AI does not fix a poorly designed process by itself
If a workflow contains pointless approvals, duplicated data and unclear responsibilities, automating it increases speed without improving the outcome. Before choosing technology, separate what should be eliminated, simplified, standardised, assisted or automated.
The best automation may be stopping a task that should never have existed.
Map the real process
Observe normal cases and exceptions. Record input, output, waiting time, active work, decision, system, owner, rework and error. Do not document only the official procedure: follow a real operation from beginning to end.
Calculate total time and the percentage that creates value. In many processes, the customer waits for days even though the actual work takes minutes. Applying AI to a small task will not solve queues caused by rules and handoffs.
Classify each activity
- Eliminate: it adds neither value nor necessary control.
- Simplify: reduce fields, variants or approvals.
- Standardise: define a common input, output and criterion.
- Assist: AI proposes and a person decides.
- Automate: the system executes within limits.
- Escalate: a person resolves the exception.
Distinguish tasks from decisions
Generative AI can summarise, draft and extract; predictive AI can score and anticipate; an agent can coordinate actions. For high-impact decisions, retain meaningful human oversight and explain which information it uses.
Define the minimum confidence needed for automation and create an intermediate review zone. A system should not treat every case the same when its confidence varies.
Design data and exceptions from the outset
Specify mandatory fields, source, quality and owner. Decide what happens when information is missing, systems do not respond or the result contradicts a rule. Exceptions determine a large part of the operating cost.
Record why the person corrects an output. This data makes it possible to improve rules and detect patterns, provided it is not used to penalise someone for preventing an error.
Measure the complete process
- Cycle time and active time.
- Cost per case.
- First-contact resolution rate.
- Errors, rework and escalations.
- Quality and satisfaction.
- Capacity released and how it is used.
Compare end to end. Saving five minutes on one task may have no impact if the case still waits two days at another stage. Make sure released capacity is reassigned to value-creating work.
Pilot the new workflow
- Establish a baseline.
- Test with a cohort and limits.
- Review failed cases every week.
- Correct the process, data and integration.
- Document controls and owners.
- Scale when the outcome improves, not just the speed.
Relate the case to a shared AI use-case portfolio. Reusable components —data, evaluations, permissions and integrations— can make the next process viable.
Canvas for redesigning a process
- Customer and expected outcome.
- Start and end of the workflow.
- Steps, decisions and waits.
- Data and systems used.
- Frequent errors and exceptions.
- Mandatory controls.
- Opportunity to eliminate, assist or automate.
- End-to-end metric.
Bring together the people who execute, receive and control the process. Mark customer value in one colour, necessary control in another and waste in a third. AI should enter after this conversation, when the problem and limits are already visible.
Frequently asked questions about redesign
Do you need to document every exception?
Start with frequent and high-impact exceptions. Keep an unknown-case category and an escalation channel; operations will reveal new variants.
What should you automate first?
Frequent, stable, measurable and reversible tasks. For uncertain or sensitive decisions, start with assistance and review.
How do you calculate released capacity?
Measure the time actually eliminated and check where it is reassigned. If the person is still waiting, reviewing or duplicating work, the theoretical saving does not exist.
The process redesign we are carrying out with I3OS
Impulsa3’s agentisation was not about placing AI on top of untouched processes. We reviewed how we worked from presales and discovery through onboarding and production, connected pieces that had previously been separate pilots and moved from isolated areas to multidisciplinary squads for each client. Technology accelerates the work, but redesigning the workflow is what turns that speed into real capacity.
If you need to redesign processes and automate them with AI without carrying inefficiencies into the new system, Impulsa3 can analyse the workflow with you and build a measurable solution.
