You already bought the tools. Most of them aren't paying for themselves yet. We do the unglamorous part: map the real process, wire the systems together, and own the exceptions.
McKinsey has run the same survey every year. The adoption line went vertical. The impact line didn't move.
The gap isn't the model. It's the workflow around it.
Source: McKinsey & Company, The state of AI. Global survey fielded 4 May to 8 June 2026.
Not theory. These are the failure modes we find when we open up a stack that was supposed to be finished.
The demo worked in March. It's still a demo. The last 20% — auth, edge cases, error states — became nobody's job.
A connector breaks silently. You find out weeks later when the numbers don't reconcile, and the fix is archaeology.
The happy path is automated. The 15% that isn't happy still lands on a human, with no queue, no visibility, and no SLA — a service level agreement, meaning an agreed maximum time before it has to be resolved. So it waits.
Someone asks why the system did that. "The model decided" is not an answer you can give a customer, a regulator, or a board.
Six subscriptions, three overlapping automations, two people who half-understand them, and no current-state map anywhere.
Nobody measured the process before it changed. So nobody can say whether it actually got better, and the budget dies quietly.
Klarna's is the most-cited automation win of the decade. It's also the most-cited correction. Both halves are public. We think you should see both.
"As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality."
— Sebastian Siemiatkowski, CEO, Klarna
Both sets of numbers are real. The automation worked on volume and failed on judgment. The design question was never "can AI handle this?" — it obviously can. The question is where you draw the line, and who catches what falls past it. Every engagement we run starts there, before a single workflow gets built.
Sources: Klarna press release via PR Newswire, 27 Feb 2024 · CX Dive, May 2025.
If automation is already running somewhere in your business, you don't need another tool. You need the parts around it that nobody enjoys building.
Current-state map with cycle times, touch counts, volumes and exception rates. The step almost everyone skips, and the reason almost everyone can't prove ROI later.
Connective tissue between the tools you already pay for. APIs, webhooks, middleware, and the auth and retry logic that keeps it alive past month three.
Invoices, POs, certificates, forms, PDFs from suppliers who still fax. Structured output with confidence thresholds and a review queue for anything below the line.
Routing, drafting, triage and multi-step agents — scoped, logged, and held to a human review step wherever a wrong answer is expensive.
Queues, alerts, retries, dead-letter handling, and a dashboard that tells you it broke before your customer does.
Runbooks, an ownership map, and a system your team can operate without us. If you can't fire us cleanly, we built it wrong.
No six-month discovery. No platform migration. We start with one process, prove it, then widen.
About an hour, however you prefer to talk. We walk one process end to end: the steps, the systems, the handoffs, where it stalls and how often it goes wrong. You keep the map whether or not you hire us. Free.
Ranked by hours recovered against build effort, with the measurement agreed up front. Includes what we won't automate and why, which is usually one or two items on any list.
First workflow live in 30 days, inside your accounts. You get the runbook, the monitoring, and the ownership map. We stay on call 60 days.
Most agencies sell you the model. We come from the operations side, where a process that leaks isn't an annoyance. It's a finding.
Extensive experience across medical device, aerospace and other regulated manufacturing. Supply chain, operations and quality systems, where documented process control is not optional.
We measure the process before we touch it. If we can't measure it, we won't claim it later. No invented percentages.
The happy path is the easy fifth of the job. We design for the messy remainder first, because that's what decides whether it survives.
We build inside your environment on tools you own. No proprietary platform, no lock-in, nothing that dies if we do.
Runbook, ownership map, monitoring. Written the way a controlled work instruction is written, because that is the discipline we come from.
Some processes should be simplified or deleted first. Automating a bad process just makes it fail faster and more consistently.
TheAIPlumbers was built by people who spent their careers making real operations run.
Our team has extensive experience inside regulated manufacturing — medical device, aerospace, diagnostics — running supply chain, operations and quality systems. In those industries a process that leaks doesn't just cost hours. It shows up as a nonconformance in the next audit, with your name on it.
That work is unglamorous and specific: map the process, measure the cycle time, find the handoff where things stall, write the procedure, prove the change held. Six Sigma discipline applied where being approximately right was never good enough.
Here's what changed recently. The tooling caught up. Work that needed a six-figure integration project ten years ago is now a careful week of wiring. But the skill that decides whether it holds is the same one it has always been: knowing how to map a process, where the exceptions hide, and what to measure before you touch anything.
Most automation projects don't fail on the technology. They fail because nobody did that part. MIT's NANDA initiative put the share of generative AI pilots delivering no measurable P&L return at around 95% — the figure is contested, the direction isn't.
Call us plumbers if you want. We like it. Plumbers find the leak, fix it, and pressure-test it afterwards. That's the whole job.
If we ever break one of these, call us on it.
We measure ourselves by the time your team gets back against a baseline we recorded together. If an automation doesn't clear it, we rip it out.
If we can't walk the person who runs the process today through the whole design, we haven't finished designing it. A complicated explanation is a symptom of a confused build.
If the process is too small, too volatile, or better fixed than automated, we'll say so on the first call. Short-term losses build long-term trust.
Accounts, data, workflows, documentation. We build inside your environment on tools you own. If you fire us tomorrow, nothing breaks.
The one everybody complains about and nobody has mapped. Sixty minutes, no cost, and you keep the map.
Contact UsWe'll reply within one business day. The first conversation is free, takes about an hour, and covers one process. You keep the current-state map either way.
Rather just write to a person?
fix@theaiplumbers.comWritten for people who already know their way around automation. If yours isn't here, ask it in the form above.
Usually not another tool. What's missing is almost always the same three things: a current-state map with real numbers, exception handling with a named owner and an agreed response time, and monitoring that tells you when a step failed. We'll often keep your existing automations exactly where they are and build the missing layer around them.
Often, yes — and if you have someone who genuinely owns it, we'll tell you to. The usual failure mode isn't capability, it's that maintenance becomes nobody's job after the person who built it moves on. We design the handover for that specific risk.
Baseline before, same measures after: cycle time, touch count, exception rate, rework, and hours per week on the process. Recorded together in that first conversation so neither of us can move the goalposts later.
Anything where the cost of a wrong answer is higher than the cost of a person doing it, unless we can put a review step in front of the output. Also anything that changes every month — automate a moving target and you'll maintain it forever.
Your accounts, your logins, your environment. We take the narrowest scopes that work. NDA on request. We don't copy your data into our systems and we don't train models on it.
That is where the team comes from. Medical device and aerospace manufacturing, where change control, traceability and documented validation are the baseline rather than a nice-to-have. If your automation has to survive an audit, say so in the first conversation.