Introduction
Konnichiwa! Welcome to the AI Automation Dojo. Today we’re working a corporate murder mystery: why do nine out of ten AI transformations quietly die, when the technology is sitting right there with a rock-solid alibi? The culprit, it turns out, is never the robot. It’s the comfortable decision – the safe-looking choice that feels wonderful in the steering committee and then strangles the whole project in the parking lot.
I’m your host, Andrzej Kinastowski, one of the founders of Office Samurai – where we’ve noticed that in transformation, the comfortable option and the correct option almost never RSVP to the same meeting.
So whether you’re a director tired of safe choices that go absolutely nowhere, or a manager who’s quietly aware that your single most important process lives in one spreadsheet, maintained by one person, who is on holiday this week, you’re in the right place.
Now grab your favorite katana – or that one mission-critical Excel file – and let’s get to it!
Why AI transformations quietly stall
Let’s start with a number that should be ruining someone’s sleep.
Depending on which gloomy report you pick up, somewhere between ninety and ninety-five percent of enterprise AI projects never make it out of the pilot stage. Let that land. Nine out of ten. And here’s the thing – they don’t fail dramatically. There’s no explosion. Nobody gets escorted out of the building. They just… idle. Indefinitely. A gorgeous little demo that gets wheeled out once a quarter to impress the board, running quietly on a server everyone is now afraid to switch off, because eighteen months ago someone important pointed at it and said the word “future”.
Now, when something fails nine times out of ten, your instinct is that the cause must be exotic. Bleeding-edge. Some impossibly hard technical problem that only three people on Earth understand.
It is not. I have sat in a lot of these post-mortems, and I can tell you the cause is almost never the technology. The models work. The platforms work. The robots, bless them, do exactly what you tell them to.
The failure is upstream. And I think it comes down to a single misread word.
We call it “AI transformation” and everyone fixates on the first word. The shiny one. The one you can put in a press release. But the load-bearing word in that phrase was always the second one – transformation – and underneath that, quietly holding up the entire building, is a word so unglamorous nobody wants to say it on stage: process.
The most important word in “AI transformation” is “process”. And companies treat it like the boring adjective instead of the actual noun.
So let me make sure I’ve got the standard playbook straight, because I want to be fair to it. We have a core business process. It is held together with tape, tribal knowledge, four hundred undocumented exceptions, and one woman in accounting named Grażyna who is the only living human who understands how any of it actually works. And the master plan… is to make that process faster. With AI.
We’re not fixing the leak. We’re buying a much more powerful pump. And then we’re standing around genuinely puzzled about why the basement is filling up quicker than before.
That’s failure number zero, the one all the others grow out of: automating a broken process doesn’t fix it. It just industrializes the chaos. You take a mess that used to happen at human speed, and now it happens at the speed of light, in the cloud, with an audit trail. Congratulations. You haven’t solved the problem. You’ve scaled it.
The 7 comfortable lies
Here’s the pattern I’ve noticed across basically every transformation that face-planted. At each fork in the road, the team had two options: the hard one that was correct, and the comfortable one that felt safe. And they reached for comfortable. Every time.
That’s the whole tragedy. In transformation, “comfortable” and “correct” almost never live at the same address. So let me walk you through the seven most comfortable lies people tell themselves on the way down. See how many you recognize. No judgment. Okay, a little judgment.
01. Lie number one: “The dashboard knows what’s happening”
Somebody bought a process-mining tool, it produced a beautiful diagram, and now the whole strategy is built on it. Here’s the problem. Process mining can only see work that leaves a digital footprint in your big logged systems – your ERP, your CRM. The second a human being alt-tabs into their email, or a chat window, or – and I promise we’ll come back to this – a spreadsheet, that work vanishes. It’s gone. The dashboard has no idea it happened.
It’s the old story about the man looking for his car keys under the streetlight. A police officer comes to help, and after twenty minutes asks: “Are you sure you dropped them here?”. The guy says, “No, I dropped them over there in the park, but the light’s much better over here.” Your process-mining dashboard is the streetlight. It is showing you, in stunning high definition, the one corner of the process that happens to be well-lit. Meanwhile the actual disaster is happening over in the dark, in Outlook, where the camera doesn’t point.
You bought a security system, and it films the lobby beautifully. The heist is taking place in the stairwell.
Lie number two: “We need to hire an AI team”
Five new headcount. Brand new titles. “AI Business Partner.” “Head of AI Enablement.” And look, I understand the impulse, it feels decisive, it looks great in the org chart.
But walk through the logic with me. The entire premise of this transformation is that there’s too much manual, redundant human effort clogging up the company. And the proposed solution to that… is to hire more humans. We are fighting headcount with headcount. We’re going to onboard five brand-new people whose job is to walk around telling everyone else they should be doing less. New headcount is not new capability. Most of the time you already have the people who understand the work – your process owners, your product people, the folks who’ve been quietly running things for years. You don’t need to hire a transformation. You need to stop ignoring the one you’ve already got.
Lie number three: “This is a cost-cutting exercise”
The whole thing gets framed around one question: how many steps can we cut, how many FTEs can we save. And here’s why that’s a trap – it locks you into the smallest possible prize. You spend three months lovingly optimizing a process, shaving off seconds, tuning it like a Formula One pit crew… for a process you should have just deleted. It’s like getting a full performance tune-up on a car you’re about to drive into a lake. Very efficient. Wrong direction.
The questions that actually move the needle are the uncomfortable ones. Does this process need to exist at all? And – the one almost nobody asks – what could we now do that we simply couldn’t before? Efficiency is the floor. Everyone stops at the floor and calls it a transformation.
Lie number four: “Let’s empower everyone – personal productivity for all!”
Give every single employee a chatbot, watch the magic happen. And people love it. Adoption numbers go up and to the right. The CEO gets a standing ovation.
I am not against this, to be clear. It’s lovely for morale and it’s great for getting people comfortable. But let’s be honest about which bucket it lives in, because everyone quietly slides it into the wrong one. Handing out personal-productivity tools is the corporate equivalent of buying a gym membership. You feel measurably healthier on the drive home, you tell everyone about it at dinner and the actual results, statistically, never arrive. Your people saving twenty minutes drafting nicer emails is real and it’s nice. It is also almost completely invisible in the P&L. If you’ve bought a few thousand AI seats and you’re sitting there waiting for the finance line to move – pull up a chair, this could be a while.
Lie number five: “But look at our beautiful process maps”
The map zealots. These are the teams that spend nine months producing a process map of breathtaking, museum-grade beauty. Colour-coded. Laminated. Possibly framed. And then… nothing. It goes on the wall. It becomes art.
A process map you don’t act on is a laminated, full-colour map of a city you are never going to visit. Gorgeous. Useless. The entire value of a map is in two things: deleting the steps that shouldn’t be there, and fixing the handoffs where work gets thrown over the wall between teams. If you’re not using the map to cut, you didn’t make a strategy document. You made a poster.
Lie number six: “We’ll split the work across three vendors to keep them honest”
This one always sounds clever in the procurement meeting. Spread the risk! Create competition! And comparing a couple of vendors during a small pilot – fine, sure, smart even. But splitting the actual execution of the real project across vendors who are incentivized to outshine each other? You’ve just hired three contractors to build one house and told them they’re not allowed to speak to one another. You are not going to get a house. You’re going to get three foundations, two roofs, no plumbing, and a very expensive group chat where everyone blames everyone else.
Lie number seven – and this is the one that’s fatal on its own: “We’ll get an executive sponsor later.”
No. Stop. The chance of a successful enterprise transformation with no real executive sponsor is not “lower.” It’s not “challenging”. It is zero. A transformation without a sponsor is a school group project where the one person who actually does the work eventually graduates and leaves – and somehow the entire rest of the group still fails the class. If nobody with genuine authority has their name welded to this thing, do not start. Save the money. Buy everyone a nice lunch instead. You’ll get more measurable value.
What actually works: showing people their real process
Alright. I’ve spent a good ten minutes describing the smoking crater. Let me throw you a rope, because there is good news, and it’s almost insultingly unglamorous.
The single most reliable thing I know in this entire field is also the cheapest. Ready? Show people their own process. That’s it. You map how the work actually happens – not the official version, not the flowchart from 2019, the real one with all the duct tape – and you show it to the people who do it every day.
And they don’t get defensive. They get horrified. There’s this beautiful, reliable moment where someone stares at the map of their own job and goes quiet and then says, “wait, we do what? We send it where? And then it comes back?” You don’t have to convince anyone of anything. Reality does the selling for you. In one mapping exercise we ran, the teams themselves launched dozens of improvement initiatives, not because a consultant told them to, but because they finally saw the whole thing end to end and couldn’t un-see it.
And here’s where the gold is buried, every single time: in the handoffs. The waste is almost never inside one team’s work. It’s in the no-man’s-land between departments – the stuff that gets thrown over the wall, sits in someone’s inbox for two days, gets re-entered into a different system by hand, and then thrown back. Nobody owns the space between the silos, so nobody ever fixes it. That gap is where your money is.
Second thing that works: you have to look with both eyes. System data and talking to actual humans. System data on its own is context-free – it tells you what happened but never why. And interviews on their own are folklore. If you only ask people how they work, you don’t get a process map, you get a horoscope. Charming, confidently delivered, and not connected to observable reality. You need both, or you’re guessing with extra steps.
Third: pick the right battle on purpose. Back-office stuff – finance, HR, admin – gives you faster, cleaner ROI, which is perfect for proving value early and getting the doubters on side. Core processes are harder, slower, and politically radioactive, but they’re worth far more in the long run. Both are valid. Just choose which game you’re playing, instead of discovering halfway through that you accidentally picked the hardest possible target because it was someone’s pet idea.
And fourth – transfer the knowledge from day one. If, at the end of all this, the only people who understand your transformation are the outside consultants, you don’t have a transformation. You have a subscription. And yes, I’m aware that vendor lock-in is, technically, great for my mortgage. I’m telling you to avoid it anyway. That’s the #NoBullshit tax. Build the capability inside your own walls so that one day you can fire us and feel nothing. That’s how you know it worked.
Why half the value evaporates and what your dashboard can’t see
I’ve saved the most important number for last, because once you see it, every failure we’ve talked about today suddenly makes sense.
When you run a proper discovery on a big organization, you’ll typically identify somewhere between ten and twenty-five percent of improvement potential. Real, documented, here’s-where-the-money-is opportunity. Sounds fantastic.
Now here’s the part nobody puts on the conference slide. Of that, companies typically realize – actually bank, actually show in the P&L – about five to ten percent. Roughly half of it.
So read that back slowly. Half the value you find evaporates somewhere between “we found it” and “it hit the books”. It’s a diet plan that correctly identifies the twenty pounds you could lose, and then stands there politely while you eat ten of them back in the car on the way home from the doctor.
That is what failure actually looks like. It’s almost never a failure of finding the opportunity. Discovery is the easy twenty percent. The failure is in the boring, grinding eighty percent – the execution, the tracking, the follow-through, the unsexy discipline of actually making the change stick while the organization desperately tries to snap back to how it’s always done things. A ten-percent year-over-year improvement is an aggressive target, but it’s genuinely achievable and only with rigorous tracking. Without it, you don’t have savings. You have a PDF of theoretical savings, which spends about as well as Monopoly money.
And let me leave you with one last uncomfortable fact, my favorite one, because it ties the whole episode together with a bow. Every single time we measure how work really happens inside a big company, leadership is genuinely shocked by one thing above all others: how much of the entire operation is secretly running on Excel.
Not the ERP they spent forty million on. Not the fancy platform. A spreadsheet. Usually several. Often the most critical number in the entire company lives in a single file, on one laptop, maintained by one person, and when that person goes on holiday, the whole department holds its breath. Your company does not run on SAP. It runs on Grażyna’s spreadsheet. And – this is the punchline – that spreadsheet is invisible to your process-mining dashboard. Remember lie number one? The most important process in your business is happening in the exact blind spot you decided not to look at. The keys were in the dark part of the park the whole time.
Your 3-step action plan
So what did we actually learn today, other than that you should probably go check what’s in Grażyna’s spreadsheet before she retires?
We learned that AI transformations almost never fail for the reason on the post-mortem slide. They don’t fail because the technology wasn’t ready, or the model wasn’t smart enough, or you bought the wrong license. They fail because, at every fork in the road, the comfortable choice and the correct choice were pointing in opposite directions and comfortable is really, really seductive when there’s a board meeting on Thursday.
The fixes aren’t clever. They won’t get you a keynote slot or a viral LinkedIn post. Fix the process before you point a robot at it. Look with both eyes. Show people the mirror and let reality do the selling. Keep a real human on the controls. Track the value all the way into the P&L, not just onto a slide. And for the love of all that is holy, get an executive sponsor before you spend a single euro.
So here’s your homework – three things, this week. One: pick one process you’re proud of and ask the genuinely heretical question, does this need to exist at all? Two: go find out how much of it is actually living in a spreadsheet your dashboards can’t see. And three: if you can’t name the executive whose neck is on the line for this transformation – stop, and go find them, because right now your odds are exactly zero and no amount of AI is going to change that arithmetic.
If you’ve spent this whole episode in a slowly rising panic, recognizing your own company in a little too much of it – good. That’s the first honest moment of the whole transformation. And it happens to be exactly what we do for a living: figuring out what’s actually worth fixing before anyone spends the big money. Go dig through our past episodes – the process discovery stuff, the playbook, the whole ladder. I promise it’s a better use of your next twenty minutes than the thought-leadership webinar currently buffering, ignored, on your second monitor.
Outro
And that’s a wrap! The comfortable lies have been named, the blind spots have been lit up, and somewhere an accountant named Grażyna has no idea she just got a shout-out.
Domo Arigato for listening. We know your time is valuable – worth more, frankly, than nine out of ten AI pilots currently gathering dust – so thank you for investing it here.
Big thanks to the team at Office Samurai for the scar tissue I get to call “content,” and to our producer, Anna Cubal, the only Human-in-the-Loop around here cleared to pull the plug on me mid-sentence. Recorded, as always, in the cozy halls of Wodzubeats Studio.
If this episode talked you out of one doomed pilot, hit subscribe and leave a five-star review. And if you do know an accountant named Grażyna, do send it to her. Her work deserves to be appreciated.
Until next time, make the correct choice, not the comfortable one. Mata ne!
