Where to Start: The Clean Baseline Your AI Transformation Is Missing
Also published on LinkedIn
Every organisation we talk to is somewhere in the AI shift. Tools bought. Pilots run. Wins claimed. And when you ask the honest question, almost all give the same answer: we're not sure where we actually are.
That's the whole problem. You can't design a transformation from a position you can't describe, and most organisations are transforming at full speed from a base they never mapped.
So this week we're putting a starting point in front of you: our Workforce AI Impact Assessment. A task-level read of how AI hits your current workforce: where the exposure is, where the capacity is hiding, where your data can't answer at all. It's the front door to everything we build, and here's why that first step matters more than any tool decision you'll make this year.
AI works at task level. Most orgs deploy at macro level.
This mismatch drives most of the failure you've read about.
AI changes tasks. Not jobs, not departments, not strategies. A role is a bundle of tasks and AI hits the bundle unevenly: some automate, some accelerate, some become more valuable, some are brand new. That's the resolution AI operates at.
But most organisations deploy at macro level. A platform for everyone. A copilot on every desk. Deployed onto an org that's only mapped at macro, the AI has no context for the actual work, so it guesses. And every guess becomes output someone downstream treats as truth.
That's how you manufacture dirty data at scale. Dirty in, dirty out, now at machine speed. 85% of failed AI projects cite poor data quality as a root cause, only 12% of organisations have AI-ready data by Gartner's count, and their forward call is blunt: 60% of AI projects lacking it will be abandoned through 2026.
Most organisations don't need another AI tool. They need an AI cleanse: a structured pass over what the work actually is, who does it, what it runs on, and where the record has drifted from reality.
The organisation brain is becoming the category
There's a reason "company brain" keeps showing up in your feed. Forbes put a name on it last week: the enterprise brain, a governed, always-learning model of how the company actually works, that both people and AI can act on. The diagnosis behind it is one we've been making for a year: AI fails in production because it lacks organisational context.
One thing the coverage keeps missing, though. Most company-brain plays start from documents: wikis, drives, tickets. But documents describe the work as someone once wrote it down, not as it happens today. The brain that matters models the work itself: tasks, roles, processes, tools, vendors, agents. Context for AI. A map for the humans deciding what changes.
Quick wins that don't compound
Plenty of organisations did move fast. Eighteen months in, the pattern is visible: the wins didn't stack. An hour saved here, a draft accelerated there, and none of it connects, because every team built alone against no shared baseline.
MIT found 95% of enterprise AI pilots delivered no measurable return. S&P Global found 42% of companies abandoned most of their AI initiatives last year. The gap isn't model quality: McKinsey found the winners are twice as likely to have redesigned the work before selecting the tools. Map first, deploy second. The 5% did it in that order.
Speed from a solid base compounds. Speed from no base just accumulates.
The invisible work
Underneath it all sits a harder truth: most of the work in your organisation has never been mapped.
Processes live in heads and team channels, not systems of record. JDs were written at hire and never touched again. Managers know broadly what their teams do, but not at task level, and task level is where AI lands. So when the board asks what the AI spend returned, nobody can answer. You can't track ROI against work that was never made visible.
Into that gap pours cultural confusion. Which tools are sanctioned, what the policy means for each role, what the new shape of the job is: unanswered, so some race ahead and others quietly opt out. Confusion isn't a comms problem. It's what an unmapped org feels like from the inside.
How you get to a clean base
Not a bigger platform decision. A sequence.
Run the gap analysis first. Find out what you don't know: where process data hides inside teams, which JDs have drifted, where record and reality disagree. The gaps are the finding, not a problem to hide.
Map the work at task level. Break roles into tasks, score each for AI exposure, connect the results to everything the task touches.
Bring your people in early. Your employees hold the truth about the work and have already started their own AI experiments. Map with them and you get accurate data and buy-in in one motion. Jam an LLM in with no map and no say, and you get neither.
Build the connective tissue. The value isn't in any single dataset. It's in the connections: this task feeds that process, runs on that tool, is owned by that role, is half-covered by that agent already. That's what turns scattered records into an organisation brain.
The Human Layer Graph
This is what we build, and we learned its shape the hard way.
We started by mapping job descriptions into task-level analysis. It worked, and it wasn't enough. Over 100 conversations with CHROs, COOs and transformation leads kept landing on the same point: the work doesn't live in roles alone.
So the platform now maps the full eight: people, roles, tasks, skills, processes, tools, vendors, and agents. Every one connected. That's the Human Layer Graph: one point of organisational truth that every analysis reads from and every change writes back to. Always on, always learning. The baseline stops being a report and becomes an asset that sharpens with every program you run.
Governance is the command centre, not the handbrake
Once the graph exists, governance stops being a policy document and becomes a command centre.
From one map: who needs which licences and who never touches theirs. Where token spend leaks against work with no return. Which processes are ready for automation and which need a human gate. ROI stops being an argument and becomes a number that traces to task-level evidence, in hours, in spend, in capacity redeployed.
Every organisation is heading to the same operating reality: human work, agent work, and AI-driven work side by side. The ones that balance that mix will do it from a live map. The ones that can't see the mix can't manage it, and that position gets harder to roll back every single day.
Start here
The Workforce AI Impact Assessment is the first pass: a task-level read of AI's impact on your workforce, and an honest audit of where your data can and can't answer. It establishes the baseline of truth, surfaces the gaps, and starts the graph.
The baseline is where the real work begins. Off the back of it run the transformation programmes: role redesign, stack consolidation, capability builds, agent deployment, each scoped against the map with outcomes defined up front. Then the traction programmes keep score: adoption, hours freed and where they went, spend against plan, drift against baseline. Each phase builds on the last, and the value compounds with every cycle.
Run the assessment.
Sources: Gartner AI-ready data and data quality research; Folio3 AI project failure analysis 2026; MIT NANDA, The GenAI Divide; S&P Global Market Intelligence, Voice of the Enterprise; McKinsey State of AI; Forbes, Enterprise Brain coverage, August 2026.