\u201cDigital twin\u201d has become one of the most misused terms in industry.<\/p>\n\n\n\n
It gets attached to 3D models, dashboards, monitoring platforms, BIM files and asset registers \u2014 often by people who genuinely believe that\u2019s what the term means. The result is that two people can sit in the same meeting, both say \u201cdigital twin,\u201d and be describing completely different things with a decimal point between their price tags.<\/p>\n\n\n\n
Most of that confusion clears up the moment you realise a digital twin isn\u2019t one thing you either have or don\u2019t have. It\u2019s a ladder with five rungs.<\/strong> Each rung does something the one below it cannot, and each costs meaningfully more to reach.<\/p>\n\n\n\n Here\u2019s the whole model in plain English, with no jargon and no vendor spin.<\/p>\n\n\n\n A 3D model with live gauges attached to it. It looks superb in a demo. Data streams in, dials move, and everyone in the room nods.<\/p>\n\n\n\n But it only displays<\/em>. It doesn\u2019t interpret anything, it doesn\u2019t know what \u201cnormal\u201d looks like for that asset, and it can\u2019t tell you anything you couldn\u2019t work out by walking the plant with a clipboard. This is a dashboard with a 3D background, and a surprising amount of what gets sold as a digital twin lives here.<\/p>\n\n\n\n It isn\u2019t on the ladder at all. That matters, because a great many organisations think they\u2019re at Level 3 when they\u2019re actually at Level 0.<\/p>\n\n\n\n The first real rung. A live mirror of the physical asset: what its temperature is, how fast it\u2019s running, how much it\u2019s vibrating, right now.<\/p>\n\n\n\n The difference between this and Level 0 is context. A descriptive twin knows the asset\u2019s structure and its normal operating envelope, so \u201c42\u00b0C\u201d means something \u2014 it\u2019s not just a number on a screen.<\/p>\n\n\n\n What it answers:<\/strong> What is happening?<\/p>\n\n\n\n This is genuinely useful and it is where most working twins sit today. It\u2019s also where a lot of programmes quietly stop, because visibility feels like success.<\/p>\n\n\n\n Now the twin doesn\u2019t just report the symptom, it finds the cause. Vibration is up on pump 4 \u2014 and the twin correlates it with bearing temperature, recent load changes and the last maintenance record to tell you it\u2019s a bearing wear signature, not an alignment issue.<\/p>\n\n\n\n What it answers:<\/strong> Why did it happen?<\/p>\n\n\n\n The jump from Level 1 to Level 2 is mostly a data and context jump rather than an AI one. You need history, you need the asset\u2019s relationships mapped, and you need enough data quality that the correlation means something. It\u2019s a smaller leap than most people expect, and often the best-value one available.<\/p>\n\n\n\n The twin stops looking backwards and starts looking forwards. Using the asset\u2019s history and behaviour, it forecasts what will happen \u2014 flagging a failure weeks before it occurs, or projecting when a component will cross its wear threshold.<\/p>\n\n\n\n What it answers:<\/strong> What is going to happen?<\/p>\n\n\n\n This is the level everyone wants, and the level most budgets are written for. It\u2019s also where the technical difficulty steps up sharply: you need enough historical failure data to learn from, models that stay calibrated as the asset changes, and confidence intervals honest enough that operators trust the warnings instead of ignoring them.<\/p>\n\n\n\n A predictive twin that cries wolf gets switched off within months. Trust is part of the engineering here, not an afterthought.<\/p>\n\n\n\n Prediction tells you a bearing will fail in three weeks. Prescription tells you what to do about it \u2014 and what each option costs.<\/p>\n\n\n\n Run to failure and accept the downtime. Intervene at the next planned shutdown. Reduce load now and stretch the window to six weeks. A prescriptive twin simulates those scenarios against your actual constraints and ranks them.<\/p>\n\n\n\n What it answers:<\/strong> What should we do?<\/p>\n\n\n\n This is where a twin stops being a monitoring tool and starts being a decision-support system. It requires simulation capability on top of prediction, plus enough operational and commercial context that its recommendations are realistic rather than theoretical.<\/p>\n\n\n\n The twin closes the loop. Within limits you define, it acts on its own \u2014 adjusting a setpoint, raising a work order, rescheduling a maintenance window, coordinating across multiple assets to protect overall output.<\/p>\n\n\n\n What it does:<\/strong> It decides, and then it does it.<\/p>\n\n\n\n Very few organisations are here, and the barrier is rarely the algorithm. Autonomy is a trust, safety and governance problem before it\u2019s a technical one. You need guardrails, audit trails, clear accountability for what the system does, and an organisation willing to let software act. Most programmes that stall between Level 3 and Level 5 stall for exactly those reasons.<\/p>\n\n\n\n Shows \u2192 Explains \u2192 Forecasts \u2192 Advises \u2192 Acts.<\/strong><\/p>\n\n\n\n That\u2019s it. If you remember nothing else, remember that sequence \u2014 it\u2019s enough to cut through almost any vendor conversation. Ask which of those five verbs their product actually does, and the picture usually gets clear very quickly.<\/p>\n\n\n\n Here\u2019s the pattern behind a lot of disappointing twin programmes: the business case is written for Level 4, the budget is approved for Level 3, and what actually gets delivered and lived with is Level 1. Nobody lied at any point. The levels were simply never named, so nobody noticed the gap until year two.<\/p>\n\n\n\n Being explicit about the level does three useful things:<\/p>\n\n\n\n That last point is worth sitting with. The goal is not to climb to Level 5. It\u2019s to reach the level that changes the decision you actually care about \u2014 and to reach it properly.<\/p>\n\n\n\n Start from the decision, not the technology. Pick the single decision you want to improve, then work out the lowest level that would genuinely change it.<\/p>\n\n\n\n If you want to stop being surprised by failures, you need Level 3. If you just need to stop arguing about why something broke, Level 2 will do it \u2014 at a fraction of the cost. If your maintenance planning is the bottleneck rather than your detection, Level 4 is where the value is.<\/p>\n\n\n\n The mistake is choosing a level because it sounds impressive, then discovering the foundations underneath it don\u2019t hold. Every level depends entirely on the one below: prediction built on data nobody trusts is just a confident guess.<\/p>\n\n\n\n That foundation question is worth answering before you commit to anything. We\u2019ve written about it separately in the Digital Twin Readiness Scorecard<\/a>, and built a free interactive maturity assessment<\/a> that scores your twin across data, model, AI, action and governance in about two minutes.<\/p>\n\n\n\n We help asset-heavy organisations work out which level they need, and then actually get there \u2014 from LPWAN sensors in the field through to twins that change decisions.<\/p>\n\n\n\nLevel 0: not actually a twin<\/h2>\n\n\n\n
Level 1: Descriptive \u2014 it shows<\/h2>\n\n\n\n
Level 2: Diagnostic \u2014 it explains<\/h2>\n\n\n\n
Level 3: Predictive \u2014 it forecasts<\/h2>\n\n\n\n
Level 4: Prescriptive \u2014 it advises<\/h2>\n\n\n\n
Level 5: Autonomous \u2014 it acts<\/h2>\n\n\n\n
The whole model in five words<\/h2>\n\n\n\n
Why knowing your level matters<\/h2>\n\n\n\n
\n
Which level do you need?<\/h2>\n\n\n\n
Where SynapseEdge fits<\/h2>\n\n\n\n