The Best of the Future Digital Twin & AI Conference — and How We’re Applying it in Reality Twin  

At the conference, much of the discussion focused on traditional “digital-only” twins and their use alongside new AI tools for industry. In contrast, SolidComp’s highly visual Reality Twin places its emphasis on shared, reality-based visual understanding rather than pure simulation. Even so, we were inspired by the innovations presented and left with ideas on how these perspectives could be developed further within Reality Twin. 

1) People First, or Your AI Won’t Last

Across the panel discussions the strongest signal was human-centric change: build AI with operators, not just for them. Speakers emphasized open collaboration, humility, and boosting AI literacy across the workforce as the starting point for any durable transformation. That’s because adoption is a change-management exercise before it’s a technological one.  

Another shared theme in the discussions was the need to design to reduce cognitive load. As one panelist put it, the point isn’t “AI for AI’s sake,” but taking friction out of field work. And that only happens when the data foundation is sound and modular for plug-and-play AI.  

Our strategy on applying this to Reality Twin is to involve field crews early via guided walkthroughs in the live 3D twin, capture their work flows (not just feature requests), and tune the workspace around the few decisions that matter most on a shift. 

2) Value > Vanity: Follow the Cash, Follow the Work

Shell’s digital twin lead described an asset-led, centrally supported model and a simple rule: “follow the cash, follow the value.” The team translates asset goals into a product roadmap, then scales capabilities that show measurable impact.  

Panelists also pressure tested the phrase “business critical” for digital twins. If you switch off the twin, the plant can still physically run, but teams would end up reverting to paper and Excel and lose competitiveness over time. This way, the real inflection is not “Would it run without?” but “Would it run without well enough to win?”  

An interesting nuance to notice was that some of the panelists were counting micro twins (unit/valve level) as “business critical.” Enterprise level twins remain fewer, but the competitive edge appears when you infuse the twin with AI and embed it in daily decisions.  

Our strategy on applying this to Reality Twin is to sequence deployments by “cash levers” (e.g., turnaround preparation, remote inspections, permit work prep) and treat each as a product increment with adoption and time-to-action metrics — and then scale the winners.  

3) Digital Twins Are the Safest Lab for AI

The most practical way to hedge the risks from AI was repeated: Use the twin as a sandbox. Teams described running deep reinforcement learning experiments in a safe digital environment and blending historical data with simulation to learn without risking assets. Reliability, safety, lower emissions and profitability are used as rewards for the learning loop.  

This pairs neatly with a broader industry pattern:  

LLMs + an industrial knowledge graph (your contextualized operations data) = higher trust and lower hallucination Gen AI.  

This formula lets you answer complex plant questions deterministically and attach the “why” to every answer.  

Our strategy on applying this to Reality Twin is to start by grounding AI in evergreen site context; up-to-date scans, linked P&IDs, tags, documents, so assistants “see” the same plant your crews do. We prioritise knowledge assistance, document navigation and action suggestions over opaque predictions.  

4) Processes Before Platforms: “Kill Asset Management (As We Know It)”

Another rising theme in the conference was a classic: Don’t pave cow paths. If you want the 10× step change, redesign workflows (zerobased if needed) and use agents to simulate decision making, not just chase dashboards. It’s the only route to more free cash flow at lower cost.  

Our strategy on applying this to Reality Twin is to codesign “tobe” playbooks with operations (hazard checks, access planning, isolation, permit preparation) and use the twin to emulate the work, not just visualise the site.  

About That “95% of AI Pilots Fail” Headline

95% of AI pilots fail? If you follow the discussion about using AI in industry, then yes, you’ve likely seen this claim. But it’s a misread of a narrow slice of projects: company-built, task specific GenAI. It does not describe the broader class of general purpose LLMs, which are being adopted widely and successfully.  Research notes high adoption for LLM pilots and even a “shadow AI economy” of worker-led tools. This is evidence that value appears when people can actually use the tech in their workflow. 

What matters for industry is context and control: bind LLMs to your operations data with a knowledge graph, deliver explainable answers, and keep data access governed. 

Digital twin studies lines up with this reality: 96% of executives say digital twins deliver value, with tracked ROIs most often >10% and operational benefits that exceed expectations (e.g., collaboration and proactive problem solving). In other words, when AI is attached to real work with real context, it pays.  

Where Reality Twin Fits

• SolidComp’s Reality Twin is built for this people and data first approach: 
• Unlimited highfidelity 3D streaming (browser & VR) so teams can align on a single, accurate view of the site. 
• Reality Twin’s asset management enables all data to move with the asset, and not with people working with it. 
• Open integrations (EAM/ERP/IoT/DMS) to ground assistants and workflows in your current systems. 
• “As built” editing & connected P&IDs so the twin stays evergreen and navigable. 
• AI modules (autotagging, knowledge assistance) to speed up the boring parts of work.  

On our recent deployments we’ve seen outcomes like 70% fewer site visits for planning and coordination at a pulp mill, because everyone can “walk the site” together in the digital twin and annotate the exact location of work before anyone travels.  

And the market context is on our side: the Nordics’ heavy industry footprint (pulp & paper, mining) is digitizing fast; scan-to-twin workflows are now 5–10 times faster and approximately 50% cheaper than traditional remodeling, which is why remote plant access is becoming standard across European operators.   

A Practical, People-in-the-loop Rollout that You Can Start Now

1. Pick one high value corridor (e.g., a mill line, a process unit, a substation). 
2. Capture + connect: scan, link P&IDs and the 10–20 documents crews actually use. 
3. Codesign the daily loop: build the operator checklist in the twin. 
4. Add assistive AI where it’s safe: knowledge search, procedure steps, permit prep. 
5. Measure adoption and impact, then expand to the next area. 

If you do nothing else, do this: put operators to you the twin early and anchor AI to their daily decisions. The rest; scale, savings, and competitive advantage, will follow. 

Subscribe to SolidComp’s Newsletter

Receive a newsletter from SolidComp about once a month. In it, we share the latest updates from the industry, SolidComp, and the SolidComp Digital Twin.

Subscribe newsletter