AI Isn't Underperforming. It's Exposing What Was Always There.
A Post I Read Recently, and Why It Wouldn't Leave Me Alone
I read a post a while back that opened with an ERP story. The author had been part of an implementation that had to be relaunched almost exactly one year after go-live. Not optimized, not tweaked, relaunched. As they told it, the software wasn't the problem. They were. Over time their team had piled up workarounds, exceptions, local optimizations, and disconnected processes, and every one of those made perfect sense to the department that built it and none of it made sense to the enterprise. They'd expected the technology to accommodate all of it, and it couldn't. The system, they wrote, couldn't manage their systemless system.
Then they made the turn that stuck with me. Someone had told them, "AI just hasn't delivered the efficiency we were promised," and their response was: I've seen this film before.
That line from the post is why I'm still thinking about it. Because they're right, and I don't think most leaders have connected the dots yet.
AI Doesn't Fail Institutions, It Reveals Them
The question the post asked is the one worth sitting with: what if AI isn't underperforming? What if it's exposing everything around it?
I'd push it one step further. The workarounds that author described weren't just clutter. They were judgment nobody wrote down. Somebody, somewhere, decided a process needed an exception, and they were probably right in the moment, and none of that reasoning made it into the system. It lived in a person's head, in a side spreadsheet, in "that's just how we do it here." When the ERP went live, it inherited the structure and lost the reasoning behind every exception built on top of it. That's not a software gap. That's an institutional memory gap wearing a software costume.
The mistake is believing AI will make yesterday's institutional design work in today's environment, and believing we don't have to adapt because AI will somehow adapt for us. Every major technological revolution has amplified the strengths and weaknesses of the institutions that adopted it, and AI won't be any different. It won't create good governance where none exists. It won't clarify confused ownership. It won't align misaligned incentives, and it won't untangle years of accumulated complexity on its own. It'll expose all of it, in real time, in front of everyone who's watching the results.
The implementation in that post didn't fail because the technology wasn't capable. It failed because the technology was asked to solve an institutional problem it was never built to solve. I suspect AI will be even less forgiving of that same mistake.
Why This Keeps Happening
Most businesses don't have people problems. What they do have are process problems, and most process problems become technology problems, and most technology problems eventually become leadership problems. The cost of inefficiency creeps and drifts quietly at first: small delays become large delays, small errors become large losses, small disconnects become organizational friction that everyone learns to route around instead of fix. Eventually entire businesses get built around compensating for bad systems, and that should never happen. Technology should make work easier.
Not harder.
The Quiet Version of the Same Failure
An ERP relaunch is a loud failure. It's public, it's budgeted, it shows up on a board slide with a red status next to it. AI failure is quieter and, honestly, easier to hide from. It looks like a pilot that never scales past three users who like it. It looks like a dashboard nobody trusts enough to actually act on. It looks like leadership calling the whole thing "underwhelming" instead of asking the harder question: what did this just show us about how our business actually runs?
Same film, different theater, same audience still surprised by the ending.
So What Actually Fixes It
Since the onset of my life in aviation, albeit then accidental, I've watched this pattern from both sides now, inside operations that were held together by tribal knowledge, and inside a company that's trying to build the thing that replaces tribal knowledge with structure instead of erasing it.
The fix was never a better dashboard. It was never a smarter model bolted onto a broken workflow after the fact. The fix is a system built to match the way the work actually happens: connected instead of fragmented, aviation-native instead of aviation-adapted, complete instead of patched together with middleware and consultants who charge you to explain your own process back to you.
That's the difference between technology that gets blamed for institutional failure and technology that closes the gap the institution couldn't close on its own. My job isn't to relitigate why somebody else's implementation didn't go as planned. My job is to build the thing that doesn't force you to fail the same way twice.
How ERP.aero Was Built to Solve This
Here's the part worth sitting with about that relaunch story: nobody sat down and designed it to fail. It failed because the work lived in twenty places and the software only understood one of them, and because the reasoning behind every workaround stayed in someone's head instead of getting captured anywhere the system could use it.
That's still the default state for most aviation parts suppliers and MRO shops today. RFQs come in from email, a website form PartsBase, ILS, Locatory, AVSpares, the145, AeroXchange, wherever they happen to land. Vendor quotes come back the same scattered way, maybe more scattered. Somebody re-types what already exists somewhere else because the system in front of them has no idea what the system next to it already knows, and the judgment about why a particular vendor gets trusted or why a particular exception gets approved never makes it into either one. That's not a people problem. That's an architecture problem, and no amount of AI sprinkled on top of a broken architecture fixes it, not even agents stringed together nor using the day's best lingo about it.
One Screen Instead of Five Systems Arguing With Each Other
The One-Screen Quoting Engine™ puts RFQs, vendor quotes, certs, inventory, pricing history, and buyer context in one place. Not ten tabs, not twenty. One page. Every tab switch is a place where institutional knowledge quietly leaks out, the thing the last person knew that the next person now has to guess at. Put it all on one screen and the guessing stops, because the quote goes out faster when the person quoting isn't hunting for the fifth piece of information they need before they feel confident hitting send.
The Data Enters Once
​iRFQ auto-captures and parses RFQs and vendor quotes from email, portals, PDFs, and marketplace replies directly into structured line items, with no manual typing required. At 500 RFQs a day, most teams burn three to five hours per resource re-entering data that already exists somewhere else, and that's not effort, that's waste dressed up as busywork. iRFQ removes the step entirely: the quote comes in already structured, and your team is quoting the next deal instead of retyping the last one. Which means the workaround that would've quietly become "how we do things here," the exact kind of undocumented habit that sank the implementation in that post, never gets the chance to form in the first place.
The Judgment That Doesn't Walk Out the Door
This is where I think the original post undersold its own point. The workarounds that team accumulated weren't random. They were built by people who'd learned, over years, which vendor actually delivers, which trace package deserves a second look, when a price feels wrong before the numbers prove it, and which exception protects the business versus which one exposes it. That's real judgment, earned the hard way, and almost none of it typically survives a system migration, a reorg, or a retirement.
SOPs preserve instructions. They don't preserve the judgment behind them, and that gap is exactly where institutions quietly bleed value every time someone experienced walks out the door or a system gets swapped out from under them.
​ELIA, the intelligence layer built into ERP.aero, is designed to help close that gap, and it doesn't live in one bolted-on screen off to the side. It lives inside the RFQ, ranking which requests are at risk of going stale and pulling up the history behind a similar deal before anyone has to dig for it. It lives inside the Quote, surfacing pricing history and vendor history at the moment a decision is being made and flagging a quote that's about to go out below margin. It lives inside Vendor Accounts, ranking which supplier actually delivers instead of just which one quoted the lowest number. It lives inside the Sales Order and the Purchase Order, watching for the exposure that turns a commitment into a missed promise before it happens. It lives inside Comments, where the judgment usually gets typed in passing and then forgotten, helping capture the context and rationale behind approvals and exceptions as they happen instead of trying to reconstruct them later from memory. And it reaches into the rest of the operating surface the same way: Parts, Customer Accounts, Invoice, Packing, Shipping Notice, ToDo, and Events, everywhere a decision gets made and a reason needs to survive it. It isn't a chatbot bolted onto a menu, and it isn't meant to replace the person making the call. It's meant to make sure the reasoning behind that call doesn't disappear the next time someone leaves the room. Keep humans in control, and let the system remember what it's been shown.
That's the deeper version of "AI won't fix institutional problems on its own." It also won't have anywhere to start if the institutional knowledge never gets captured in the first place. ELIA is ERP.aero's answer to that specific gap: preserve the judgment while the work is happening, not after somebody's already gone.
Going Live Without Betting the Business
This is where that relaunch story matters most, because most ERP buyers have been burned exactly that way before: sold a big-bang cutover, then living inside the fallout for a year until somebody finally admits it needs to be relaunched.
The Go-Live Confidence Plan™ is milestone-based, with acceptance criteria at every phase. You sign off before we advance, and if a milestone isn't met, we don't move forward until it is. Rollback-safe implementation means pilot activation and phased rollout, so your current operation stays live while ERP.aero comes online, and you never have to choose between running the business and implementing new software. The Data Liberation Clause™ means your data is yours, with a clean, structured export available on request at any time. No lock-in, no hostage data, no renegotiation required to get back what you put in.
None of that is decoration. It's the direct answer to the exact failure mode in the story that opened this piece: technology asked to solve an institutional problem, with no way to pull back when it couldn't.
If You Were Designing Your Institution Today
Looking back, that ERP implementation didn't fail because the technology wasn't capable. It failed because the team asked technology to solve institutional problems technology was never built to solve, and the judgment that could've prevented the failure was never captured anywhere the system could reach it.
AI will be even less forgiving of that same mistake. It doesn't patch over confused ownership, and it doesn't average out a misaligned incentive. It shows you exactly where the design breaks, in real time, at scale, in front of everyone.
Respect the past, understand the present, build the future, and never confuse familiarity with progress.
The question that closed the post I read is the one I keep coming back to: if you were designing your institution, knowing what you know now, would you build it the same way?
​ERP.aero was built for the suppliers, distributors, and MRO shops that already know the answer is no. Banner Aircraft International scaled quoting capacity 30% with no added headcount. Mountain Air Parts answers 20 RFQs in under five minutes. Vendor sourcing cycles run 42% faster, invoice posting is 83% faster, and inventory accuracy sits at 99.99%, with the judgment behind those transactions captured instead of walking out the door with whoever made the call.
None of that happens by accident. It happens because ELIA is where the institutional knowledge actually lives now. Not in one person's head. Not in a drawer of notes nobody reads after they're gone. Not in a workaround someone will forget to explain before they retire. In the system, still reachable, the next time someone needs it.
One screen. No bolt-ons. Built for aviation.
If you want to see what a system built to match the work, instead of asking the work to bend around the system, actually looks like, book a 20-minute live quoting walkthrough. Use code KIT for implementation credit, or grab time directly at calendly.com/ralphmerhi.
-About ERP.Aero
​ERP.Aero is an aviation-first ERP platform built to connect RFQs, quoting, inventory, purchasing, repairs, work orders, compliance, shipping, invoicing, reporting, and customer intelligence in one continuous workflow. ERP.Aero helps aviation companies respond faster, work with fewer handoffs, and make better decisions from the same page. While you're there, check out .../build.
​