TL;DR — Key Takeaways
- Between 2019 and 2026 our annual output index rose about 30% while headcount fell from about 200 to about 120 people, so output per employee more than doubled (index 100 to 217).
- The transformation ran in four phases—machine layer, data layer, process layer, then people and AI layer—because attempting all four at once destroys shop-floor trust.
- Two of our early projects failed outright, and those failures taught us more than the successes: a dashboard nobody used and a sensor program we cancelled after 14 months.
- Headcount reduction came from four years of natural attrition plus more than 30 people moving into new digital roles; training hours per employee rose from about 12 to about 68 per year.
- Three metrics barely moved—small-batch changeover improved only about 6%—and knowing what automation cannot fix is what makes a supplier's claims verifiable.
Between 2019 and 2026, this factory raised its annual output index by about 30% while the payroll fell from roughly 200 people to about 120—because the transformation was never mainly about machines. It was about moving decisions out of human memory and into data that machines, planners and buyers can all see. Output per employee more than doubled. Order lead time fell 15%. The defective product rate dropped 17.6%. Slab utilization improved 22.58%. Those are the numbers, and I am proud of them.
But you have read the glossy version of this story before—from us, from every competitor, from every automation vendor. What I want to write instead is the version I would want if I were a procurement director trying to judge whether a stone supplier is genuinely digital or merely newly painted. That version includes the two projects we killed, the workforce conversation, and the three things four years of automation did not fix. Because a supplier who can tell you what they failed at is far more useful to you than one who cannot.
What Exactly Changed Between 2019 and 2026?
The honest answer is that a handful of operational metrics moved a lot, several moved modestly, and a few did not move at all. The real headline is not output growth; it is that output per employee more than doubled, from index 100 in 2019 to index 217 in 2026. Here is the full record, indexed where internal baselines are sensitive:
| Metric | 2019 baseline | 2026 current | Change |
|---|---|---|---|
| Annual output index | 100 | 130 | +30% |
| Headcount | ~200 | ~120 | −40% |
| Output per employee (derived) | 100 | 217 | +117% |
| Order lead time | Baseline | Baseline −15% | −15% |
| Defective product rate | Baseline | Baseline −17.6% | −17.6% |
| Material utilization | Baseline | Baseline +22.58% | +22.58% |
| Training hours per employee per year | ~12 h | ~68 h | +467% |
| Current capacity reference | — | 40,000 m²/month (360,000 m²/year) | — |
Table 1: Transformation record, Ruifengyuan Stone 2019–2026. Baseline indexed where internal data is commercially sensitive. Last verified: 2026-09-21.
Read that table the way a buyer should read it, and one line stands out above all others: the headcount line. Because output rose while headcount fell, the productivity gain per person exceeded the output gain by a factor of nearly four—and that is the number that decides whether a supplier can hold a price and a delivery date during peak season.
The Four-Phase Timeline: What We Installed, and When
We ran the transformation in four deliberate phases over seven years, and the sequencing mattered more than any single machine. Smart manufacturing in a stone factory means machine data, order data and quality data flow through one connected system so decisions come from measurement instead of memory. Getting there took four layers, built in this order:
Phase 1 (2019–2020): The machine layer
We replaced manual cutting and hand measurement with six intelligent bridge saws and two Italian five-axis machining centers, and we made one decision I would repeat today: we wrote acceptance tests into the purchase contracts, using ISO 10791 test conditions for machining centres as the reference for accuracy verification. Buying a machine without an acceptance test is buying a promise; the ISO 10791 series exists precisely to convert the vendor's brochure numbers into measured ones. Phase 1 taught us a hard lesson, though: throughput barely moved for the first eight months, because better machines fed by the same manual scheduling simply sat idle in different patterns.
Phase 2 (2021): The data layer
Then we connected things. Work orders moved into ERP with MES tracking on the floor, slabs received digital identities, and machine output started flowing continuously rather than being recorded by hand at shift end. This is the layer the NIST Digital Thread for Manufacturing program describes as a two-way information thread between engineering, manufacturing and quality—and that two-way word is the whole point, because a factory that only reports upward gains dashboards, not control.
Phase 3 (2022): The process layer
With data flowing, we could finally fix processes instead of blaming people. Vein matching and slab layout moved into CAD with scanned digital twins, and the layout area became a dry-lay verification station rather than a storage yard. Quality inspection gained an AI visual layer developed with the Chinese Academy of Sciences' Institute of Automation. Because the pattern and the quality standard were now defined digitally, the same defect argument stopped repeating on every project—and the site disputes we used to absorb became rare enough to count.
Phase 4 (2023–2026): The people and AI layer
This is the phase nobody budgets for and the one that determines everything. We retrained operators into programmers, inspectors into adjudicators, and foremen into data readers. I will come back to the workforce question below, because it deserves its own section rather than a bullet point.
| Phase | Years | What we installed or changed | What it delivered |
|---|---|---|---|
| 1 — Machine layer | 2019–2020 | 6 intelligent bridge saws, 2 Italian five-axis centers, ISO 10791-style acceptance testing | Machine accuracy verified in writing, but throughput flat for the first 8 months |
| 2 — Data layer | 2021 | ERP–MES integration, digital slab identities, continuous machine output collection | Scheduling decisions moved from shift-end memory to live status |
| 3 — Process layer | 2022 | CAD vein matching on scanned twins, dry-lay verification station, AI visual inspection with CAS | Utilization +22.58%, defective rate −17.6%, site disputes reduced |
| 4 — People & AI layer | 2023–2026 | Role migration into digital work, structured annual training, AI-assisted adjudication | Output index +30% at 120 employees; lead time −15% |
Table 2: Four-phase summary of the transformation. Source: internal project records. Last verified: 2026-09-21.
Why Do Most Smart Manufacturing Projects Stall in the Pilot Phase?
They stall for organizational reasons, not technical ones, and the industry data backs that up. According to the World Economic Forum's Global Lighthouse Network, more than 70% of companies investing in advanced analytics, AI or digital solutions fail to move beyond the pilot phase. That network now spans 238 factories worldwide—and those are the survivors. I can tell you from the inside why the other majority drown.
The dashboard nobody used. In 2021 we built a beautiful real-time production board that displayed machine states, utilization and shift output on a 65-inch screen in the hall. Within six weeks the foremen had stopped looking at it. The reason was embarrassingly simple: it measured what management wanted to know, not what foremen needed to decide. Nobody on the floor was measured on machine utilization; they were measured on orders shipped complete. We rebuilt it around order readiness instead, and adoption followed in days. If a digital tool does not answer a question its user already asks, it is decoration.
The sensor program we cancelled after 14 months. We instrumented vibration and current draw on older saws, intending predictive maintenance. The models kept flagging false positives because the biggest variable was not tool wear—it was which material batch was being cut that afternoon. We killed the project. What survived is far less glamorous and far more useful: a simple maintenance schedule derived from blade-hours per stone type. Not every smart tool deserves to live; some deserve to die early.
Both failures pointed at the same rule, which I now apply to every investment proposal on my desk. Because a transformation project only survives if the person using it is better off the same week it launches, any pilot whose benefit arrives six months later is already dead—it just has not been told yet.
What Happened to the 80 People Who Left the Payroll?
This is the question clients ask me privately, and it deserves a direct answer. Our headcount fell from about 200 to about 120 between 2019 and 2026, and that reduction happened through four years of natural attrition—not a redundancy program; more than 30 employees moved into new digital roles instead of leaving.
Three role families absorbed the migration. Former saw operators became CNC programmers, learning to read cutting files rather than chalk lines. Measuring-and-marking hands became layout engineers working in CAD on scanned slab twins. Inspectors became adjudicators, reviewing the borderline cases the AI vision line flags rather than scanning every square meter with their eyes. Training hours per employee rose from roughly 12 to 68 per year, and I will say plainly: that training bill was the largest single line item in our four-year transformation budget—larger than any machine we bought.
Was it friction-free? No. We lost people who did not want to sit at a screen, and I respect that honesty more than I respect a forced retraining that ends in quiet failure. If you are a distributor or developer reading this, there is a practical takeaway hidden in the labor story: a factory that has retrained its workforce can absorb a schedule shock, while a factory that only bought machines can only absorb it with overtime—which is where the price increases and the missed dates come from.
The Metrics That Moved—and the Three That Didn't
Here is the section most supplier websites skip entirely. Four years of automation left three areas essentially untouched: small-batch changeover, hand carving, and fitting done by touch. Our changeover time between small batches improved only about 6%—measured, not estimated—and that number embarrasses me slightly less each year now that I understand why.
| Area | Change 2019–2026 | Why automation did not deliver |
|---|---|---|
| Small-batch changeover | Approximately −6% | Changeover cost is dominated by material handling and tooling change, not by machine speed |
| Hand carving and sculptural work | Effectively unchanged | The work is tactile and judgment-based; no sensor we tested replaces an experienced eye |
| Complex fitting at installation | Effectively unchanged | Tolerances depend on site conditions and touch; digital twins help planning, not execution |
| Standard-format cutting and polishing | Large improvement | Repetitive, measurable, machine-readable work—exactly what automation is for |
| Quality inspection coverage | From sampling to 100% of pieces | Machine vision does not fatigue across an 8-hour shift |
Table 3: Where the four-year transformation did and did not pay. Source: internal MES and maintenance records. Last verified: 2026-09-21.
When a supplier tells you their factory is "fully automated," ask what did not improve. A supplier with a real transformation record has a specific, slightly uncomfortable answer ready. A supplier without one has only adjectives.
Interactive: Factory Digital Maturity Scorecard
Six questions, 24 points. Score a factory you are evaluating—or your own operation—by selecting the description that matches reality, not ambition. The result updates as you choose.
Reactive operation: decisions live in individuals and paper. Start with slab identity and per-piece QC records—the two foundations every later phase depends on. (0 of 6 questions answered.)
If a Supplier Claims Smart Manufacturing, How Do You Verify It?
Ask for artifacts rather than descriptions, and ask for them in this order. A genuine transformation leaves documents behind; a marketing claim leaves only pictures of machines.
- Request a per-piece QC report from a delivered order, with timestamps, measurement overlays and a stated defect taxonomy. Real records cluster; typed summaries repeat.
- Request one machine output record covering a specific production week, then compare the reported capacity with the schedule you were quoted.
- Request a traceability sample that follows a single order from slab ID through cutting and crating—the digital thread has to survive past the machine, or it is not a thread.
- Ask which processes did not improve. A supplier with a real program names changeover times, carving, or fitting without prompting.
- Ask for a client who permitted a factory audit in the last 12 months, and ask that client what the dry-lay and acceptance process felt like.
- Ask what happened to the workforce. It sounds like an ESG question; it is actually a capacity question. Retrained teams absorb schedule shocks, and unre-trained ones pass them to you.
None of this requires you to become a manufacturing engineer. It requires only that you insist on evidence that survives leaving the showroom. If your supplier cannot produce it, the right conclusion is not that they are dishonest—it is that their capacity planning is still running on memory, and memory fails exactly when your deadline is tightest.
Frequently Asked Questions
What is smart manufacturing in a stone factory?
Smart manufacturing in a stone factory means machine data, order data and quality data flow through one connected system so that decisions are made from measurement instead of memory. In practice it shows up as scanned slab identities, MES-tracked work orders, machine output collected continuously, and per-piece QC records—not as a room full of robots.
How long does a stone factory digital transformation take?
Did automation replace workers at your factory?
Why do most smart manufacturing projects fail?
How can a buyer verify a supplier's smart manufacturing claims?
Which parts of stone production still resist automation?
Visit the Factory the Numbers Came From
Seven years of transformation produce a lot of documentation—and I would rather you audit it than take my word for it. Send us an elevation drawing or a product list, and we will return a production plan with realistic lead times, the quality records we can commit to in writing, and an open invitation to inspect the line yourself.
Request a Production & QC Capability Review →
Or explore the product catalog and our large-format marble slab range for hospitality and residential projects.
Related reading: ERP + MES: Order-to-Delivery Digitization in a Stone Factory · CAD Pattern Tracking & Vein Matching · AI Visual Inspection in Stone Quality Control · Intelligent Bridge Cutting & Slab Utilization
Post time: Sep-21-2026