A company decomposed to the subprocess
Operations are broken into 11 business areas, then processes, then 133 subprocesses — the level where work actually gets done and where a tool, a team, or an AI agent is actually chosen. The taxonomy is aligned to the APQC Process Classification Framework (PCF®) so it maps to how enterprises already describe themselves.
Each subprocess gets the same five-section profile. We publish a subprocess once its profile has real substance (tasks, executors, and an AI mix) — not before.
Tag the work before naming any tool
Every task is tagged on four business-friendly dimensions — what kind of work it is and how it has to be supervised — before AI enters the conversation. The kind of work is what later decides which solution, if any, fits.
- Nature — what kind of work each task is:
- Oversight — how hands-off it can run: unattended → handles exceptions → approves → a person decides.
- Volume and Risk / control — the economics and the stakes, which decide how far automation is worth pushing.
A subprocess is summarized by its binding step — the single hardest task that gates everything downstream — not by an average. Get that step right and the rest tends to follow. The framework is research-grounded — built on established workforce-research and AI-adoption frameworks — and used as scaffolding rather than gospel.
The same five questions, every time
- Overview — what the work actually is, task by task, and where its difficulty sits.
- Improvement potential — how much better best-in-class runs it (quality, efficiency, cost, speed), and how teams get there.
- Who can run it — the five delivery levers: internal staff, your ERP/platform, specialized SaaS, AI agents, or a BPO.
- Where AI fits — the AI mix, as a result of matching solutions to the work — plus its target value and how proven it is.
- How to choose — matching the approach to your volume, variability, control needs, and appetite to operate a system.
The AI mix is a readout, not the starting point
We don't start from "where can we put AI." We start from the nature of the work, then match the right kind of solution to each kind of task — reading → document AI, deterministic steps → automation/RPA, prediction → ML, drafting → generative AI, and so on. The "AI mix" you see is simply the weighted result of that matching, concentrated where the work concentrates. As the hype recedes, AI stays one layer of the answer, not the whole answer.
Grounded, and clear about what's measured vs. estimated
- No fabrication. Every fact, number, vendor, and example comes from our database or cited research — never general knowledge or placeholders.
- The AI mix is directional. The dominant technology is research-supported; the exact percentages are a calibrated estimate, not a measured split.
- The value is graded. Improvement and AI-value figures carry an evidence grade (the strength of the research behind them), and we lead with the best-in-class frontier from the most credible named source.
- Vendor-neutral. Executors are listed as real options with their trade-offs, not as recommendations.