Workflow automation means letting software run, on its own, the sequence of tasks, decisions and handovers of a work circuit: it triggers when an event arrives, routes each task to the right person or system, applies the rules, tracks deadlines and keeps a record of every step. Put another way: work flows by itself, instead of waiting in email inboxes and shared spreadsheets.
This guide explains what a workflow is and what it means to automate it, the six patterns that cover the vast majority of cases, how it differs from BPM, from an iPaaS like Zapier or Make and from RPA, when to custom-code and when to use a low-code platform, which metrics to watch and which seven workflows to start with to see results in weeks.
A workflow is the ordered sequence of tasks, decisions and handovers that carry a piece of work from start to finish: a leave request that is submitted, approved and recorded; an invoice that is received, validated, approved and posted. Automating it means handing that sequence to software so it runs by itself: triggering at the right moment, routing each task to the right person or system, applying the rules, watching the deadlines and recording who did what and when.
It is worth not confusing three levels that everyday language blurs together:
The smallest unit: "approve", "sign", "fill in the form". It is done by a person, a system or an AI agent. A standalone task is not a workflow.
The chaining of tasks of one specific circuit, with its order, its decisions and its owners. This is what gets automated.
The set of workflows, data and rules that deliver value end to end (e.g. "hiring an employee"). This is the domain of BPM.
A workflow is therefore the basic unit of the process, and automating it is the first rung of a broader practice of process management. If you want the full framework —lifecycle, BPMN 2.0 and continuous improvement— we develop it in the pillar guide What is BPM?. Here we stay on the practical plane of the workflow.
Almost any real circuit is a combination of six patterns. Recognising them saves many hours of design: instead of inventing the flow from scratch, you identify which patterns make it up.
When to use it: tasks that must be done in a fixed order, one after another. Example: log an incident → diagnose → resolve → close. It is the base pattern; the rest are variations on it.
When to use it: someone must sign off before continuing, and it should not get stuck. Example: an expense request goes to the manager; if they do not respond within 48 h, it automatically escalates to their superior. Escalation is what stops a forgotten approval from blocking the circuit for weeks.
When to use it: several independent tasks that can happen at once and must converge before continuing. Example: in employee onboarding, HR prepares the contract while IT creates the accounts; "day one" does not arrive until both branches finish. It shortens total time without losing control.
When to use it: work does not advance in a straight line but moves through states with allowed transitions. Example: a case file that goes from "draft" to "under review", can return to "draft", move to "approved" or "rejected". Ideal for tickets, case files and cases where back-and-forth is the norm.
When to use it: meeting a response time matters and has to be measured. Example: an incident with a 4 h response commitment and 24 h resolution; the workflow warns before breaching and records whether the SLA was met. It turns a service agreement into something measurable, not a promise.
When to use it: an AI proposes or executes a step, but a person validates when confidence is low or the amount is high. Example: the AI extracts the data from an invoice and, if the amount exceeds a threshold or confidence drops, it sends it to human review before posting. It is the pattern that combines machine speed with human judgement and that we cover in depth in BPM with AI agents.
Beyond the diagram, a workflow that works in production always has the same seven pieces. If any is missing, the circuit breaks on the first odd case or stops being auditable.
What starts the circuit: someone submits a form, an email or a message from another system arrives, a date falls due, or an event from your ERP/CRM. A good trigger is unambiguous and makes clear who the requester is.
The information that travels through the flow. Capturing the data well at the start (with validations) avoids most of the later back-and-forth. The data is what the rules use to decide.
The automatic decisions: from what amount double approval is required, which cases go through the fast lane, when something is rejected without human intervention. Externalising the rules (instead of "hiding" them in the code) lets you change them without rebuilding the flow.
Each task goes to a role (the requester's manager, the finance team), not to a named individual. That way the workflow survives holidays, leave and reorganisations without touching the design.
Each human task carries a deadline; when it expires, it reminds, reassigns or escalates. It is the difference between a circuit that moves by itself and one that depends on someone remembering to check their inbox.
The calls to other systems: read a value from the ERP, create a user, send for signature, notify by email. An isolated workflow forces copy-pasting; an integrated one moves the information by itself.
The record of who did what, when and with which document at hand. It is what turns an audit (GDPR, ISO, tax) into a query rather than archaeology through emails. Without this piece, there is no governance.
Four categories that get confused constantly because they all "automate". The useful way to tell them apart is by what they solve and by their honest limits:
| Category | What it solves best | Where it falls short | Example |
|---|---|---|---|
| iPaaS (Zapier / Make) | Connecting SaaS apps: "when X happens in one app, do Y in another" | Human tasks, approvals, deadlines, role-based governance and volume at a predictable cost | Create a row in a sheet when a lead arrives |
| Workflow / BPM | Orchestrating circuits with people, rules, deadlines, documents and auditable evidence | For 2-3 trivial automations between apps it can be overkill | Approve invoices with escalation and signature |
| RPA | Running a repetitive task in a system without an API, mimicking clicks | It does not orchestrate the process or manage people; it is fragile to screen changes | Dump data into a legacy ERP without an API |
| BPM (discipline) | Designing, measuring and improving all processes continuously and under governance | It requires a management practice, not just a tool | Full onboarding, from offer to day one |
To be honest: if your need is to connect two or three SaaS apps with simple automations, Zapier or Make have capacity to spare and are the fastest and cheapest option. The inflection point comes when human tasks with inboxes and approvals appear, along with deadlines and escalations, complex forms, role-based permissions, auditable evidence, high volume where the price per operation spikes, or the requirement that the data live in the EU. From that point on a workflow/BPM platform is the better fit. We develop this in detail —without sugar-coating when migration is NOT needed— in Zapier alternative and Make alternative.
Once you have decided you need more than an iPaaS, the classic question remains: do we custom-code it or use a platform? Both options are legitimate, but their cost is spread very differently over time.
Maximum flexibility and control. But the cost is not in building it: it is in maintaining it. Every rule change, every new integration, every changing regulation requires touching code and deploying. The knowledge sits in few hands and the IT backlog becomes the bottleneck of every operational improvement.
The form, the rules, the deadlines and the standard integrations come ready-made; publishing a new version does not require deploying. The cost concentrates into a predictable subscription and the first productive workflow is measured in weeks. It empowers the citizen developer: the analyst who knows the circuit builds it, with IT governing the platform.
The honest way to decide is to compare the three-year total cost —build + maintenance + the opportunity cost of the processes that never get automated because "it is not worth opening a development"—. That is precisely where low-code changed the market: small processes that would never have justified a custom project now get automated in days. You can see it in a platform like Dokuflex low-code BPM.
Do not start with the most complex process in the house. Start with one that has high volume, clear rules and visible pain. These seven are the ones we see most in production and the ones that deliver results fastest:
Request → manager approval with escalation → balance recording. The perfect candidate for the first workflow: high volume and clear rules. See solution →
Receipt and OCR reading → validation → approval by amount → posting, with tax compliance. Invoices and AEAT →
A pure parallel pattern: HR, IT and the team prepare everything for day one in parallel. BPM for HR →
Receipt capture → validation → approval → reimbursement. High volume and simple rules by amount. Expense management →
Document generation → review → electronic signature → archiving with evidence. Digital signature →
Incidents are the textbook example of the state-based pattern; supplier onboarding combines a form, validations and approval. Two simple circuits to build momentum.
Would you rather start from a ready-made model than a blank page? The catalogue of workflow templates brings these circuits ready to activate and adapt.
Automating without measuring is stopping halfway. The great advantage of an automated workflow is that every run leaves data; these four metrics are the ones worth reading from day one.
How long the circuit takes from start to finish. It is the metric the business notices most and the one that usually justifies the project ("from 9 days to 2 in approving an invoice").
Of the cycle time, how much is real work (touch time) and how much is waiting in an inbox (wait time). In email-driven processes, wait time is usually the vast majority: that is where almost all the room for improvement lies, and it is what automation attacks.
What percentage of cases leave the happy path (rejections, incomplete data, reassignments). A high and stable rate signals a badly set rule or a form that captures the data poorly at the start.
The percentage of cases resolved within the committed deadline. It is only measurable if the workflow tracks deadlines (pattern 5); it turns a service agreement into a management data point, not a feeling.
That this is not theory is shown by the case studies: at the Sant Pau Hospital, automating documentary circuits saves around 12,000 hours a year by eliminating much of the wait time and the repetitive manual work.
AI has entered workflow automation through three doors, in order of maturity, and they all make sense only if the AI acts inside the flow, with its context and its log:
The AI reads invoices, contracts and forms without templates and pours the data into the start of the workflow. It is the most mature case and the one with the fastest return, because it attacks manual data capture at the source.
Deciding which queue a ticket goes to, which request takes the fast lane, which document belongs to which case file. The AI classifies and the workflow routes accordingly, reducing manual triage.
An agent executes a complete step —it reasons over the case, decides with justification and acts— always with a confidence threshold and escalation to a human (pattern 6, human-in-the-loop). It is the 2026 frontier and we cover it in depth in the guide BPM with AI agents.
The rule that separates value from risk is the same as always: AI acts inside the workflow, never on the side. Inside, it inherits permissions, leaves an auditable log and has oversight; outside, it is a loose tool without governance.
Workflow automation means letting software run, on its own, the sequence of tasks, decisions and handovers of a work circuit: it triggers when an event arrives, routes each task to the right person or system, applies rules, tracks deadlines and keeps a record of every step. The goal is for work to stop waiting in email inboxes and shared spreadsheets.
A workflow is the sequence of tasks of one specific circuit (for example, approving a leave request). BPM (business process management) is the full discipline: it includes the workflow as its basic unit, but adds standard BPMN modelling, business rules, integrations, metrics, auditing and continuous improvement across all the organisation's processes. In practice, automating workflows is the first step towards BPM.
Zapier and Make are very good at connecting SaaS apps and at task-to-task automations (when X happens in one app, do Y in another). They fall short when the workflow needs human tasks with inboxes and approvals, deadlines and escalations, complex forms, governance and role-based permissions, auditable evidence, high volume at a predictable cost or data residency in the EU. From that point on a workflow/BPM platform is the better fit.
It depends on the approach. Custom-coding it carries a high upfront cost and recurring maintenance every time a rule or an integration changes. With a low-code platform, the cost concentrates into a predictable subscription and the first scoped workflow is usually in production within weeks. The honest way to estimate it is to compare the 3-year total cost (licence + build + maintenance) against the hours it saves; the plans and pricing detail are on the pricing page.
Not with a low-code workflow platform. A business analyst who knows the circuit can build the form, the rules, the role-based assignments and the deadlines visually, without writing code. IT governs the platform (security, integrations, data) instead of coding each circuit. Only very specific integration cases call for some development.
Bring a real circuit —leave, invoices, onboarding— and we will show it automated in Dokuflex in a 30-minute demo. No commitment and with your case, not a lab one.