Where to Start Automating in an Asset-Heavy Operation

Every automation program I have watched fail started the same way: someone opened a workshop by asking the team to "list the quick wins." Two hours later there is a whiteboard full of sticky notes, everyone feels productive, and the process that gets picked is whichever one the loudest person in the room complains about most. That is not prioritisation. It is theatre. In an asset-heavy operation, where work orders, permits, requisitions and meter data all cross several systems, a bad first choice does not just waste a few weeks. It burns your credibility, and credibility is the only budget you actually get to spend twice.

Why "find the quick wins" is bad advice

"Quick win" is a feeling, not a measurement. It rewards the process that is most annoying, most visible, or most recently blamed for a missed SLA. None of those tell you whether the process is a good candidate. A task can be painful and still be a terrible first automation, because pain often comes from a broken input, a disputed approval chain, or two systems that disagree about the truth. Automate on top of that and you have built a faster way to produce the wrong answer.

The alternative is not complicated. Score every candidate the same way, on the same axes, and let the numbers argue. You are not trying to be scientific to three decimal places. You are trying to replace "who shouted loudest" with a defensible ranking you can put in front of a steering committee. That is the whole job of a prioritisation model.

Score every process on five things

List your candidate processes, then score each one from 1 to 5 on the five dimensions below. Be honest, and score with the people who actually run the process, not the managers who think they know how it runs.

• Volume. How often does this run? Daily meter reads score high, a monthly report scores low. High volume multiplies every other benefit, but see the warning further down.
• Touch time. How much human handling does one instance take? A permit that needs three sign-offs and manual data re-keying scores high. A one-click approval scores low.
• Error rate. How often does a human get it wrong, or produce something that has to be corrected downstream? Manual data entry into two systems is a classic error factory.
• Rework cost. When it goes wrong, how expensive is the cleanup? A mismatched invoice can freeze a payment run. A wrong meter read gets quietly overwritten next month. Same error rate, very different cost.
• Systems crossed. How many systems does one instance touch? This one is double-edged. More systems means more value from removing the manual bridge, but also more fragility, because every integration point is a place the data can disagree.

Add the first four into a single Value score. Treat Systems crossed as your primary Fragility signal, adjusted up when the data those systems hold is disputed or duplicated. Now you have two numbers per process, and two numbers is all a matrix needs.

The value-versus-fragility matrix

Plot value up the left and fragility across the bottom. Four quadrants fall out, and each one comes with a different instruction.

The trap quadrant is the top right: high value, high fragility. These are the processes everyone wants automated because they hurt, but they hurt precisely because the data underneath is a mess. Your instinct will be to start there. Resist it. The correct move for the top right is to fix the foundation first, then automate. The correct place to begin is the top left.

Eight facilities processes, scored

Here are eight processes you will find in almost any facilities operation, scored on the model above. Value is the sum of volume, touch time, error rate and rework cost (out of 20). Fragility is scored out of 5. The recommendation follows from where each lands on the matrix, not from how loudly anyone complains about it.

Notice what the scores did. The two processes that scream loudest, invoice matching and work request intake, both landed in "fix first," not "start here." The winners are PPM scheduling and meter reading: mid-volume, single owner, clean data source. That is not a coincidence. It is the pattern.

Why the busiest process is the wrong place to start

The highest-volume process looks like the obvious first target. It is not, for two reasons that have nothing to do with technology.

First, the busiest process is the most politically defended. High volume means lots of people, several team leaders, and at least one manager whose headcount is justified by it. Touch that process and you are not automating a workflow, you are threatening an org chart. Every exception becomes a reason to pause, every edge case becomes a committee. You will spend your first ninety days in meetings instead of shipping.

Second, high volume usually means high fragility. Processes get busy because they sit at the centre of the operation, which means they cross the most systems and inherit the most disputed data. Work request intake is the perfect example: enormous volume, and precisely because of that volume it has grown five informal channels that nobody owns end to end.

PPM scheduling wins for exactly this reason. It is important without being political, it draws from a single authoritative asset register, and one planner owns the outcome. You can automate it, prove the model, and produce a clean before-and-after number that funds the next wave. For how to build that number credibly, see my note on measuring automation ROI honestly.

The "do not automate this yet" list

Three categories of process should be kept out of your first wave no matter how good the value score looks. Automating any of them just encodes a problem you have not solved.

• Anything with an undefined approval matrix. If you cannot write down, on one page, who approves what and at which value threshold, you cannot automate the approval. Permit to work and requisition approval both fail here until the matrix is agreed and signed. Automating a workflow whose rules are still being argued about just hard-codes today's argument.
• Anything where two systems disagree on master data. If your EAM says an asset is decommissioned and finance still depreciates it, automating across that gap produces confident, fast, wrong output. Invoice matching sits here. Reconcile the master data first, then automate. This is the failure mode I see most in asset-heavy shops, because the systems were bought at different times by different teams.
• Anything a manager is currently doing to compensate for a broken upstream process. When someone is manually massaging data every morning to cover for a bad feed, that manual step is a symptom, not a task. Automate it and you make the upstream breakage permanent and invisible. Fix the upstream feed instead, and the compensating work disappears on its own.

The common thread: never automate over an unresolved disagreement, whether the disagreement is about who approves, what the data says, or why the manual patch exists in the first place. Where those problems are absent, tools like Microsoft Power Automate make the build genuinely quick. Where they are present, no tool saves you.

A 90-day first wave

Here is how I sequence the first ninety days once the matrix has named the winners. The point is to finish one thing, prove it, and earn the mandate for the fragile-but-valuable processes waiting in the top-right quadrant.

That is the whole discipline. You do not need a hyperautomation platform or a center of excellence to start, though it is worth understanding where the market is heading; the Gartner hyperautomation glossary is a reasonable primer on the vocabulary vendors will use on you. What you need is one honest score, one uncontested win, and the patience to fix foundations before you build on them. If you want to see where a first wave sits in the bigger picture, my automation maturity model maps the stages, and the workflow automation buyer's guide covers tool selection once you know what you are automating.

Where this leaves you

Stop hunting for quick wins. Score every candidate on volume, touch time, error rate, rework cost and systems crossed, collapse that into value and fragility, and plot it. Start in the top-left quadrant: the mid-volume, single-owner, clean-data process that nobody will fight you over. Keep the politically defended giants and the disputed-data processes out of wave one, no matter how much they hurt. Finish one thing, measure it honestly, and let that win pay for the next. In an asset-heavy operation, the order you automate in matters more than the tools you automate with.

A note on independence: I do not resell, and I am not paid by any of the vendors mentioned here. The scoring model is my own, drawn from delivering automation across ERP, EAM and CAFM environments. Treat the numbers in the worked table as illustrative; run the scoring yourself, with your own people, on your own processes.