CConstruction AI Maturity Index

Know where AI pays off — and where it’s liability-gated.

CAMI is an independent, expert-led AI-maturity audit for architecture, engineering and construction consultancies. Through personal one-to-one interviews, a Chartered Engineer and arbitrator establishes where AI pays off in your firm, where it must not be used, and how to capture the wins — grounded in your organisation, not a template.

Independent · expert-led · person-confidential · AEC
The audit

Four things a chatbot cannot give you

A six-dimension maturity framework is a scaffold; any capable model can produce one. The audit is built on what only an expert in the room can establish.

01 · Observed

Ground truth about your firm

The bespoke workflows and workarounds a firm actually runs on — the junior's script that quietly does the QA, the cost data that lives in one person's spreadsheet. Not in any model's training data; only obtainable in the room.

02 · Evidenced

The gap between what you think you do and what you do

Observed evidence set against self-report: a retrieval test, a system walkthrough, the senior's account of a handoff beside the junior's. The divergence is the finding.

03 · Accountable

Liability calls with a name on them

Where AI must not be used, with the statutory reasoning, signed by a Chartered Engineer and arbitrator. Defensible to a board or a professional-indemnity insurer.

04 · Buildable

A specific automation design

For the workflows that pay off: the topology — full, partial, human-in-the-loop or hybrid — the decomposition, and the tooling per part. Detailed enough to act on, by your own team or whoever you choose.

The engagement

Two weeks, one connected slice of your organisation

Not a survey and not a sample of unrelated people: a cluster whose workflows join at handoffs, traced across whatever levels those handoffs cross. The shape below describes the standard audit; every engagement is scoped with its sponsor.

  1. 00Before the week

    Kickoff

    With the sponsor: the strategic question the readout must answer; an org-map of the slice — who does what, which hats they actually wear, the handoffs between them; scope, consent, and the confidentiality promise.

  2. 01Week one

    Discovery interviews

    Personal one-to-one interviews across a connected cluster of people — senior first for the official process, junior last for the real one. Each inventories that person's workflows, the handoffs in and out, and the evidence. Typically five to ten people, at most an hour each.

  3. 02Week two

    Synthesis

    Scores on the six dimensions; the workflow graph with its confirmed handoffs, blind spots and overlaps; the integrated heat-map; the liability calls; automation designs for the workflows that pay off. One or two senior follow-ups close the gaps.

  4. 03Live

    Readout

    A presentation in person or on a call: what is unique here, the radar, the heat-map, the buildable blueprint, and the honest account of where AI cannot be used.

  5. 04Some months on

    Follow-up

    A check-in: did the first change land, does anything need help. Support, not extraction — and what worked or failed is how the index learns.

The benchmark engine

Six dimensions, one ladder

Underneath the discovery work sits a fixed structure, so that every engagement is scored the same way and can later be read against others. Scores are set on a five-step ladder, to one decimal, always against 5.0. The rubric itself is not published.

D1Know
How the firm captures, finds and reuses what it knows — drawings, details, decisions, precedents.
D2Flow
How work and information move between people and systems: the handoffs, the re-keying, the workarounds.
D3Risk
Who is accountable for AI use, what is written down, which gates are operational, how liability is held.
D4Tools
What is in use, what is sanctioned, what runs in the shadows, and how well any of it is integrated.
D5People
Skills, willingness and ability across roles; where adoption stalls, and why.
D6Money
What is spent on tooling, what it returns, and whether the spend is landing.
The maturity ladder
  1. 1Ad hoc
  2. 2Emerging
  3. 3Defined
  4. 4Managed
  5. 5Optimised

Single-hue by design: a level states where a firm is, it does not grade it.

Feasibility × impact

Where AI pays off — and where it can’t

Every workflow in an engagement — standard and discovered — is plotted by present-day feasibility against delivery impact. Deeper green is higher opportunity.

Hatched cells are liability-gated: high-value work that remains a licensed human responsibility, where AI is advisory only, whatever its impact. Stating that line clearly is the point of the audit.

Impact →
Design sign-off
Stamped-drawing review
Safety-critical checks
Tender document triage
Drawing & detail retrieval
Contract notices
Design option appraisal
Change-control drafting
Cost data extraction
Meeting minutes & actions
Programme risk flags
Specification first pass
Schedule of rates matching
Report assembly
Correspondence summaries
Site progress capture
Precedent search
Transcription & tagging
Formatting & filing
1Feasibility →5
Opportunitylow → AI pays off
Liability-gated — AI advisory only
Illustrative — generic AEC workflow categories, not client data and not a measured result. Hatched cells are liability-gated: licensed human responsibility, AI advisory only.
Principles

What holds on every engagement

Read the principles in full
  • 01

    Independent

    CAMI recommends; it does not sell software, and takes no fee from anyone it recommends.

  • 02

    Person-confidential

    Interviews are private. The report carries findings, not who said what; willingness and ability are reported as the firm's aggregate shape, never by name.

  • 03

    The scoring layer is not published

    You receive scores, evidence and reasoning. The instrument stays with CAMI, so that it cannot be gamed and every firm is measured the same way.

  • 04

    Anonymised patterns build the index

    Under a consent clause in the engagement terms, de-identified and aggregated findings join the benchmark that later audits are read against. A client may decline; the terms say what is stripped, how to withdraw, and when a pattern may be served.

  • 05

    Liability calls carry a name

    Every “must not” is reasoned from the statute and signed by the assessor.

The assessor

Conducted by the engineer who signs it

Musa Yılmaz

Chartered Civil Engineer · CIArb · APM

Fifteen years in architecture, engineering and construction — a decade with contractors, then consultancy — across claims, cost, programme and project controls. The audit is designed and conducted by the assessor whose name is on the liability calls.