- Market
- Saudi Arabia
- Written for
- COO, Head of Digital
- Decision
- Automate a back-office workflow with rules-based software rather than manual effort
How we approach it
How we automate a back-office workflow without hard-coding today's mess
The order protects you from the most common failure: pouring concrete over a process that should have been changed first. Understanding and fixing come before a line of automation.
Map the process as it truly runs
We follow the workflow as people actually perform it, including the workarounds and the informal steps, because the documented process and the real one are rarely the same and the real one is what you would automate. A true picture, not the tidy one on the wall.
Fix it on paper before code
We remove the steps that exist only out of habit and simplify the routing before anything is built, because it is far cheaper to change a process in a diagram than in a system that has encoded it. A process worth making permanent.
Encode the rules that are settled
We automate the parts that follow clear rules — who approves what at which threshold, where a request goes next — so the predictable majority of the work runs without a person shepherding it. The routine handled the same way every time.
Keep the exception path human
We route the cases the rules cannot settle to a person with the context to decide, because a process that pretends every case is routine breaks on the one that is not and hides the break. Exceptions seen and decided, not swallowed.
Not an AI project
Why we reach for rules before we reach for a model
A large share of back-office work does not need anything that learns. An approval that depends on an amount and a role, a routing decision that follows a table, a reconciliation that matches two records by a key — these are rules, and rules have real advantages a model does not. They are explainable: you can point at exactly why a request was approved or refused, which matters when an auditor or a regulator asks. They are predictable: the same input gives the same output every time. And they are cheap to run and to change, because changing a rule is editing a rule, not retraining and revalidating a model.
This is a different service from AI-driven automation, and the honest thing is to say so. Where a task genuinely involves judgement over unstructured input — reading a free-text complaint, interpreting a document that never follows a template — a model may earn its place, and that is a separate conversation with its own costs and risks. But reaching for a model to do what a clear rule does better is a way to make a simple, auditable process expensive and opaque. We start from rules and software, use them for the large predictable majority of a workflow, and only raise the question of anything more where the work actually requires it.
One request's path
What happens to an approval request from submitted to recorded
A single request shows where rules do the work and where a person still has to. The automation carries the predictable path; the design is in how it hands the unpredictable one to someone who can decide.
Submitted
Someone raises a request — a purchase, a leave day, a discount — and today waits for it to be noticed.
The request is captured in a structured form so the rules have the fields they need, instead of arriving as an email someone must interpret.
- Intake form
Routed by rule
The request needs to reach the right approver, which today depends on someone knowing the org chart and the thresholds.
A rule sends it to the correct approver by amount, department and policy, so routing is consistent rather than dependent on who happens to be asked.
- Rules engine
Decided or escalated
Most requests fit the policy cleanly; a few sit outside it and need a judgement the rules do not hold.
The clear ones are approved against the rules automatically, and the ambiguous ones are escalated to a person with the full context to decide, not bounced back with no reason.
- Rules engine
- Case queue
Recorded and actioned
Once decided, the outcome has to reach the systems that act on it and leave a trail an auditor can follow.
The decision is written to the systems it affects and logged with who decided and on what basis, so the process is accountable rather than a black box.
- Systems of record
- Audit log
Our role
What we automate and what stays a human decision
- 01
The process, examined and fixed
We map how the workflow really runs and help you simplify it before automating, because we would rather tell you a process needs fixing than sell you the automation of a mess.
- 02
The rules, automated and visible
We build the rule-driven majority of the workflow so it runs consistently and explainably, and instrument it so you can see where work is stuck rather than discover it in a complaint.
- 03
Judgement stays with your people
The cases that need judgement are routed to your team, not automated away; we are clear that automation handles the routine and makes the exceptions visible, it does not replace the decisions only a person should make.
Which decisions must stay with a person, and how the process should work once simplified, are yours to set; we automate the rule-driven majority, route the exceptions to your team with their context, and leave every outcome logged so the process is faster without becoming a decision nobody can account for.
Common questions
What buyers ask
Is business process automation the same as AI automation, or something different?+
Different, and it is worth being precise. The automation we are describing here is rules-based: the process follows clear, written logic — this amount goes to this approver, this record matches that one by a key — and software runs that logic quickly and consistently. It does not learn and it does not guess; the same input always gives the same output, which is exactly what you want for an approval an auditor may review. AI-driven automation is a separate thing, for tasks that involve judgement over unstructured input, like interpreting free text, and it brings its own costs and risks. Most back-office work is genuinely rules, so we start there, and only raise a model where the work truly cannot be expressed as rules.
Should we automate a process before or after we fix it?+
After, almost always, because automation locks a process in and multiplies whatever it locks. If you automate a workflow that carries steps existing only out of habit, approvals that add no control, or routing that sends work in circles, you get all of that faster and harder to change, and you have spent money making a bad process permanent. The cheap moment to fix a process is on paper, before it is encoded; the expensive moment is after, when the fix means changing a system people now depend on. So we map the real workflow, simplify it with you first, and automate the version worth keeping. Occasionally a process is already clean and can be automated as it is — but that is a conclusion we reach by looking, not an assumption we start from.
Where does RPA fit, and where does it become a liability?+
RPA — software that drives a system through its user interface, as a person would — fits where you need to automate work in a system that has no API to integrate with and cannot be changed, often an older or third-party application. Used there, deliberately, it is a pragmatic bridge. It becomes a liability when it is treated as a permanent substitute for a real integration, because it depends on the screen staying exactly as it is: a layout change, a renamed button or an extra dialog can break the automation silently, and now you maintain a brittle robot that mimics a person clicking. Our approach is to integrate properly wherever a system offers a way to, and to use RPA narrowly and knowingly where nothing else is available, with monitoring so a break is caught quickly rather than discovered when the numbers are wrong.
In this cluster
Next step
Map a process to automate
A mapping session over one workflow that separates the rule-driven steps worth automating from the exceptions that need a person, before any automation is built.

