Shopify recently mandated that employees prove AI can't do a job before requesting additional headcount. Microsoft's Copilot generates $2 billion annually helping people write documents faster. Fiverr reported an 18,347% surge in AI agent-related searches as they position themselves as "AI First".
The corporate world is racing to transform with AI.
Within software companies, product managers feed requirements into ChatGPT. Engineers use Copilot to generate technical specifications. QA teams prompt Claude for test plans.
Documents that took days now take minutes. Yet the fundamental workflow remains unchanged. Requirements document, then technical spec, then test plan, then sprint planning, then development, then review, then deploy.
We've automated the artifacts. We haven't questioned why the artifacts exist.
The 74% failure pattern
Boston Consulting Group (BCG) 's latest research reveals an uncomfortable truth: 74% of companies have yet to show tangible value from their AI investments.
Walmart deployed AI assistants to 75,000 employees while JPMorganChase estimates it will deliver $1.5B in "business value" through AI adoption. Whilst these are real efficiency gains, they're gains in helping employees search documents faster, summarize meetings, draft emails, generate reports.
The pattern is consistent: companies take their current processes, designed entirely around human limitations, and attempt to make them faster.
But faster wrong is still wrong.
The funny reality is that in software companies the requirements documents authored by product managers were historically a result of a lot of sweat and tears. PMs knew every line by heart because of the hours put into the document. Now they can provide a high level overview and then have AI generate the body of the document to fill out the required template. This then gets handed to an engineer who glosses over it, probably copy and pasting it into ChatGPT and asking for a dot point summary. They then follow a similar process: scope out the work at a high level, have AI generate the technical spec. There might be some engineers that ask a few questions still at the moment when reviewing the spec. Give it a year though, those specifications will be copied directly into a prompt and work will commence without anyone actually fully reading what is being built. Hilarious. Scary.
This disconnect runs deeper than just document generation.
Most companies don't even know what their processes actually are. They have what people do, which is different from what the official documentation says, which is different from what actually creates value.
Ask a typical enterprise software company to map their product development process end-to-end. You'd get blank stares. Someone will link you to an outdated Confluence page and, if you're lucky, you'll also find a slide deck from a consultant that helped establish "agile" in the workplace in 2017.
The companies seeing real AI transformation share one characteristic: they stopped and mapped their actual workflows first. Not the idealised versions. Not the ones in the employee handbook. The real ones with all their workarounds, exceptions, jargon, and tribal knowledge.
Every business process exists to solve a problem. But most solve problems that disappeared years ago. And when you throw generative AI into the mix almost all of the existing business processes will also disappear or evolve.
Traditional processes exist because humans need scaffolding. We need documentation because we forget. Status meetings because we lose context. Staged reviews because we make mistakes. Asynchronous handoffs because we're not always available.
AI has none of these limitations.
AI can hold entire systems in context, remember every decision, trace impacts instantly. Yet we ask it to produce the same documents, attend the same meetings, follow the same approval chains.
Two paths diverging
The market is splitting. Some companies will add AI to each existing step, celebrate efficiency gains, preserve current structures, and wonder why transformation feels incremental.
Others map what actually happens, question every step, design around AI capabilities, and accept that many current processes become obsolete.
Most enterprises choose the first path. It feels safe. It preserves jobs and org charts. It fits into existing budgets. It lets them claim transformation without transforming anything.
But the companies seeing exponential gains are on the second path.
They're not asking "How can AI help us do this faster?" They're asking "Why do we do this at all?"
Some are using the opportunity to dig through layers of accumulated practices to find their original purposes. What problem does this solve? Often, the problem vanished years ago, but the solution remains. Who actually uses this output? Frequently, nobody. The report gets generated because it always has been. What would happen if we stopped? Usually, nothing. Or something better emerges naturally.
It's the same everywhere. Follow a customer request from start to finish. Count the handoffs. Time the delays. Document the rework. Map reality, not policy. Include the Excel files, chat groups, daily stand-ups, and "quick syncs". For every step, ask what human limitation it addresses. If it's about memory, coordination, or processing limits, it's a candidate for elimination.
Start from desired outcomes and work backwards. What's the minimum viable process if AI handles this task? Imagine the AI-native version. The gap between current and possible is the real opportunity.
The path forward
Companies claiming AI transformation without doing process archaeology are building on sand. They're layering technology onto workflows they can't fully describe, optimising processes that shouldn't exist.
Real transformation is discovering that most of the work doesn't need to be done. The question isn't which AI platform to choose, it's how to get from A to B with as little human intervention as possible so you can circumvent the laws of scaling and enable your company to reach its full potential.