I want to open today with the most suspicious document in corporate history. It’s not a contract. It’s not an NDA. It’s a timesheet. Specifically, it’s the timesheet you get back whenever you ask employees to measure their own work. And it is a thing of beauty.
Every task takes a lovely round number – thirty minutes, one hour, two hours. No task in recorded history has ever taken seventeen minutes. And every single day – every day! – adds up to exactly eight hours. Not seven and a half. Not eight and a quarter. Eight. Thousands of rows, and mathematics has never been so cooperative.
Nobody works like that. But everybody reports like that.
Now if you’re planning to roll out process intelligence in your company, you’ve probably already had The Meeting. The one where somebody leans back, crosses their arms, and asks: “So… we’re going to spy on people?” And I know what the standard playbook whispers in your ear at that moment: keep it quiet. Run a stealth pilot. Handle the communication later, once you have results to show.
Today I want to convince you that this instinct – data first, trust later – is the single most expensive mistake you can make with this technology. Because fear doesn’t just hurt morale. Fear corrupts the data itself. Scared people produce fake numbers and then you make real decisions based on them.
By the end of this episode, you’ll know the five configuration decisions that answer the question “what exactly are you collecting?” – honestly, out loud, and before the agent lands on a single laptop.
Introduction
Konichiwa! Welcome to the AI Automation Dojo. I’m your host, Andrzej Kinastowski, one of the founders of Office Samurai, where we know that a timesheet that adds up to exactly eight hours isn’t data. It’s fiction with a total row.
So whether you’re an automation lead whose process intelligence pilot has been stuck in legal review since spring, or an employee who just heard there’s an “agent” coming to their laptop and quietly opened LinkedIn in another tab – you’re in the right place.
Now grab your favorite katana – or your suspiciously tidy timesheet – and let’s get to it!
Why traditional time tracking always fails
Today’s story comes from Zuzanna Pamuła, who runs our process improvement and business analysis practice – a certified Lean Six Sigma Black Belt who, before joining Office Samurai, spent twelve years inside the shared services centers of big international corporations, measuring processes for a living.
Which also means: she spent twelve years watching measurement fail. Not once. Systematically. In three escalating acts.
Act one: the stopwatch era. Picture a young analyst – fresh black belt, full of Lean enthusiasm – literally sitting next to an employee with a notebook on her lap, a pen in one hand and a stopwatch in the other. No laptop yet. The methodology: watch, click, scribble and then spend the evening transferring your notes into something digital before they get tangled up with the three other observations you did that day. And it works, sort of for one desk. Try scaling it to a two-hundred-person operations floor and you’ll discover you need a second career just to hold the stopwatch. And there’s a subtler problem: people behave beautifully when a black belt is sitting next to them. You’re not measuring the process. You’re measuring the performance of the process. Dinner-with-the-in-laws behavior.
Act two: delegation. Fine – we can’t watch everyone, so let’s have people report their own time. The team built what Zuzanna describes as a bulletproof Excel template. Dropdowns, formulas, locked formatting, the works. You may recognize this as a close cousin of the Excel Trap we’ve talked about on this show before. Two things happened. First, employees demonstrated a truly inspiring creativity in destroying preset formulas which cost Zuzanna hours, sometimes days, of soul-sucking cleanup just to make the inputs usable again. And second… the numbers came back round. All of them. Every task a clean half hour or hour. Every day summing to exactly eight hours. Zuzanna’s comment on this, and I quote: “Amazing, right?”
Act three: automation or at least, 2010s-flavored automation. Her manager had an idea: an Excel tool with VBA macros. You pick your task from a dropdown, you click Start, you click Stop, and the record writes itself. No manual entry, no rounding, times corresponding to reality. In theory – brilliant. In practice, they met the human factor: people forgot to click Start. Or Stop. Or both. And so the Process Excellence team – the people whose entire job was eliminating waste – ended up running a hidden factory whose only product was manually repairing broken time records.
The measurement of waste, generating its own waste. I’d call it ironic if it weren’t so universal.
Now, here’s the part that actually matters. Three methods, three failures and the conclusion Zuzanna nearly drew, the mantra she heard repeated for twelve straight years: business processes just can’t be measured. Every case is unique. People are messy. Forget it.
That conclusion is wrong and her own diagnosis explains why. The reports were based on how people said they worked, not how they actually worked. And the gap wasn’t laziness, and it wasn’t sabotage. People simply didn’t know how the results would be used, or what the consequences would be – for them. So they did the entirely rational thing: they reported the official version of their job. The SOP version. The version that’s safe to show a manager. Zuzanna puts it in one sentence I want you to keep: when people work inefficiently, it’s almost never bad will. It’s bad tools, badly designed processes, or poor knowledge transfer. But if people believe the stopwatch is looking for someone to blame, they will protect themselves and your data pays the price.
And I can hear some of you already: “Andrzej, this is ancient history. The software solved this. The sensor doesn’t forget to click Start.”
True. The modern agent doesn’t round to the nearest half hour, doesn’t skip lunch in the logs, and doesn’t need an evening dedicated to un-ruining a spreadsheet. But fear didn’t retire when the stopwatch did – it just bought new equipment. Mouse jigglers. Auto-clickers. The laptop mouse taped to a Roomba, sweeping the apartment and keeping the status light green. And yes – pattern-based tools see through most of that theater, because a thousand clicks that produce no outcome look exactly like what they are. But that’s not the point. The point is: if your people are scared enough to perform for the measurement, then congratulations – you have spent enterprise money to digitize the same old fiction, just in higher resolution. Garbage in, and – you know the rule by now – garbage clogs the pipelines.
So the privacy conversation – “what exactly are you collecting?” – is not the HR garnish on your rollout plan. It is a data-quality requirement. Every question you leave unanswered turns into fear. Fear turns into changed behavior. Changed behavior turns into numbers you cannot trust. Which is why you answer it out loud, before anyone installs anything.
And here’s the good news: “what are you collecting?” isn’t really one question. It’s five decisions. And every single one of them belongs to you, not to the vendor.
Let’s take them one by one.
5 configuration decisions for Process Intelligence
Here’s the uncomfortable truth about the question “what exactly are you collecting?”: the honest answer is “whatever we decide.” These platforms are configurable, which is precisely why the question scares people, and precisely why you can’t answer it with a vendor brochure. You answer it with decisions. There are five of them.
Decision one: content, or metadata.
Let’s start with full disclosure, because this is the Dojo and we don’t do fine print. Yes, tools in this category can capture keystrokes. They can capture screen content. They can take screenshots. If you switch it on, the machine will happily oblige. The vendor puts the power in your hands, which means the responsibility is in your hands too.
Our advice, after more than a dozen deployments we’ve run: don’t switch it on.
Not because it’s illegal – in many setups it can be made legal. Because you don’t need it, and the trust it costs you is wildly out of proportion to what it buys you.
I know what you’re thinking: “But if we’re not reading screens, how do we know what people are actually doing?” Here’s the thing, the findings that actually make money live at the metadata level. Application-level data: which tools are in use, for how long, in what sequence. Remember the big three patterns from our opportunity-identification episode? The copy-paste table. Application tandems – that four-minutes-in-Oracle, jump-to-SAP dance, fifty times a day. Manual text entry volumes. Every one of them is visible without reading a single word of anyone’s screen. Our favorite example: the insurance company where the copy-paste table exposed people frantically pasting from a sticky-notes app into the ERP, because the ERP was too slow to type into during customer calls. We didn’t need screenshots to find that. We needed the pattern.
Think of it like a building’s badge system. It knows you were in the building. It doesn’t know what you whispered in the hallway. That’s the level you want, because that’s the level where process problems live.
Decision two: whitelist, not blacklist.
When we had Adam Bujak from KYP on the show, he said the thing most people get exactly backwards, so let me repeat it: at the beginning, the tool sees nothing. Nothing. You then decide what gets whitelisted – SAP goes in, Excel goes in, Outlook goes in, because that’s where the process lives. What never goes on the list – your bank, your private email, the fantasy football league you are definitely, definitely not managing during work hours – doesn’t get “collected but tactfully ignored.” It produces zero data. There is no file. There is nothing to leak, nothing to subpoena, nothing for a curious manager to peek at. And crucially: the whitelist isn’t a secret. Publish it. When people can read exactly which applications are in scope, the question “are you watching me?” gets a boring, verifiable answer and boring, verifiable answers are what kill fear.
Decision three: names, or patterns – the resolution question.
Now, here’s where you probably expect me to say: “anonymize everything, aggregate everything, and you’re safe.” A lot of consultants will tell you exactly that, because it’s the comfortable answer. It’s also lazy. Because out of the five decisions, this is the one where I won’t give you a default – the right resolution of your data depends entirely on what you’re trying to achieve, and on what your people are ready to sign up for.
If your goal is process improvement – finding the copy-paste tables, unclogging the bottlenecks – then yes: anonymize at the source and aggregate at team level. Your analyst sees agent 007, not Grażyna from accounting; you learn what happens inside the team, not inside one person’s Tuesday. For pure process work, individual identities are noise. Don’t collect resolution you have no use for.
But. We have a customer whose main reason for deploying the platform was to build an incentive program. Every employee doing transactional work sees their own dashboard, knows exactly how they’re being evaluated and the bonuses follow the data. You cannot pay a bonus to agent 007. That program requires individual-level data, with names on it. And it works, because the deal is explicit and people signed up for it: my numbers, my dashboard, my bonus.
And there’s a second case for names, one I care about a lot: protecting the quiet heroes. In every operation there’s somebody carrying a workload nobody sees – the person quietly absorbing volume, working long and intense, invisible to management-by-walking-around because they’re too busy to perform busyness. Aggregate data can tell you “this team is overloaded.” It cannot tell you “this specific person is drowning and needs relief before they resign.” If you want the measurement to defend individuals, the measurement has to see individuals.
So decision three isn’t “anonymization: yes or no.” It’s: what resolution does your goal actually require and is that deal on the table, openly, before you collect a single click? Anonymization is not a virtue. Names are not a crime. Resolution follows purpose and the purpose is agreed with the people in the data. And whatever you decide, role-based access still applies: the resolution you collect is not the resolution everyone gets to see. That part isn’t a philosophical debate, it’s a configuration screen.
Decision four: snapshots, or patterns over time.
Everyone has slow days. I’m recording this on one of them. Configure for patterns, and the slow Tuesday simply doesn’t exist as a finding, a sustained bottleneck does. This matters more than it sounds, because it flips the most feared conversation in the building. When the dashboard shows a team at, say, forty percent passive time, the rookie reading is “these people are slacking.” The correct reading – and what the data almost always shows on inspection – is waiting: for approvals, for the system to unfreeze, for a handoff that arrives late. That’s not a people problem. That’s a process design problem. And the conversation changes from “why were you idle?” to “what’s blocking you?” – which, you may notice, is the first genuinely useful question anyone has asked that team in years.
Decision five – this one, you get to guess.
Let’s play. In our experience, one single decision defuses more resistance than everything else combined. It’s not anonymization. It’s not the whitelist. It’s not even giving the works council a seat at the table — though do that too. I’ll give you a second.
Decision five is who gets measured and the answer is: management gets the agent too.
If leadership deploys the sensor on everyone below a certain pay grade and exempts itself, you have just confirmed, in writing, every suspicion in the room: this is a tool pointed down. But run it on the whole org chart and two things happen. First, the message becomes credible, we’re measuring the process, and management is part of the process. Second – and this is the part executives don’t expect – the most valuable findings are often up there. Meeting overload, Zuzanna calls it meetingorrhea, and I refuse to use a better word because there isn’t one. Decision bottlenecks. Approval queues where a process sits for four days waiting for one signature. Those are leadership diseases, and they’re invisible until the sensor points in both directions. And if the steering committee refuses to be measured? Write that down. It’s your first finding.
So: metadata, not content. Whitelist, not blacklist. Resolution that follows purpose. Trends, not snapshots. Everyone, not just the people who can’t say no.
Notice something about these five decisions: not one of them is technical. The vendor can execute any combination you pick, the configuration screen is really a values screen. And that’s exactly why “what are you collecting?” deserves an answer from you, not from a datasheet.
But even when you get all five right, there’s a deeper question underneath, the one people are really asking when they ask about collection: “Fine. But what are you going to DO with it?” Hold that question. We’ll deal with it in a minute and then properly, because it deserves more than a minute. First, let me show you what you get for making these five decisions out loud. Because the strangest thing about doing this right isn’t that the resistance fades. It’s that the resistance switches sides.
Turning employee resistance into process support
We’ve seen this moment in our deployments, and it never stops being satisfying. The scope is published, people get access to their own data and within days, somebody points at their screen and says: “Look. THIS. This is that stupid, repetitive work I’ve been complaining about for months. Can we finally get rid of it?” For years, that complaint was an opinion, one voice against a manager’s gut feeling. Now it’s a number. The tool they were warned about becomes the first witness they’ve ever had.
And remember Zuzanna’s sentence from the beginning, bad tools, bad processes, poor knowledge transfer, almost never bad will? There’s an uncomfortable symmetry hiding in it: when the real data finally arrives, it indicts management’s choices far more often than employees’ effort. The people who should be nervous about measurement are usually not the ones who are.
Now, one question is still hanging in the air, and I haven’t forgotten it. Even with a perfect scope, somebody – probably your works council, doing their job well – will ask: “Fine. Metadata, team-level, transparent. But what are you going to DO with it? Performance reviews? Restructuring?” And that is the right question. It’s also a different question – rules of use, not rules of collection – and it deserves a full episode, not a footnote. It’s going to get one, soon. For today, let’s close the loop we opened.
3-sentence summary for your steering committee
So let me say the whole episode in three sentences – the version you can quote at your steering committee.
One: fear corrupts your data before it corrupts your culture – scared people produce fake numbers, whether the sensor is a stopwatch or an enterprise agent. Two: the privacy conversation is not the obstacle standing between you and the measurement – it’s the thing standing between you and measurement that’s actually true. And three: “what exactly are you collecting?” is not a vendor question. It’s five decisions that belong to you – made out loud, before the agent lands on a single laptop.
5-sentence rollout action plan
Your homework – small enough to actually happen. Tomorrow, before you book a single vendor demo, take today’s five decisions and write one sentence for each. Content or metadata. What’s on the whitelist. What resolution, and for what purpose. Patterns over what period. And who’s in scope – all the way up.
Five sentences. If you can write them – congratulations: you’re holding your rollout communication plan, and it took ten minutes. If you can’t – you’re not ready to deploy, and no tool on the market will fix that for you.
And if a sixth sentence is itching to be written – “this data will never be used for…” – good instinct. Save it. That’s a whole episode of its own, and it’s on its way.
