METHOD · 6 MIN READ
One bottleneck, one number: why focused AI beats the big programme
How picking a single constraint gets a measurable AI result inside a quarter, while broad transformation programmes stall.
LAUNCH AGENTS ·
Why big AI programmes stall
Most enterprise AI programmes start with a long list. Dozens of use cases are scored, a roadmap is drawn, and a platform is chosen. Twelve months later the organisation has pilots in many functions and a result in none. The problem is rarely the technology. It is that effort is spread across work that was never slowing the business down.
A workflow behaves like a pipe with one narrow section. Speeding up any other section does nothing to the flow. Automating the intake of invoices is impressive, but if invoices then wait four days for an approver, the business pays suppliers no faster than before.
Every workflow has one constraint
The Theory of Constraints, developed for factory operations, applies cleanly to knowledge work. In any process there is one step whose capacity limits the output of the whole. In quoting, it is often the senior engineer who signs off every bid. In finance, it is the reconciliation that only two people know how to do. In customer service, it is the specialist queue behind the front line.
Finding that step takes days, not months. You walk the workflow with the people who run it, watch where work piles up, and ask where they wait. The constraint is usually obvious to the team and invisible to leadership.
What focus changes
- Speed. One workflow can be redesigned and put live with agents in about 45 days, because the scope is narrow and the people involved are few.
- Proof. A single agreed metric, such as quote turnaround or days to close, shows clearly whether the change worked.
- Adoption. One team changes how it works at a time, so change management is concrete rather than abstract.
- Compounding. When the constraint is lifted, it moves somewhere else. The next cycle starts from a stronger base.
How to run the first cycle
Pick a function where leadership already watches a number and is unhappy with it. Spend 10 to 15 days mapping how work, information and cash actually move. Name the constraint. Redesign the flow so that the constraint only does the work that truly needs it, then build agents to carry the rest. Agree the metric and its baseline before you build, and review it on a fixed cadence after go-live.
Then do it again. One bottleneck, one number, then the next. That is how AI becomes part of the operating rhythm instead of a programme on a slide.
Key takeaways
- Every workflow has one constraint that sets its pace; improving anything else does not move the result.
- A single function can be redesigned and running with agents in about 45 days.
- Agree one metric and its baseline before building, then review it on a fixed cadence.
- When the constraint lifts, it moves. Repeat the cycle on the next one.
Frequently asked questions
What is a workflow bottleneck?
A bottleneck, or constraint, is the single step in a workflow whose limited capacity sets the pace of the whole process. Work piles up in front of it and everything after it waits.
How long does a focused AI project take?
A focused cycle typically takes about 60 days: 10 to 15 days of discovery to find the constraint, then 45 days to redesign the workflow and put agents live.
Why not run many AI use cases at once?
Spreading effort across many steps that are not limiting produces activity without a measurable result. Focusing on the constraint is what moves the business number.