Pilot graveyards are the real maturity baseline
For many smart factory programs, the honest digital maturity baseline is a graveyard of pilots without operating owners. Score what still runs after the innovation budget closes, not the roadmap slide.
What a pilot graveyard is in plant language
A pilot graveyard is a set of proofs of concept that once passed a demo gate and then lost a named owner, a production budget path, or a decision they were supposed to change. The stack still exists: cameras still stream, vibration models still score, energy tiles still refresh, and a digital twin hall still opens in a browser. What is missing is everyday use that survives staff moves and year-two OPEX.
Industry writing often calls this pilot purgatory: activity without scaled impact. Maturity and readiness models (Acatech-style indexes, RAMI-framed assessments, and published reviews of Industry 4.0 models) frequently score technology and IT dimensions well. Those scores can outrun process and ops reality when the plant still runs on the systems that never absorbed the pilot. The same gap shows up in the partial twin layers covered in most plants already own some version of a digital twin. Maturity scores also inflate when assessors credit new pilots and ignore the MES, SCADA, QMS, and CMMS stacks already running on most floors, the starting set covered in the systems already running on most floors.
Four PoC failure modes that keep repeating
Research on industrial twin and manufacturing AI pilots keeps returning to organizational failure modes more than missing algorithms. The first is no data owner after go-live. The integrator scope ends at acceptance. The champion moves to another site. Twelve to twenty-four months later the model still shows assets that were already replaced, stale labels, or last-campaign readings, and users stop opening it. Data governance here means a named person whose job includes keeping each data layer current in year two, not a steering committee slide.
The second is as-designed geometry or process truth standing in for as-built reality. A twin or layout shell built from drawings and design intent disagrees with the plant after decades of tie-ins, relocated valves, moved cells, and redlines that never returned to the drawing office. Field users lose trust on first contact. A BIM shell alone is not an operational twin of the shift.
The third is a pilot funded from innovation or capital with no signed path into the production operating budget. The PoC " succeeds" and quietly ends because nobody owns licences, connector care, label refresh, or support under OPEX. The fourth is no named decision the pilot changes. Visibility is not a decision. Without a decider, a date, and an acceptance criterion, scope grows and no decision ever gets forced to a close. Those four modes show up across domains. They are not logistics-only problems.
The same modes in quality, maintenance, energy, and twin demos
Vision quality control is a common graveyard entry. A computer vision station catches defects on a curated line under good light. Production brings drift, edge cases, and operators who override noisy alerts. If verdicts never land in the QMS as owned dispositions, the camera is a slideshow. The failure is missing operating ownership and a named release-or-hold decision, not missing pixels.
Predictive maintenance PoCs fail the same way when alerts never open trusted work in the CMMS. A model trained on clean historian slices meets brownfield sensor drift, uneven tagging, and technicians who ignore false positives. Without a maintenance owner, a false-positive budget, and a production support line, the pilot becomes another dashboard. Energy dashboards fail when utilities tiles have no owner who can change setpoints, schedules, or idle policy, and when the board never named which kilowatt-hour (or BTU) decision the tile was supposed to change. Multi-site programs fail the same way when one plant's successful PoC never gets a named owner, budget path, and decision rule at the next site, so the roadmap marks the enterprise green while each hall still runs a private demo.
Twin demos fail when a polished 3D shell or slow map proves " we have a twin" without live sync, a stored place-and-time record, or a shift decision someone will change. Live operations, a spatially oriented dataset, and correlation and causality analytics are equal reasons to keep an operational digital twin alive. A demo that only tours geometry has not finished any of those jobs. Internal logistics can be one equal example beside production monitoring, quality, maintenance, and energy. It should not be the only spine of the maturity story.
Why roadmap slides overstate maturity
Smart manufacturing roadmaps often mark a domain green after a successful PoC review. That mark confuses three different states: the technology can work in controlled conditions, the plant can operate it under exceptions, and the enterprise can repeat the pattern on the next line or site. Industry practice and published pilot-to-production writing treat those as separate gates. Crossing the first gate alone is not maturity.
IT-heavy scores rise when connectors, cloud tenancy, and model accuracy look good. Ops reality stays flat when MES, SCADA, QMS, and CMMS still do not share a governed live path, or when nobody owns keeping those connections current after the vendor leaves. Brownfield interoperability (proprietary protocols, gateways, OPC UA or MQTT bridges) is real work. It does not replace naming the operating owner. That gap begins at the starting point is still a collection of disconnected systems. Later foresight still waits on the stored history described in predictive digital twins wait on stored history.
How teams set an honest baseline
1. Inventory pilots by operating owner, not by slide title - For each vision QC, predictive maintenance, energy, twin, or logistics PoC, name who still owns data refresh after go-live. No name means the pilot counts as graveyard stock for maturity purposes.
2. Separate as-designed artifacts from as-built trust - CAD, BIM, and design twins are useful. They are not proof the live floor matches the model. Flag unverified geometry and process truth explicitly.
3. Require a signed production operating-budget line before the next PoC starts - Licences, connector care, label and model refresh, and support sit under OPEX or an equivalent operating commitment. Innovation CapEx alone is how successful pilots die quietly.
4. Name the decision, the decider, and the date - Release vs hold, open a CMMS work order, change an energy setpoint, reroute a starved loop, or cancel a trip from the twin. Visibility without a decision is not a pass criterion. Bound the learn cost the way what a two-to-four-week hall pilot actually costs to learn describes.
5. Count equal twin jobs when a twin pilot is in the list - Live reaction on a live hall map, the spatially oriented dataset under that view, and later correlation and causality work are equal. A tour that stores nothing and changes no decision is still a graveyard candidate.
What changes once the baseline is honest
An honest baseline shrinks the roadmap and raises the bar for the next buy. New work starts only where an owner, a production budget path, and a named decision already exist on paper. That discipline is how a smart factory program stops collecting demos and starts keeping systems that still run under real owners and year-two budgets.
twinzo belongs after that honesty gate, not before it. It sits as an operational layer for production monitoring and internal logistics optimization while building the place-and-time record that correlation work needs later. Capability depth is on the features overview. Get in touch if you want to map which pilots still have operating owners and which belong in the maturity baseline as graveyard stock.