (REST API)](/hubs/metals-manufacturing-glossary/). Authentication runs through OAuth or your existing Microsoft Entra single sign-on. Data stays hosted in UK Azure. GoSmarter's own models train on your certificate data to improve extraction. Your data is never shared to train Microsoft's underlying models. Set up the metals-specific fields your ERP probably does not hold properly:

-   **Section type** such as UB, UC, RHS, plate, or rebar
-   **Weight per metre**
-   Nominal and actual dimensions
-   Heat number

Set default units to metric: mm, m, and tonnes. Set currency to **GBP (£)**. Make sure both systems use **dd/mm/yyyy**. If they do not, you will spend your time sorting out reconciliation errors instead of fixing the process.

The pilot usually runs for **four to eight weeks**, and most teams see clean, linked certificate data within the first week of running MillCert Reader. During that time, run MillCert Reader on a fair batch of incoming mill certs for that product line. Check the extracted fields against your current manual entries. Track a few simple key performance indicators:

-   Time to find a heat-linked piece
-   Manual data entry minutes saved per shift
-   Scrap value logged in tonnes and £ against earlier estimates

At the end, review the results with production, quality, and planning teams before you decide how far to roll it out across the factory.

Pilot one product line and one live stock area first.

## Run a GoSmarter pilot on your next batch of mill certs and one live stock area

With scope and key performance indicators ([KPIs](/hubs/metals-manufacturing-glossary/)) fixed, run the pilot on **one live stock area** and the [next batch of mill certs](https://gosmarter.ai/blog/how-to-automate-mill-certificate-management-in-5-steps/). GoSmarter, built by Nightingale HQ, works best when you don't try to boil the ocean on day one.

Pick **one product family** and **one live stock area** before you switch anything on. Keep the boundary small. It should fit on one page. If it doesn't, you've made it too big.

Set ownership now. If you don't, the pilot turns into **end-of-shift admin**. That's where good projects go to die.

| Pilot role | Primary responsibility | Key tool |
| --- | --- | --- |
| Pilot Lead | Scope and sign-off; chairs weekly reviews | GoSmarter Admin |
| Process Engineer | Validate cert-to-stock links and scrap data | MillCert Reader |
| Stores Supervisor | Log receipts, moves, and issues as live events | Metals Manager |
| Information Technology (IT) / Data Specialist | Fix integration and data-format alignment | Comma-Separated Values (CSV) exports |
| Quality Engineer | Check AI-extracted fields against EN and BS specs | MillCert Reader |

> Collect your baseline numbers before switching anything on. Without them, you cannot prove the pilot worked.

In the week before launch, capture three things:

-   How many mill certificates arrive as Portable Document Format (PDF) files that need manual retyping
-   How accurate stock records are in the chosen area
-   How long it takes to find a heat-linked piece for an urgent job

Set clear targets for a single stock zone. Cut manual cert-handling time from roughly **25-30 minutes per bundle** to **under five minutes**. Push stock accuracy from the **mid-80s** to **97% or above**. Hit **95-99% [traceability completeness](https://gosmarter.ai/docs/what-is-steel-traceability/)** for every item in the pilot area.

Run weekly check-ins with the full group. Keep them short: current KPI values, one win, one obstacle. No one needs a one-hour meeting about a stock bay.

At the end of **30 days**, compare the numbers to your baseline. If **three or four metrics** improve, move to the next bay or product family.

Use the 30-day results to decide the rollout path in the conclusion.

## Conclusion: Build real-time inventory tracking one step at a time

Real-time [Artificial Intelligence (AI)](/hubs/metals-manufacturing-glossary/) inventory tracking follows a simple order. Clean your master data first. Record every movement when it happens. Then let AI flag the odd stuff and put material where it needs to be. That is how you turn a noisy yard into something your team can trust.

Receipts, moves, cuts, issues, and scrap need to become live events at the point of work. You keep heat-number traceability at every step. Then AI scans those events and flags the things your team usually finds too late: **mismatched stock, [missing heat numbers](/blog/ai-mill-test-report-traceability/), and odd scrap patterns**.

The pilot in this guide keeps risk low. Start with one product line, one stock area, and one batch of mill certificates. GoSmarter, built by Nightingale HQ, links heat numbers to live stock through MillCert Reader. Metals Manager keeps orders and offcuts in one view. **Live certs. Live stock. Fewer manual checks.** If the pilot shows the numbers stack up, scale by bay and by product line.

Use the pilot result to decide if you should expand. Check if stock variance has fallen, scrap is lower, and reconciliation time is down. Then scale bit by bit until every tonne and metre of stock is visible in real time, and your team trusts what the system shows.

## FAQs

{{< faq question="What data needs cleaning before using AI for live inventory?" >}}
Before you use GoSmarter, built by Nightingale HQ, for live inventory, **sort out your data first**. Clean your master and transaction data. Tidy your parts and material master lists. Remove duplicates. Standardise supplier and stock records.

If you skip this bit, the software just chews through the same old mess faster. That's not progress. That's just **quicker confusion**.

You also need incoming materials set up with consistent, structured mill test or certificate fields, plus **accurate density factors** so your weight and dimension checks don't go sideways. Flag missing or suspect values. Check them against expected ranges. Then link the extracted data to the right inventory items in real time. Keep **remnant and offcut tracking** that carries each offcut's heat number and full history, or it turns into mystery metal that ends up ignored, misused, or scrapped.

That matters because live inventory falls apart when one cert says "S275JR", another says "S275 JR", and a third has half the fields missing. The metal's fine. The paperwork is the problem.
{{< /faq >}}

{{< faq question="When should we use RFID instead of barcodes?" >}}
Use Radio Frequency Identification ([RFID](/hubs/metals-manufacturing-glossary/)) when you need **hands-free tracking** and you cannot rely on direct line-of-sight scanning. It lets you monitor stock and movement in the background, without asking someone to stop and scan every item.

Barcodes and mobile devices still make sense for manual, point-of-activity capture. Think goods-in or production issues. They work well when someone already stands at the job and can scan there and then.

RFID earns its keep in **high-throughput** environments. If your team handles bundles or coils all day, scanning each one soon turns into a bottleneck. That is where RFID cuts the faff and keeps material moving.
{{< /faq >}}

{{< faq question="How long should a real-time inventory pilot run?" >}}
Run the pilot for **30 days**. Keep the scope tight. Start with one area, like **goods receivable** or **stock intake**.

Track a few plain measures:

-   **Cert admin time**
-   **Rework incidents**
-   **Release speed**

That gives you enough time to set a baseline, see what changed, and decide what to automate next. It also shows you what to fix before you roll it out more widely.
{{< /faq >}}

{{< faq question="Can AI improve traceability without replacing your ERP or WMS?" >}}
Yes. GoSmarter, built by Nightingale HQ, sits on top of your existing Enterprise Resource Planning ([ERP](/hubs/metals-manufacturing-glossary/)) or Warehouse Management System ([WMS](/hubs/metals-manufacturing-glossary/)). You improve traceability without ripping out the system that still runs invoicing and procurement.

It uses Comma-Separated Values ([CSV](/hubs/metals-manufacturing-glossary/)) imports and exports or a Representational State Transfer Application Programming Interface ([REST API](/hubs/metals-manufacturing-glossary/)), authenticated through an API key or Microsoft Entra single sign-on (SSO). GoSmarter digitises certificates, then links heat numbers and test reports to stock records. Your data stays hosted in UK Azure. GoSmarter's own models do train on your certificate data to keep extraction accurate. Microsoft's underlying models are excluded from that training. Your core system keeps doing the day job. GoSmarter deals with the paperwork mess.
{{< /faq >}}