Qatreh AI Logoقطره
Back to blog
AI Adoption and Management

Redesign the Process, Do Not Add a Tool

Six practical steps: pick one process, measure the current state, find the real constraint which is usually waiting rather than work, and choose the tool last.

تیم قطرهSeptember 6, 20266 min read
Redesign the Process, Do Not Add a Tool

"Workflow redesign" is a phrase that appears constantly in management reporting and is rarely unpacked into what it actually involves. This is that, step by step.

The short answer

Redesign means measuring the current process precisely before choosing any tool, finding its real constraint, and asking which stage is no longer necessary given information now available. The intelligent tool is selected at the end of that work, not at the start. In McKinsey's 2026 global survey, that ordering was the strongest predictor of financial impact.

Step one: pick a process, not a department

"Making the sales department intelligent" is not a project; it is a territory. "Telephone order to issued invoice" is a process: defined start, defined end, and countable.

Three criteria for the first one: it repeats, it is measurable today, and someone owns it who has the authority to change it. The third defeats more projects than the other two combined.

Step two: measure the current state before changing anything

This is the most-skipped step and the most expensive to skip. Without a baseline, no claim of improvement is provable at the end.

The minimum to record:

  • Cases per week
  • Total elapsed time from start to finish, not just hands-on time
  • Waiting time between stages, recorded separately
  • Error or rework rate
  • Person-hours the stage consumes

Two weeks of data is enough. One week usually is not representative, because there is always a week busier than the others.

Step three: find the constraint, not the busiest stage

The distinction is subtle and decisive. The busiest stage occupies the most people; the constraint is what makes the whole process slow. They are usually not the same stage.

In administrative processes the constraint is almost always waiting rather than work: waiting for an approval, for a customer reply, for data from another unit. When total waiting exceeds total hands-on time — which is common — speeding up the hands-on part produces no visible change.

Step four: ask which stage is no longer needed

This is where redesign actually happens, and where the decision is managerial rather than technical.

Many approval and review stages exist to compensate for information that was not available. If that information is now available continuously, the question becomes whether the stage is still required or can become a sampled review instead.

Three patterns that produce the most effect in practice:

  • Turning hundred-percent review into risk-based review — ordinary cases proceed automatically and only doubtful ones reach a human
  • Removing manual re-keying between two systems — wherever someone reads one screen and types into another, it costs time and manufactures errors
  • Moving a decision earlier — a large share of rework comes from defects discovered at the end of the process

Step five: only now choose the tool

Only at this point is it clear what is needed: recognition from images, prediction from data, automated answering, or none of them. A substantial share of redesigns need no model at all and are resolved by reordering the stages.

If automated answering is needed, the architecture is a separate decision, examined in language model or retrieval from your own knowledge. And if the tool depends on an external service, define its fallback from the outset — the reasoning is in AI service outages and business continuity.

Step six: read the step-two measures again

Thirty to sixty days after deployment, measure the same numbers the same way. If they have not moved, the project does not scale; you return to step three, because the constraint was probably misidentified.

That simple discipline is what separates the successful organisations from the rest — the figures behind it are in why an AI budget does not turn into profit.

Running this in Karaj and Alborz

For a manufacturer in Karaj or a service operation elsewhere in Alborz province, steps two and three cannot be done remotely. Real waiting time is not recorded on any process diagram; it lives in the memory of the person who does the work every day.

Our method is two on-site sessions: one beside the work to see what actually happens, and one with a manager who has the authority to change the process. Without the second, the output degrades into "adding a tool".

Frequently asked questions

How long does a workflow redesign take?

Two weeks for the baseline, typically one to two weeks for analysis and design, and thirty to sixty days after deployment to measure the result. Deployment time itself depends on scope.

Should we redesign every process at once?

No, and this is among the most common mistakes. One process, end to end, with a number. The experience gained makes the second process several times faster.

Who should own the project?

A manager with the authority to change the process itself. If the project lives only in IT, that department cannot remove an approval stage and is left with adding tools.

What if staff resist?

Resistance usually comes from ambiguity rather than opposition. In practice: include the people who do the work in step three, because they know the constraint better than anyone. Training the team is part of delivery, not something for afterwards.

How do we know we identified the right constraint?

A simple test: if that stage were halved, would total elapsed time fall noticeably? If not, it is not the constraint.

Does this only apply to administrative processes?

No. The same six steps work for a production line, a warehouse, customer support and training operations. What changes is the measure, not the order of the steps.

Related services from Qatreh