Pilot-to-production value

The distance between a pilot that worked and a number in the P&L.

24documents on this topic
18organizations represented
2issues named
3sourced citations
0sourced statistics

The state of it

One of 7 topics within ROI measurement.

24 documents from 18 organizations address the distance between a pilot that worked and a number in the accounts. The single most quoted figure in the whole ROI literature lives here - Gartner's forecast, carried in EY's material, that more than 40% of agentic AI projects will be cancelled by the end of 2027, for escalating cost, unclear business value or inadequate controls.

What is striking across the set is how much of the reporting is use-case counts. Several of the largest documents here are catalogues of hundreds of use cases. A count of use cases is a measure of activity; it becomes a measure of value only when someone states how many of them are in production and what they returned.

The issues, by agreement

How many independent organizations name each issue as a problem. An issue is only as real as the number of separate publishers that identify it, so the count is the ranking. Bars are organizations, not documents. Where the count reads ours, no publisher here states the issue and the analysis is our own.

Where they disagree

Points where the research contradicts itself. These are the entries worth the most attention, because no single document reveals them - they only appear when the sources are read against each other.

Use-case counts are being reported instead of production systems

Catalogues of hundreds of use cases are presented as evidence of progress. They measure exploration, not value, and the two are being conflated in board material.

The dissentDecember 2024

Boston Consulting Group — Argues directly against the catalogue approach: leaders pursue few high-value opportunities with roughly double the ROI impact rather than hundreds of use cases.

The issues in full

Each issue carries the organizations that name it, the numbers behind it, and the remedies proposed - with the concrete steps under each. Every citation points at a section of a named document, so any count here can be checked.

Issue 011 organization name it2026 evidence

The base rate is cancellation, and cost is the leading cause

The forecast every plan is competing against is that more than four in ten agentic projects will be cancelled - and the first named reason is escalating cost, not capability.

How to fix it — 1 approach, 3 steps

Set the kill criteria at approval

Decide in advance what result would end the project, and who is allowed to call it. Projects without kill criteria are cancelled late and expensively.

Done when The approval paper itself states the kill criteria and the review date, names someone other than the sponsor who can stop it, and the review was held on the date.

  1. Write the kill criteria and the review date into the approval paper itself.0-30 daysCFO
  2. Name who can stop it, and make that a different person from the sponsor.0-30 daysCEO
  3. Hold the review on the date regardless of momentum.ongoingCFO
The evidence — 2 documents
OrganizationDocumentPosition
EYConsultancy · June 2026Unlocking agentic value: a new investment disciplineOur reading Relays the forecast that more than 40% of agentic projects are abandoned before 2028, over cost, unproven value or weak risk controls.Investment discipline; Gartner forecastnames it
Boston Consulting GroupConsultancy · December 2024AI value creation in leading enterprisesOur reading Proposes concentrating on a few high-value opportunities with roughly double the ROI impact instead of hundreds of use cases.Investment concentrationproposes a fix

Issue 02Our analysis1 disputes itnewest evidence Dec 2024

Use-case counts are being reported instead of production systems

Catalogues of hundreds of use cases are presented as evidence of progress. They measure exploration, not value, and the two are being conflated in board material.

How to fix it — 1 approach, 3 steps

Report production systems and their return, nothing else

Stop reporting use-case counts upward. Report the number in production, the measured return of each, and the number retired.

Done when Production is defined by four conditions, the inventory has been reclassified against it with the drop reported once, and the report shows systems in production, the measured return of each, and the number retired.

  1. Define production: a named owner, a measured baseline, live users, a rollback path.0-30 daysCIO
  2. Reclassify the current inventory against that definition and report the drop once, openly.30-90 daysCIO
  3. Add a retired count to the same report so the portfolio is visibly managed.ongoingCIO
The evidence — 1 document
OrganizationDocumentPosition
Boston Consulting GroupConsultancy · December 2024AI value creation in leading enterprisesOur reading Argues directly against the catalogue approach: leaders pursue few high-value opportunities with roughly double the ROI impact rather than hundreds of use cases.Investment concentrationdisputes it

Who is represented

This dossier is drawn from 18 organizations working on the subject, 2 of which are cited directly in the issues above.

Consultancy — 7

Boston Consulting Group 2 EY 2 Capgemini 3 McKinsey & Company 3 Deloitte 1 Genpact 1 PwC Malaysia 1

Institution — 1

IAB 1

Academic — 2

MIT Sloan 1 Stanford Hoover Institution 1

Hyperscaler — 6

Google Cloud 2 IBM 2 AWS 1 Fujitsu 1 Microsoft 1 OpenText 1

Frontier lab — 1

OpenAI 1

Vendor — 1

Writer 1