Pilot Purgatory Diagnostic

1

Portfolio Inventory

List every active AI initiative — name, owner, stage

Instruction: Write down every active AI initiative. Include anything in evaluation, pilot, approval, or stalled. Writing this down often reveals something immediately — the portfolio is more fragmented than leadership realized.

Initiative	Department	Owner	Stage	Time at current stage
Contract Review Assistant	Legal Operations	Sarah M.	Piloting	3 months
Meeting Summarization Tool	HR / All-org	IT Digital Team	Evaluating vendors	1 month
Customer Inquiry Classifier	Customer Service	CX Transformation	Stalled post-pilot	14 months
Code Review Assistant	Software Engineering	DevOps Lead	Active · scaling	6 months
Procurement Risk Scoring	Finance / Procurement	AI CoE	Proposed	Just started

Observation: Customer Inquiry Classifier has been stalled for 14 months with no clear path to resolution. Two separate initiatives (Meeting Summarization + Code Review Assistant) are progressing in parallel without any coordination or shared learnings.

2

Failure Mode Assessment

8 questions per initiative across three dimensions

Instruction: For each initiative, answer 8 questions across three dimensions: Technology, Integration, and People & Organization. The goal is not a score — it is to surface where the real constraint sits.

Customer Inquiry Classifier

CX Transformation · Stalled post-pilot · 14 months

Needs attention

Technology

Does this capability exist in commercially available tools today?

Yes

Did the pilot output meet your minimum accuracy threshold for production use?

Yes — 91%

Integration

Does the tool fit into the workflow where the agent actually makes decisions (not a separate step)?

Partially

Have you observed real agents using the tool in their actual workflow, not a demo?

Yes

Is time-to-output under 30 seconds from the point of need?

Yes

People & Organization

Is there a named person responsible for driving adoption — not just the technical owner?

No

Did the agents who will use this tool have meaningful input in designing how it works?

No

Is the executive sponsor still actively engaged and accountable for scaling outcomes?

No — sponsor moved roles

Technology

2 / 2

Integration

2.5 / 3

People & Org

0 / 3

Primary failure mode

People & Organization

Contract Review Assistant

Legal Operations · Piloting · 3 months

Monitor

Technology

Does this capability exist in commercially available tools today?

Yes

Did the pilot output meet your minimum accuracy threshold?

Partially — 84%, target 90%

Integration

Does the tool fit into the workflow at the point of decision?

Yes

Have you observed real users working with the tool in their actual workflow?

Yes

Is time-to-output under 30 seconds from point of need?

Yes

People & Organization

Is there a named person responsible for driving adoption?

Yes — Sarah M.

Did the lawyers using this tool have input in designing it?

Yes

Is the executive sponsor still actively engaged?

Yes

Technology

1.5 / 2

Integration

3 / 3

People & Org

3 / 3

Primary failure mode

Technology (minor)

Repeat this assessment for each initiative in the portfolio inventory. Remaining 3 initiatives not shown.

3

Governance Gap Audit

What employees can do with AI personally that they cannot do officially

Instruction: List five things employees in your organization can do with AI personally today that they cannot do through official, company-approved channels. For each, estimate how long the standard approval process would take. This is not an indictment — it is a benchmark.

Upload and summarize meeting recordings

ChatGPT / Otter.ai — widely used personally. No approved internal equivalent.

8 months

estimated approval time
Draft contract clauses or policy language

Claude / GPT-4 — used in personal time. Officially blocked by data policy.

12 months

estimated approval time
Translate client documents

DeepL / Google Translate — used freely. Company policy unclear on data classification.

3 months

estimated approval time
Generate presentation outlines from strategic memos

ChatGPT / Claude — commonly used. Microsoft 365 Copilot in procurement queue.

6 months

estimated approval time
Run sentiment analysis on customer feedback

Personal API keys used by data team. No sanctioned internal tool approved yet.

9 months

estimated approval time

Governance speed benchmark

Average approval time across these 5 items: 7.6 months. Major model capability releases are now occurring every 2–4 months. At current governance velocity, approvals for today's tools will complete when those tools are already two generations old.

4

The Output

One initiative to examine differently. One governance process to accelerate.

Instruction: Based on sections 1–3, identify the single highest-value action in each category. This is enough to make the next steering meeting a different kind of conversation.

Initiative to re-examine

Customer Inquiry Classifier

Technology works (91% accuracy). Integration is mostly solved. All three People & Org questions failed: no adoption owner, no user input in design, no active sponsor. This is not a technology problem — it never was. Fourteen months of waiting for a technical fix that was never the constraint.

Proposed next step: Assign a named adoption owner with a clear mandate. Run a two-hour co-design session with three frontline agents before any further technical work.

Governance process to accelerate

Meeting Summarization Approval

Current estimated approval time: 8 months. Employees are already using Otter.ai and ChatGPT for this — creating exactly the data governance exposure the approval process is designed to prevent. The slow process is not making the organization safer; it is pushing the risk into unmonitored personal channels.

Proposed next step: Fast-track a 90-day pilot with an approved vendor under a data processing agreement. Target: approved tool in production before employees find three more workarounds.