engineering notes
How AI compresses operational processes
How AI removes intermediate steps, approvals and waits in operational processes and what it gives in cycle time.
In brief for executives. AI’s effect in operations is not “minus headcount” but a shorter cycle and fewer approvals. Most working time is eaten not by the work but by its coordination: waits, handoffs, reconciliations. AI compresses exactly that part. The right effect metric is cycle time and the number of approvals before and after, not “how many people were cut”.
When people say “AI will raise operational efficiency”, they usually picture replacing a human on a step. In practice the main effect comes from elsewhere — from compressing the gaps between steps, not from speeding up the steps themselves.
AI cuts not people, but the waits between steps.
Hypothesis: AI cuts steps and waits, not people
In an operational process the actual work takes the smaller share of time. The larger one is handoffs between people, waiting for a response, status reconciliation, re-gathering context. The AI layer removes exactly these gaps: the process compresses not because steps got faster but because time stopped being lost between them.
More than half the time goes not to the work itself but to coordinating it. That is the layer an AI environment compresses — fewer approvals and handoffs, not «replacing people».
More than half the time is coordination, not creation. That is what gets compressed.
Problem: long chains of approvals and waits
A typical operational process is not “work” but a queue of waits: a request waits for classification, classification waits for an owner, the owner waits for data from another system, the result waits for approval. The sum of waits is many times the sum of work. Automating one step here gives almost nothing — the bottleneck is in the gaps.
Why the usual approaches don’t work
“Speed up a step with AI” doesn’t help if 80% of cycle time is waits between steps, not the step itself.
“Add people” doesn’t help: more people means more handoffs and approvals, coordination load grows faster than throughput.
“Put a bot on this section” doesn’t help if the process isn’t rebuilt: the bot closes a step, but the chain of waits around it remains.
Engineering model: rebuilding the process around AI
Removing handoffs. Classification, routing, gathering data from systems are done automatically and immediately — the wait of “who will take this” and “send the data” disappears.
Reconciling context. The needed information is gathered to the step automatically, not requested by people in circles — the re-gathering of context disappears.
Parallelism instead of a queue. Events let independent things be done in parallel rather than lined up into a queue of waits.
A human on expensive decisions. People stay where the error cost is high — but with context already gathered, without preparatory waits.
The key: the effect comes from rebuilding the process around AI, not inserting AI into the old process. An automated step in an unchanged chain of waits barely moves cycle time.
Impact comes not «a little everywhere» but concentrated in a few functions with repeated operations. That is the hint of which process to automate first.
That is why the effect concentrates in a few operational functions with repeated handoffs — where there is something to compress.
Practical takeaway for business
Measure cycle time and number of approvals, not “how many steps were automated”. The effect is “the process took N days, now it takes M” and “there were K approvals, now L”, not “AI adopted on a section”.
Be ready to rebuild the process, not decorate the old one. The greatest effect is where there are the most waits between steps; exactly those gaps must be removed, not a single step sped up.
The goal is to scale operations without a linear headcount increase. Compressing the cycle lets the same staff do more; that is a different framing than “cut people”, and a more robust one.
Apply this to your processes — .
Open questions
Where the limit of compression without decision-quality loss lies depends on the process and is checked on the pilot. Which approvals are truly needed and which are a historical ritual is a process-review question, not a technology one. How to count the effect when the gain is removed coordination is a cycle-time model, not an industry standard.
If in your process there is more waiting than work between steps — it can be compressed. — we’ll work out where time goes and what is removed first.