Use case
Automate a process without losing control.
The process works. It simply consumes a lot of human time on repetitive handling: re-keying, files moved by hand, chasing people for an approval. Automating does not mean taking people out of the loop — it means giving them back the steps where their judgement matters.
You may recognise some of this
These signs show up by watching a working day, not by reading a procedure.
- The same information is entered into two or three different tools.
- A spreadsheet circulates by email and several people treat it as the reference.
- Documents are uploaded, renamed and moved by hand at every step.
- Approvals run through email chains, and nobody knows where a case stands without asking.
- Tracking depends on one person who has "the overview" in a personal spreadsheet.
- When an error surfaces, finding its origin means walking back through several exchanges.
What happens if nothing is done
The process keeps working. It is its margins that degrade.
- Time goes into handling, not into the work
- Qualified people spend part of their day copying, renaming and chasing. That time does not show up in a budget, but it is paid for.
- Errors are hard to trace
- Without a record of who changed what and when, fixing an error means reconstructing the history from emails. The real cost of an incident is the investigation, not the correction.
- The process does not scale with volume
- Manual handling absorbs a gradual increase until a breaking point, usually reached at the worst moment: a busy period, an absence, an important new client.
- The knowledge stays in habits
- When the rule is written nowhere, a new joiner needs weeks of shadowing, and someone leaving takes the exceptions with them.
Not everything should be automated
A process breaks down into steps. Each one belongs to one of these three regimes, and the value comes from telling them apart correctly — not from pushing everything towards automatic.
| Not everything should be automated | When it is the right regime | What it assumes |
|---|---|---|
| Automatic | The task is stable, the rules are known and can be stated, and an error is catchable by a check. | Writing the rules out explicitly, edge cases included, and deciding what happens when an input is invalid. |
| Assisted | The system can prepare the work — gather, pre-fill, flag an anomaly — but the decision commits the business. | A fast review interface, with the context in front of the person, and the ability to correct without starting over. |
| Manual | The case is rare, ambiguous, sensitive, or needs a judgement call that cannot be coded. | Owning it: an explicit manual path, traced like the others, beats an automation that people quietly work around. |
A fully automated process that gets bypassed at the first edge case is worth less than an assisted one everybody actually uses.
The same process, before and after
What changes is not that steps disappear: it is that they stop being carried by manual handling and chasing.
Today
- Spreadsheet
- Copy
- Tool A
- Check
- Tool B
- Chase
Afterwards
- 01
One entry point
A form, a document upload or an incoming message: a single arrival point, timestamped and traced.
- 02
Rules applied
Consistency checks, calculations, routing to the right handling path according to written rules.
- 03
Review where it commits
A person decides on what matters, with the case already prepared. Compliant cases go through without a pointless stop.
- 04
Written into the systems
Data lands in your existing tools, with no re-keying and no intermediate file.
- 05
History and notification
Who did what, when, and on what basis. The people concerned are told without having to be chased.
The review step is not a leftover from the manual process: it is designed in, and it is what makes automation acceptable to the teams.
How we approach it
We start by watching the process as it actually runs, not as a procedure describes it.
- 01
Observe the real process
Following one case end to end with the people who handle it. The workarounds they have invented are the most useful output of this step.
- 02
Spot the waits, the re-keying and the exceptions
Where a case waits for someone, where the same data gets written twice, and how often cases fall outside the rule.
- 03
Choose what deserves automating
Not all steps are equal. A rare, complex step is expensive to automate for a small gain; a mundane one repeated a hundred times a week is the opposite.
- 04
Connect the systems involved
Most re-keying exists because two tools do not talk. Dealing with the exchanges often comes before the workflow itself.
- 05
Automate in stages
One portion of the process at a time, actually put into service, with the manual path still available during the transition.
- 06
Measure and adjust
Volume handled, share of cases sent to review, remaining bottlenecks. Without measurement, there is no way to know whether the next step is worth it.
What this can look like
Depending on the process and the tools in place, automation takes quite different shapes.
- An internal workflow replacing a chain of approvals by email
- A portal where clients or partners upload what used to arrive by inbox
- An integration between two tools that removes a daily re-keying task
- Automatic handling of incoming documents, with human review below a confidence threshold
- Documents generated from the data instead of templates filled in by hand
- Notifications triggered by process events, instead of chasing
These describe possible shapes of a solution, not delivered projects presented as references.
This path may draw on
Analysing the process comes first. The other areas come in depending on what it concludes.
Frequently asked questions
Do we have to replace our existing software?
Usually not. Most of the gain comes from connecting what exists and organising the handover from one step to the next, not from changing the business tools. Replacing a system is a separate decision, taken for its own reasons.
Can a process with many exceptions be automated?
Yes, as long as you do not try to automate all of it. You handle the main path, which usually covers the large majority of cases, and route exceptions explicitly to a human. The gain comes from volume, not from completeness.
How do we keep a human approval?
By designing it as a step in the process from the start, not as a safeguard bolted on afterwards. That means an interface where the person sees the prepared case and the context they need, can correct it, and whose action is traced like the automatic steps.
What is the difference between classic automation and AI?
Classic automation applies rules you can state: if this condition, then that action. It is predictable and verifiable. AI is for when the rule cannot be stated — reading an unstructured document, classifying a free-text request. It is useful on those specific steps, and it needs a confidence threshold and a review path.
How do we start without launching a big project?
By taking a single step — the most repetitive and the best understood — and handling it completely. That gives a usable result in a few weeks, a real measurement of the gain, and a basis for deciding whether the next step is worth it.
Let us talk about your process
Tell us which step your teams repeat the most. That is almost always the right starting point.