Decision guide
Automation vs doing it manually
Not every repetitive task is worth automating. The deciding factors are how stable the rules are and what an error actually costs.
Automation is usually sold on hours saved, which is the easy half of the calculation. The half that gets skipped is that a bot is software: it has a maintenance burden, it breaks when an upstream system changes its layout, and somebody has to own it. A process automated without an owner becomes a silent failure waiting to happen.
The useful test is not "is this repetitive?" — it is "are the rules stable, and can a machine tell when it is unsure?" A process with clear rules and a clean way to escalate the ambiguous cases automates well. A process where a human is quietly applying judgement on every item will produce a bot that is confidently wrong at scale.
Side by side
| Criterion | Keep it manual | Automate it |
|---|---|---|
| Best suited to | Low volume, high judgement, frequently changing rules. | High volume, stable rules, decisions that can be written down. |
| Cost shape | Linear. Twice the volume needs roughly twice the people. | Up-front build plus ongoing maintenance, then near-flat as volume grows. |
| Error behaviour | Errors are individual and usually caught by the person doing the work. | Errors are systematic — one bad rule affects every item until someone notices. |
| Handling exceptions | Native. People notice when something looks wrong. | Must be designed. A bot needs an explicit confidence threshold and a human review queue. |
| When upstream changes | People adapt, often without telling anyone. | Breaks, and needs a fix. This is the maintenance cost most projections omit. |
| Audit trail | Depends on discipline and memory. | Complete by construction — every decision is logged. |
Keep it manual if
- The volume is genuinely low and the work is not growing.
- The rules change often enough that the bot would need constant rework.
- Every item needs real judgement rather than rule-following.
- Nobody would own the automation once it is built.
Automate if
- Volume is high enough that people are the bottleneck.
- The rules are stable and someone can state them precisely.
- The work is a queue with a clear start and finish, not open-ended.
- An audit trail is valuable — or required.
Our take
The mistake to avoid is automating the process you have rather than the process you need. Map it first: a surprising share of steps in long-running manual processes exist to work around a limitation that no longer applies. Automate what survives that review, always with a confidence threshold and a human queue for the exceptions a bot should not be deciding on its own.
Questions we get asked
How do we know if a process is a good automation candidate?
Ask whether someone can write the rules down completely. If the answer involves "it depends" more than occasionally, the judgement is doing real work and the process either needs simplifying first or should stay with a person.
What happens when the system a bot depends on changes?
The bot breaks, which is why maintenance has to be budgeted rather than assumed away. We build monitoring and alerting alongside the automation so a failure is visible immediately, not discovered at month end.
Will automation replace the people doing this work?
In the engagements we run it usually moves them. Data entry becomes exception handling and supplier or customer work — the tasks that need the judgement a bot cannot supply.
What does a first automation cost?
A first production bot typically runs $4k–$11k. We model the payback during discovery so the decision is made on your numbers rather than a generic claim.
More decision guides
In-house hiring vs a dedicated team
Both give you engineers. They differ in how fast capacity arrives, what happens when the roadmap changes, and where the knowledge ends up.
Rewrite vs modernize a legacy system
The rewrite is almost always the more appealing option and almost never the safer one. Here is how to tell which situation you are actually in.
Still weighing it up?
Tell us the constraint you are working against and we will tell you which way we would go.