Finding The Work

Finding the work

Do not hold a workshop. Walk the floor.

---

The fastest way to pick the wrong project

Lock a room full of executives together and ask them to brainstorm automation opportunities.

They will pitch you their most ambitious problems, not their most solvable ones. Everyone in the room is incentivised to name something impressive, and nobody in the room does the work.

The alternative costs less and works better: go and watch.

The high-value targets are usually hiding in plain sight. The analyst who copies data between three systems every morning. The specialist who checks the same external portal eleven times a day. The person maintaining a spreadsheet nobody else can read, which four downstream processes depend on.

None of those get pitched in a workshop, because none of them sound like a strategy.

The Sigh Test

Watch the hands. Listen for the sigh.

The sigh is a real signal and it is more reliable than a process map, because a process map records what is supposed to happen and a sigh records what does.

What you are looking for: a task the person performing it visibly resents, that requires real cognitive effort, and that produces nothing anybody values. That combination is rare, obvious in person, and invisible on a slide.

The opportunity score

Every candidate is scored before it earns a pilot slot. This exists to stop the loudest voice in the room picking the hardest problem as the first project.

Opportunity Score = (Volume × Empathy) ÷ Complexity

	What it measures	Low end	High end
Volume	How often does this happen?	Rarely	Hundreds of times a day
Empathy	The Sigh Test. Does the person doing it hate it?	They find it meaningful	They are quitting over it
Complexity	How clear is the procedure? Could you write it on a napkin?	Rule-based, clear if-then logic	Expert judgement, no consistent rules

A qualifying threshold sits in the low double digits on a five-point-per-factor scale. Calibrate it once against work you already know, then leave it alone: a threshold that moves per candidate is a negotiation, not a filter.

An instant disqualifier, regardless of score: anything touching consequential unstructured judgement without a qualified human review loop.

Why Empathy is in the numerator

It is the unusual term and it is the one that earns the formula its place.

Volume and complexity are the obvious axes and most prioritisation stops there. That gives you a list of things that are frequent and easy, which is also a list of things nobody minds doing.

Empathy adds the human signal, and it predicts two things nothing else does. Adoption, because people help a system that removes work they hate. And honesty during discovery, because someone who wants the task gone will tell you about the exceptions, the workarounds and the reason step four exists. Someone indifferent will describe the happy path.

This is the same mechanism as the dignity clause, arriving one stage earlier. The people whose work this touches are either helping you or quietly not, and Empathy is the term that notices which.

Building your own heatmap

The source gives a populated value-chain table for one industry. A borrowed heatmap is somebody else's operation. The method transfers; the table does not.

One. List your value chains. Four to eight, at the level a leader would recognise.

Two. Score each on four axes, coarsely: volume of repetitive work · potential for judgement-based automation · regulatory sensitivity · human impact. High, medium, low is enough.

Three. The candidates are where volume is high, regulatory sensitivity is not critical, and human impact is high. That last column is the one people leave out and it is the one that decides whether it gets used.

Four. Start there and walk the floor. The heatmap tells you which floor to walk, not what to build.

A worked example, in logistics. Order intake scores high on repetitive volume and low on regulatory sensitivity. Customs documentation scores high on both, which makes it a later project rather than a first one. Exception handling in delivery scores medium on volume and very high on human impact, because it is the work the team most dislikes. The first candidate is order intake; the one worth doing second is exceptions, and it will produce more goodwill than either of the others.

Instrumenting the walk

Watching does not scale past a handful of people, and it does not have to.

Task capture on a small sample of workstations in the target area compresses weeks of discovery workshops into a couple of weeks of observation, and surfaces the highest-frequency repeated sequences without anyone having to remember them.

Two cautions. It records what people do, not why, so it finds sequences and misses reasons. And it is surveillance if it is not consented to, which will cost you more in withheld cooperation than the discovery is worth. Ask first, share the output with the people it observed, and use it to prompt the conversation rather than to replace it.

How this relates to BXT

They are different instruments at different moments and they are often conflated.

	Answers	When
Opportunity score	Is there work here worth automating?	Before there is a candidate. It finds candidates
BXT	Should we fund this candidate?	Once one exists. It ranks and gates them

Running BXT without ever running the opportunity score means scoring a list somebody guessed. The Select stage of the spine needs both: one produces the list, the other chooses from it.

_{Adapted from the AI CoE and Agent Factory Playbook, based on The Augmented Enterprise framework. The heatmap method and the worked example are the author's own; the source's single-vertical table is not reproduced.}

_{Back to Home · The three proof gates · The framework spine}