There is a seat at every AI deliberation that is almost always empty.
Every institution deploying an AI system moves through the same five decisions. At each one, the measure of success is usually written by whoever has a stake if the answer favors them — whether an external vendor selling software or an internal team championing a build. Iceberg Tribe Advisory exists to hold that measure independently instead — before capital, code, or contracts are committed.
Historical Parallels
An old problem, condensed
This isn’t a new problem. Four major industries faced this exact breakdown and arrived at the same structural conclusion.
Capital Markets
Split sell-side research from buy-side analysis.
Construction
Owner's engineer certifies work and pours none of the concrete.
Audit & Accounting
Separated certifier and certified by law post-Arthur Andersen.
“Whoever defines and verifies a measure must not profit from the outcome of the measurement. AI has the stakes these industries had. It doesn’t yet have the seat.”
Market Reality
What the interval has cost
95%
Pilot Failure Rate
Enterprise AI pilots delivering no measurable P&L impact.
Case Study 01 / Quality Rollback
The 700-Person Automation Claim
A celebrated AI deployment announced automating 700 jobs. Within two years, its CEO conceded quality suffered and began quietly rehiring.
Case Study 02 / Audit Vacuum
The $1.5B Platform Illusion
A platform sold “AI-automated” work powered by hundreds of human engineers — hidden for years because clients lacked audit rights.
None of this required dishonesty. It only required that the party being measured also write the measure.
Structural Conflicts
Why existing seats can’t hold this
The market already has advisors, but their compensation models create inherent conflicts of interest.
Systems Integrators
Bid for the implementation they're advising on.
Global Consultancies
Formally allied with the vendors they recommend.
Internal AI Leads
Mandate is measured in deployments shipped (championing & auditing conflict).
The seat needs an independent authority with no commercial stake in software sales, no billable implementation hours, and no reputational bias toward defending an internal build.
How measurement authority holds: The five decisions
Every AI deployment moves through the same five decision gates. At each step, the critical question is simple: who wrote the measure?
Select any decision gate below to inspect how measurement authority is tested, audited, and preserved at that specific stage of your deployment.
Problem framing
GATE 01Where AI is going wrong before you even start →
The use cases on the table are often the ones the vendor has a playbook for — not the ones that touch your actual constraint.
ROI definition
GATE 02Before you sign the business case →
The return is usually modeled on assumptions the vendor supplied, against a baseline nobody froze in writing.
Buy vs. build
GATE 03Build or buy — whose numbers are you using →
The buy-versus-build comparison is priced, more often than not, by the party selling "buy" — or framed around what an internal team already knows how to build.
Selection
GATE 04The vendor evaluation →
Vendors are graded on benchmarks they chose, running demos they designed.
Verification
GATE 05What happens after signature →
The party that built the system and operates it — whether an external vendor or an internal engineering team — is usually also the party reporting how well it's working.
“The measure decides the outcome. The only question that matters is whose measure it is.”
Wherever your own initiative sits in this sequence — pick the gate above, and check whether the measure it’s using was written by someone with something to gain from the answer.