How to Digitalise a Factory Step by Step

Digitalising a factory does not start with buying software: it starts with knowing what the plant loses and where. This guide lays out the complete journey in five steps (diagnostic, data connection, visualisation, decision automation and artificial intelligence), explains what not to do first, who needs to be involved in each phase and how the project is funded in the Valencia region of Spain.

Where to start (and where not to)

The most common question from an industrial general manager is not whether to digitalise, but in what order. The short answer: in the order in which each step generates the data the next one needs. A dashboard without reliable data shows pretty, false numbers. An AI algorithm trained on incomplete manual records learns the mistakes of the paperwork. And an ERP implemented without first reviewing the processes digitalises the existing chaos, just faster.

That is why the journey has a natural sequence: first understand the plant, then connect it, then see it, then automate the repetitive work and only at the end apply artificial intelligence. Each phase rests on the previous one. Skipping a step does not accelerate the project: it makes it more expensive, because it forces you to go back when the advanced phase discovers the foundation does not exist.

This sequence is not consultancy theory. It is the repeated lesson of projects that fail: most do not fail because of the technology chosen, but because they started with the roof. Let us go step by step.

Step 1. Diagnostic: know where you stand before deciding where to go

The first step requires buying nothing. It consists of answering three questions with data, not intuition: where the plant loses time and money, what information already exists (even if it lives on paper, in spreadsheets and in the shift manager’s head) and what systems are installed and in what state.

A serious operational diagnostic produces three concrete deliverables:

• Process and loss map. What the plant produces, with what flow, and where stoppages, waiting, rework and waste accumulate. If the plant measures OEE, this map starts from it; if it does not, the diagnostic usually begins by establishing that measurement (the OEE guide explains how to do it rigorously).
• Data and systems inventory. Which machines have PLCs and of what generation, which signals are available, what software exists (ERP, spreadsheets, bespoke applications) and what information is recorded by hand.
• Prioritised roadmap. A short list of projects ordered by expected return and technical dependencies, not a five-year master plan nobody is going to execute.

The trap in this phase is turning it into an endless study. A useful diagnostic is tightly scoped: it observes the plant in operation, interviews the people who run it and produces an actionable roadmap. That is exactly the scope of our operational diagnostic service: leaving with a prioritised list of what to do, in what order and why.

Step 2. Connect the shop-floor data

With the diagnostic done, the second step is to make the data flow on its own. Two worlds live here that historically do not talk to each other: OT (the machines, the PLCs, the sensors) and IT (the ERP, the databases, the office software). Connecting them is the infrastructure project underpinning all subsequent digitalisation: OT/IT integration is the foundation the rest is built on.

Connecting does not mean wiring the whole factory at once. It means, in this order:

1. Capture the critical signals. For most plants, three signals per machine are enough to start: run/stop status, piece counter and reject counter. With that you can already calculate an automatic OEE.
2. Centralise in a common data layer. Have the signals from every machine arrive at a single point with coherent naming, instead of point-to-point integrations that multiply with every new system. That common layer is what we call an industrial data platform.
3. Connect shop floor and management. Have the ERP’s works orders come down to the plant and the production reports go up on their own, without manual retranscription.

An important nuance: you do not need to replace old machines to connect them. A 1990s machine without an accessible PLC is connected with external sensors (photocells, electricity consumption meters) for a fraction of the cost of replacing it. Connectivity is designed for the plant’s real machine fleet, not for the manufacturer’s catalogue.

Step 3. Visualise: indicators people actually use

The third step turns connected data into information that changes decisions. This is where the project becomes visible to the whole organisation, and where internal credibility is won (or lost).

The golden rule: few indicators, in real time, visible where decisions are made. A shop-floor panel showing the shift’s OEE and the cause of the last stoppage changes behaviour that very day. A forty-page monthly report nobody opens changes nothing. Well-designed operational dashboards answer specific questions from specific people: the operator wants to know whether they are on pace, the shift manager where the bottleneck is stuck, and management how the month is closing.

This phase also has a valuable side effect: it uncovers data-quality problems. When the panel shows a performance of 120%, someone has the ideal cycle badly defined. Fixing those inconsistencies now is cheap; discovering them at step 5, with an AI model trained on faulty data, is expensive.

Step 4. Automate decisions and workflows

With reliable, visible data, the fourth step is to eliminate repetitive work and decisions that no longer need a person. This phase usually has two fronts:

Automation of management workflows. Production reports that generate themselves, automatic warnings when a machine has been stopped too long, replenishment triggered when a minimum stock level is crossed, delivery notes and invoices that are not transcribed twice. This is where a well-implemented industrial ERP makes the difference: if the plant runs on Odoo or is considering it, our page on Odoo for industry details how management connects with the shop floor, and the article on how much implementing Odoo in an SME costs breaks down that project’s cost drivers.

Automation of operational decisions. Rules that act without waiting for anyone: if the furnace temperature drifts, adjust; if a line’s reject rate exceeds a threshold, notify quality; if electricity consumption spikes while idling, alert maintenance. These are simple, deterministic automations, no AI involved. And they solve a surprisingly large share of daily problems.

Step 5. Artificial intelligence on reliable data

AI is the last step, not the first, for a practical reason: models learn from historical data, and until steps 2 and 3 have been running for a while, that history does not exist or is not reliable. With the foundation built, the industrial use cases with the most mileage are concrete:

• Predictive maintenance: anticipating breakdowns from patterns in vibration, temperature or consumption, instead of waiting for the failure or swapping parts by calendar.
• Demand forecasting and planning: adjusting production and purchasing to what will actually sell, not to last year’s average.
• Automatic visual inspection: cameras detecting defects at line speed, with a consistency human inspection cannot sustain shift after shift.

The criterion for choosing the first AI use case is the same as for the whole project: start with the problem that hurts most and that already has data, demonstrate value in a tight scope and scale afterwards. An AI pilot on one well-instrumented line is worth more than a corporate artificial intelligence plan on an unconnected plant.

What not to do first: the three shortcuts that end up costing dearly

Do not start by buying the tool. Software chosen before the diagnostic is chosen by demo, not by need. The classic symptom is the plant with an expensive MES used as a report log, or the powerful ERP of which 15% is used. First the problem, then the tool.

Do not start with AI. It is the most tempting shortcut and the most expensive. A predictive maintenance project on unsensored machines forces you to do steps 1 to 3 anyway, but in a rush, without an assigned budget and with the project’s credibility already spent on the initial promise.

Do not digitalise everything at once. The programme spanning the whole factory in every dimension never finishes and burns out the organisation. The tightly scoped sequence works better: one pilot line, measurable results, and extension to the rest of the plant with what has been learnt. The total cost of the journey, and above all the factors that push it up or down, are covered in detail in how much digitalising an industrial company costs.

Who needs to be involved in each phase

Digitalisation fails more because of people than because of technology, so it pays to be clear about who is indispensable at each step:

One cross-cutting pattern: every phase needs an internal project owner with real time allocated. Not a symbolic title, but someone who dedicates weekly hours to deciding, unblocking and demanding. Projects without that counterpart drag on hopelessly, however good the vendor is.

How long each phase takes (in relative terms)

Absolute timescales depend on the size of the plant, the state of the machine fleet and the internal dedication, so we give proportions, not dates:

• The diagnostic is the shortest phase of the project, by far. If it is dragging on, it is a sign it has turned into the endless study to be avoided.
• Data connection is the longest of the foundational phases: it depends on the variety of the machine fleet and on the shutdown windows available to work on it. The more heterogeneous the fleet, the heavier it weighs.
• Visualisation is fast if phase 2 was done well: with data centralised, standing up the first panels is a matter of weeks, not months.
• Automation is incremental by nature: it is rolled out workflow by workflow and coexists with daily operations. It does not have an end; it has a pace.
• AI demands history before starting: between the start of data capture and the first useful model lies the time needed to accumulate representative data, including the business’s seasonal cycles.

The practical consequence: the first visible results (panels with real data, the first automatic alerts) arrive well before the end of the journey. A good project is designed so that each phase pays for the next, in results and in internal confidence.

How it is funded: grants and investment order

In the Valencia region of Spain, industrial digitalisation has specific public funding lines. The main one is DIGITALIZA-CV, run by IVACE (the Valencian regional agency for business innovation): in its 2026 call it subsidised digitalisation projects of industrial SMEs with up to 175,000 euros per company, at an intensity of between 35% and 45% of the project, a minimum eligible budget of 20,000 euros and an advance payment of 75% of the grant. It is an annual call: the 2026 edition was open from 28 May to 3 July, and future editions are published each year with their own deadlines and conditions. The full details, with eligibility requirements and eligible costs, are in our guide to DIGITALIZA-CV and IVACE grants.

Two practical notes on funding. First: grants condition the calendar, not the strategy. The order of the five steps does not change because a call is open; what changes is which phase is submitted for subsidy and when it is executed. Second: management software qualifies too. In the 2026 DIGITALIZA-CV call, ERP implementation was an eligible cost, which makes it especially efficient to combine the automation phase (step 4) with a call.

As for your own money, the rule is to invest in the order of the steps: the diagnostic is the smallest investment and the one that makes all the others pay off best, and data connectivity is the infrastructure investment the later phases amortise over and over again. The market ranges and the factors that make each type of project cheaper or more expensive are analysed in the cost series, starting with how much automating a production line costs.

Frequently asked questions about digitalising a factory

If you want to know where your plant stands and what your specific roadmap would be, the starting point is an operational diagnostic: at Captia Technology we run it on site, with your team, and it ends in a prioritised list of projects, not a report for the shelf.