Your team re-enters the same data in three tools, chases approvals by email and waits for a spreadsheet to be "up to date." You know you need to act, but the question is not only what to improve: it is how. Should you automate, integrate or replace a system? And which process should you start with?
This article is written for managers and IT leads in organizations who need to set priorities without launching a major overhaul. It proposes a six-step approach: map, score, clean up, choose the right approach, test with a pilot, keep a human in control.
Step 1: map before you decide
You cannot prioritize what you cannot see. For each candidate process (invoice processing, onboarding a new member or client, an access request, a monthly report), note the following on a single page:
- The trigger: what starts it?
- The actual steps, as people carry them out, not as the manual describes them.
- The tools and data involved at each step.
- The people who take part and the approvals required.
- The exceptions: what happens when a case falls outside the ordinary?
- The end: how do you know the process is complete?
Ask the people who do the work to review the map. The gap between the official procedure and actual practice is often the first useful discovery.
Step 2: score by frequency, volume and risk
Once five to ten processes are mapped, compare them using a simple grid. Guides on the subject converge on a few criteria. According to the selection criteria checklist from Decisions (a page first published under the ProcessMaker name), good candidates are high-volume, manual and repetitive processes that are standardized (few variations), governed by clear rules and fed by structured digital data. The same page also mentions the need for measurable results, systems that offer integration points (APIs), quality data and resources to maintain the solution.
In its guide to prioritizing processes for automation, Activepieces proposes scoring by volume (how often it runs and time per run), complexity (number of conditional branches and external integrations) and business value, stressing that the cost of human error can outweigh the cost of labour. Both sources are publications from tool vendors: treat them as frameworks for thinking, not as standards.
Here is an adaptable grid to fill in as a team (scores from 1 to 3, as an example of method):
| Criterion | Question to ask |
|---|---|
| Frequency | How often does the process run (daily, weekly, seasonal)? |
| Volume | How many cases per run, and how many minutes per case? |
| Risk of error | Could a mistake result in financial loss, loss of trust or failure to meet a legal obligation? |
| Variability | What proportion of cases fall outside the usual path? |
| Technical complexity | How many systems and rules are involved? |
| Sensitivity | Is personal information or are important decisions involved? |
The numbers in the grid only have value in your own context: the goal is to rank the processes against one another, not to produce a universal score.
Step 3: do not automate a bad process
Automating a muddled process does not fix it: it just runs faster. The guides consulted say so in their own ways. Decisions points out that not everything should be automated and that high exception rates increase complexity and cost; the page suggests, as a guideline, an exception rate below 15–20% to identify standardized processes. This range is that vendor's recommendation, not a universal standard: whether automation makes sense also depends on the nature, predictability and handling cost of the exceptions. It also recommends starting with simple, high-value processes before tackling complex flows.
Before automating, ask yourself these questions:
- Does each step still serve a purpose, or does it exist out of habit?
- Are the decision rules written down and accepted by everyone?
- Are the exceptions known, or discovered through incidents?
- Is the quality of the input data sufficient?
If the answer is "no" to several of them, simplify first. This is sometimes the project's biggest gain, and it requires no software.
Step 4: automate, integrate or replace?
Three levers address three different problems.
Automate: when the task is repetitive and well defined
Automating means having a system perform steps that people do by hand: sending a reminder, generating a document, transferring data according to a rule. It makes sense when the process is stable, the rules are clear and the volume justifies the effort.
Integrate: when the systems are good but isolated
Integrating means connecting existing systems so that data flows without re-entry. It is often the right answer when each tool does its job well but information gets lost in between. A word of caution, though: Activepieces points out that adding applications can increase the number of possible points of failure; this is a tool vendor's claim, and the actual risk depends on the architecture of the integrations, how they are monitored and the recovery mechanisms in place. For its part, Decisions notes that custom integrations with legacy systems can be complex and costly.
Replace: when the system itself is the problem
Replacing is justified when the tool can no longer evolve, does not integrate, is no longer supported or forces permanent workarounds. It is the heaviest option (data migration, training, transition). Do not choose it out of a "start over from scratch" reflex: choose it when integrating or automating does not address the cause.
A simple decision tree
- Is the process muddled or contested? Simplify first.
- Is the process clear but manual? Automate.
- Are the tools adequate but disconnected? Integrate.
- Is the tool blocking the process, no longer supported or unable to evolve? Consider replacing it, with a migration plan.
In practice, these approaches are combined: you replace a core system, integrate it with the rest, then automate the repetitive tasks that follow from it.
Step 5: validate with a pilot
Choose a process with clear value and limited risk, then frame the pilot:
- Reduced scope: one department, one type of case, a defined period.
- Success criteria written before the start (for example: fewer re-entries, fewer follow-ups, no undetected amount errors). Define them yourself based on your starting situation.
- A baseline measurement: without a record of the current state, you will not know what improved.
- A rollback plan: how do you return to the manual method if the pilot fails?
- An end-of-pilot review: what worked, what was surprising, what do we roll out more widely?
A successful pilot gives you arguments; a failed pilot gives you inexpensive lessons.
Step 6: keep a human in control
Automating a step does not remove responsibility. For decisions that involve money, access rights, personal information or people, keep human validation and a record of what happened. A good general practice:
- Escalation thresholds: beyond an amount or risk level that you define, a human validates.
- Logging: who did what, when, and under which rule.
- Exception handling: an unforeseen case is queued for a person, never decided silently.
- A designated owner: one person is accountable for the proper functioning of each automation.
- Periodic review: rules, connected systems and volumes change; the automation must be reviewed.
Checklist before launching a project
- The process is mapped and validated by the people who carry it out.
- The candidates are ranked using a common grid.
- The rules and exceptions are written down; the process is simplified as needed.
- The choice (automate, integrate, replace) is justified by the real problem.
- A pilot with a limited scope and written success criteria is planned.
- The rollback is defined.
- The human validation points and logging are planned.
- One person is responsible for keeping it in service.
Frequently asked questions
Which process should we start with?
A frequent, well-understood process with clear rules and limited risk. The sources consulted recommend starting simple before tackling complex flows.
Should we always replace an old system?
No. If the system still does its job well and can be connected to the others, integration is often the more measured choice. Replacement is justified when the system prevents you from reaching the goal.
Can everything be automated?
No. Processes that require judgment, that are unstable or that involve many exceptions are poorly suited to full automation. Partial automation with human validation is often more appropriate.
How do we know whether the pilot succeeded?
By comparing the results with the criteria defined before the start and with the baseline measurement of the initial state. Without those two reference points, the conclusion remains an impression.
Conclusion
Prioritizing starts with seeing clearly: map, compare by frequency, volume and risk, then improve processes that are not worth equipping in their current state before introducing new tools. The choice between automating, integrating or replacing follows from the real problem, not from the technology of the moment. A well-framed pilot and well-placed human oversight reduce unpleasant surprises.
Learn more
- Automation
- Systems integration
- Talk about my project: describe your situation and we will discuss it.

