WORKSHOP · PROCUREMENT · WS·05

The AI Grifter Test

Five red flags that catch more bad AI deals than any technical due diligence — spotted in a single week, by a non-technical CEO. 12 slides, 30 minutes.

12 SLIDES · ~30 MIN · PLAYBOOK OVERVIEW
Key Takeaways
  • The CEO is the last line of defense. Five red flags, spotted in a single week, will catch more bad deals than any technical due diligence ever will.
  • 10% of AI value is the algorithm — the model itself. This is the part vendors love to demo.
  • 20% is the data — preparation, access, permissions, pipelines. The unglamorous, unavoidable middle.
  • 70% is process and governance change — interviewing operators, redesigning workflows, training people, rewriting policy. The part that actually pays.
  • Pull the last AI proposal in your inbox and run the five-flag scorecard on it before lunch. Twenty minutes, not a project. The first proposal you score will tell you whether you have a vendor problem or a process problem.
WORKSHOP · PROCUREMENT DEFENSE

The AI Grifter Test

How to buy AI services without getting taken. Five red flags. One week. One signature you don't put down.

BuildClub Academy
01 · WHY NOW

The grift works because the failure rate gives it cover.

95%
Zero P&L return
of enterprise generative AI pilots deliver no measurable P&L return. When failure is the base rate, a vendor who delivers nothing looks statistically normal. (MIT Project NANDA, 2025.)
42%
Scrapped initiatives
of companies scrapped their AI initiatives in 2025 — up from 17% the year before. The average sunk cost per abandoned project runs into the millions. (S&P Global.)
40%
Agentic projects canceled
or more of agentic AI projects will be canceled by 2027 — escalating cost, unclear business value, inadequate risk controls. (Gartner.)
67%
Vendor partnerships succeed
buying from specialized vendors / partnerships succeeds at roughly two-thirds rate; internal builds succeed at one-third of that. The number that flips the script in your favor. (MIT NANDA.)

The base rate is brutal — but it is brutal for a reason. Bad integration, not bad models. The grifter sells you the model story to distract you from the integration work.

02 · THESIS

The CEO is the last line of defense.
Five red flags, spotted in a single week,
will catch more bad deals than any
technical due diligence ever will.

The Thesis
03 · MENTAL MODEL

You're not buying technology. You're buying organizational change that happens to use technology.

Takeaways
  • 1 10% of AI value is the algorithm — the model itself. This is the part vendors love to demo.
  • 2 20% is the data — preparation, access, permissions, pipelines. The unglamorous, unavoidable middle.
  • 3 70% is process and governance change — interviewing operators, redesigning workflows, training people, rewriting policy. The part that actually pays.
  • 4 Grifters always sell the 10% wrapped in the language of the 70%. The model demo is dressed up as 'transformation.' Your job is to ask: where in this proposal is the 70%?

Pick your partner like you pick an ERP, not a model. The model is becoming a commodity; the service and integration layer is the lock-in. Evaluate the relationship and the exit, not the demo.

04 · THE FIVE RED FLAGS

Red flags 1 and 2: speed and shape of what they're proposing.

1
They closed you in one or two meetings
Real workflow transformation requires process mapping — interviewing operators, watching where humans act as middleware. A vendor who scopes a transformation from one discovery call has not looked at your workflows. They have a template. Speed of close is inversely correlated with quality of diagnosis.
Tell
What you're looking for on Flag 1
No request to interview your operators. No artifact — process map, swim lane, day-in-the-life — that proves they understood your specific process. Glossy proposal, generic verbs, no named people.
2
They're proposing to build you something proprietary
MIT data: internal and custom builds fail at roughly twice the rate of vendor-led, platform-based solutions. A vendor steering you to a bespoke build is either inexperienced or manufacturing lock-in and billable hours. The course thesis holds: buy the platform, build the integration.
Tell
What you're looking for on Flag 2
The language is 'we'll build you a custom model / agent / platform' instead of 'we'll integrate a proven platform into your workflow.' The named platform is conspicuously absent. Their IP, not theirs-plus-yours.

Flags 1 and 2 are the easiest to catch — they're visible in the first meeting. Most grifts die here if you're paying attention.

05 · THE FIVE RED FLAGS, CONTINUED

Red flags 3 and 4: exit and economics.

3
The deliverable is murky and the technology is opaque
If you cannot see how you would leave their platform, assume the opacity is intentional. Gartner's 40%-discontinued number is the downstream cost of opacity — projects nobody can evaluate eventually get killed. Portability is the antidote. The exit clause is the smoke test.
Tell
What you're looking for on Flag 3
You cannot get a one-sentence answer to 'How do I migrate off you?' or 'What do I keep if we part ways?' The vendor pivots to features when you ask about exit. Their data model is undocumented. Your data lives on their substrate.
4
They want significant payment up front
Serious vendors stage payments against verifiable deliverables. Large up-front payment transfers all the risk to you before any value is proven. The payment schedule is a confidence test the vendor takes on their own product. Front-loaded cash means they want their money before you find out what you bought.
Tell
What you're looking for on Flag 4
Money is front-loaded. Milestones are vague. There is no 'we don't get paid for phase 2 until phase 1 is verified by you.' Discount offered for paying earlier — the vendor is hedging against you walking away mid-project.

Flags 3 and 4 are about asymmetry. They want all the leverage — opaque tech and front-loaded cash — and they want you to give it to them on faith.

06 · THE FIVE RED FLAGS, COMPLETE

Red flag 5, and the synthesis: any two and you walk.

5
The proposal has no real data-work line item
Industry consensus: data preparation is 70–80% of any real AI project. If data work is not a budgeted, staffed line item, the vendor either does not understand the work or is hiding the hardest, least glamorous part of it until you have already signed the contract.
Tell
What you're looking for on Flag 5
A polished model and UI scope with no data audit, no cleaning, no pipeline work, no access-permission scope priced in. The data work is described as 'leveraging your existing data' — a phrase that means nothing and budgets nothing.
Sum
The synthesis — any two, walk
Closed fast · proprietary build · opaque exit · front-loaded cash · no data line. Any two of these in the same proposal and you walk to the elevator politely. Real vendors fail one of these occasionally and acknowledge it. Grifters fail multiple and try to explain them away.
Why
Why two and not one
Even good vendors sometimes propose a custom build, or want partial up-front cash for a small engagement. One flag is a conversation; two is a pattern. The pattern is the signal — the grift is structural, not incidental.

Borrowed cadence from the company-brain rule: if a vendor can't tell you in one sentence how their system inherits your permissions, walk them to the elevator. Same energy here.

07 · DECISION FRAMEWORK

The buyer's scorecard: five questions, one page, go or no-go.

Diagnosis
"Walk me through how you'll map our workflow before scoping." Clean: named operators, ride-alongs, a delivered process artifact. Dirty: "We'll start with a discovery call."
Buy vs. Build
"Which proven platform are you integrating, and why that one?" Clean: a named platform plus an integration plan. Dirty: "We'll build you a custom solution."
Exit
"In one sentence: how do we leave you, and what do we keep?" Clean: a real migration path; our data stays ours. Dirty: a long pivot to features.
Payment
"Tie payment to milestones — what's verified before phase 2?" Clean: staged payments against deliverables. Dirty: front-loaded cash, vague milestones.
Data
"Show me the data-preparation line item." Clean: a staffed, budgeted 60–80% data scope. Dirty: "We'll leverage your existing data."
08 · FAILURE MODES

Even clean deals oversell. Even careful buyers get burned.

1
Klarna — the walk-back
Best-in-class platform-native AI deployment. Replaced ~700 service roles with AI, then rehired humans after customer satisfaction quality dropped. CEO Sebastian Siemiatkowski on the record: 'We focused too much on efficiency and cost. The result was lower quality, and that's not sustainable.' Even a clean vendor can oversell autonomy. Scope for human-in-the-loop. Stage the cut.
2
IBM Watson Health — the enterprise ancestor
$62M spent, zero patients treated. The grifter's enterprise-scale ancestor. Wrong leadership (sales-led, not workflow-led), wrong scope, no operational integration. The lesson isn't that AI is hard — it's that proposals divorced from real workflows fail at the same rate whether the vendor is a grifter or a Fortune 500 partner.
3
The internal-build trap
Hiring the wrong leader — a systems-builder CTO instead of a workflow-redesign leader — produces an 18-month pilot that never touches a real workflow. The same failure a grifter sells you, just self-inflicted. The 10-20-70 rule applies to your own team too. If your AI lead is selling you the 10%, fire them or reframe the role.

Three different failure paths, one common cause: the proposal (external or internal) was built around the model, not the workflow. The grifter test catches the external version. The CTO question catches the internal version.

09 · PROCUREMENT UPGRADE

Procurement upgrade: build a framework, not a roster.

1
2
3
4
5
1
One capability bar
Define what good AI vendor work looks like — the same five red flags applied as positive criteria. This is the qualifying bar every vendor must clear before they can quote a workload.
2
One contract template
Staged payment, exit/portability clause, named data-preparation scope, IP ownership, audit rights. Write it once, sign it five times with five vendors. The legal work is amortized; the negotiation is structural.
3
Multiple vendors qualified
Don't pick a vendor — qualify three to five. Each can quote any workload. The competitive market does the price discovery. The framework does the quality control.
4
Workloads portable
Every contract requires the work product to be portable — data formats, integration code, model artifacts. A new vendor signs and gets work; a laggard falls behind and gets de-prioritized. No emergency board meeting required.
5
Renewal is the lever
Use-case scoping and portability go into the contract at renewal, not into a policy document nobody enforces. Renewal is the only time you have leverage. Use it.

Borrowed from 'Eight AI Vendors, One Customer — The Procurement Lesson.' The CIA, the Pentagon, the Fortune 50 — the buyers who get this right don't pick a vendor. They build a market, then they shop in it.

10 · CEO INSPECTION

Five questions to ask before your next signature.

1
2
3
4
5
1
How will you map our workflow before scoping?
Listen for named operators, ride-alongs, a process artifact. Vague answers mean templates, not diagnosis. Templates are the grifter's tell.
2
Which proven platform are you integrating, and why that one?
Listen for a named platform with a real integration story. 'We'll build you something custom' is a 2x failure rate dressed up as innovation.
3
In one sentence, how do we leave you and what do we keep?
Listen for a real migration path. Pivots to features or 'we'll cross that bridge' mean the lock-in is the product.
4
How is payment staged against verified deliverables?
Listen for milestone gates — 'phase 2 starts when you verify phase 1.' Front-loaded cash and vague milestones are the structural grift.
5
Show me the data-preparation line item.
Listen for a budgeted, staffed 60–80% scope. 'We'll leverage your existing data' means they haven't priced the actual work.

Ask all five in the same meeting. Watch how the vendor reacts. The reaction is the data — more useful than any answer.

11 · MONDAY-MORNING ACTIONS

Three moves for Monday morning.

Monday Morning
  1. 1
    Pull the last AI proposal in your inbox and run the five-flag scorecard on it before lunch. Twenty minutes, not a project. The first proposal you score will tell you whether you have a vendor problem or a process problem.
  2. 2
    Add three clauses to your AI procurement template at the next renewal: staged payment against verified deliverables, a one-sentence exit and portability clause, and a named data-preparation line item with a budget.
  3. 3
    For the next AI vendor conversation, ask for the workflow process-change plan — the 70% — not the model demo. If they can't produce it, you have your answer before the proposal arrives.
WS·05 · Questions CEOs Ask

Frequently Asked Questions

What is the core idea of The AI Grifter Test?
The CEO is the last line of defense. Five red flags, spotted in a single week, will catch more bad deals than any technical due diligence ever will.
What does the data say a CEO should pay attention to?
10% of AI value is the algorithm — the model itself. This is the part vendors love to demo. 20% is the data — preparation, access, permissions, pipelines. The unglamorous, unavoidable middle.
What should a CEO do Monday morning after reading The AI Grifter Test?
Start here: Pull the last AI proposal in your inbox and run the five-flag scorecard on it before lunch. Twenty minutes, not a project. The first proposal you score will tell you whether you have a vendor problem or a process problem; Add three clauses to your AI procurement template at the next renewal: staged payment against verified deliverables, a one-sentence exit and portability clause, and a named data-preparation line item with a budget; For the next AI vendor conversation, ask for the workflow process-change plan — the 70% — not the model demo. If they can't produce it, you have your answer before the proposal arrives.
What are the steps in The AI Grifter Test?
1) One capability bar; 2) One contract template; 3) Multiple vendors qualified; 4) Workloads portable; 5) Renewal is the lever.
Where do the claims in The AI Grifter Test come from?
The playbook cites MIT Project NANDA — The GenAI Divide: State of AI in Business 2025; Fortune — MIT NANDA 95% report; HBR — Beware the AI Experimentation Trap; Trullion summary — Gartner 40% canceled.
WS·05 · Sources

All Sources in This Playbook

00 / 12