Every CEO is being asked for more AI money
Do you know whether your AI is winning or losing your company customers?
AIPS is the one number that tells you — asked by the AI itself, the moment the conversation ends. Independent, comparable, trendable.
Learn more about a Sprint →“Which of my AI-powered high-value journeys are trusted enough to scale — and which do I need to fix first, or abandon altogether?”
Today nobody can answer it with confidence. The instrument below can.
The budget
The first wave of AI budgets was approved on the promise.
Spend is still rising, but the next tranche now has to clear cost and ROI review — and nobody can show which customer-facing AI is earning it.
Wave 1 · approved
Wave 2 · under review
of CEOs still report no significant financial benefit from AI; 12% see both cost and revenue gains
PwC Global CEO Survey, Jan 2026
of organizations can attribute any EBIT impact to AI — flat since 2025; just 6% are high performers
McKinsey State of AI, Aug 2026
have scaled AI across business units — and yet 85% plan to spend more in 2026
Gartner, Sep 2026
of customers used the company’s chatbot in their last service interaction — unchanged since 2022
Gartner customer survey, Sep 2026
The gap
AI now sits between you and your customers — and nobody measures what they think of it
It answers, decides and acts in your brand’s name, on borrowed judgment. One conversation sets the opinion; the opinion becomes your numbers — retention, referral, complaints, growth.
AI maturity models
Measure capability inward — strategy, data, talent, governance. Not one asks the customer.
NPS and CSAT
One brand number that blends human and machine. The AI’s share cannot be separated out.
Vendor dashboards
Volumes, resolution rates, deflection — activity and efficiency, not whether the customer trusted the outcome.
The customer’s side of the machine is not measured. That is the gap — and the whole idea.
Three decisions
After every AI conversation, the customer decides three things
Nobody asks them. Each decision shows up later — in the numbers, in the press, or not at all.
Will I recommend you?
Brand love and trust in the AI drift apart. The brand average hides the machine.
Example: Klarna →Do I trust what it did?
A working agent is often mistaken for a trusted one. When the judgment around it fails, it fails in public.
Detail: judgment →Will I come back?
Customers quit AI quietly. They don’t complain — they just don’t return.
Detail: ten reasons →27% of customers would try a chatbot again after one bad experience — and they rarely say so. Gartner, 2026
Decision 1 · Will I recommend you?
Klarna — what the public record shows, a year early
High advocacy and low trust do happen together. Advocacy is about the brand; trust is about the one conversation where the AI decided for you.
Klarna — advocacy high, trust low
Trustpilot · 600,000 reviews · 4.5
~4× one-star reviews naming the chatbot in 2024 — about 87,000 one-star reviews in total, clustered on refunds, disputes, charges and fees.
Star split: Trustpilot, Sep 2026. Chatbot share: our coding of one-star reviews, Sep 2026.
Feb 2024 · The cost case
2.3M conversations in month one — the work of ~700 agents. Projected $40M profit improvement.
2024 · The public record
4.5 stars, 76% five-star — yet 15% one-star, and the share naming the chatbot almost quadrupled.
May 2025 · The walk-back
The CEO called cost “a too predominant evaluation factor”. Human agents recruited back.
The cost axis alone green-lit scale. A trust axis would have said “fix first” a year before the public walk-back.
Decision 2 · Do I trust what it did?
Intelligence is not judgment — three things must stay human
Every agent acts on borrowed judgment. An agent without the right judgment does not stop; it makes something up — and then you scale fiction.
The steering wheel
Whose judgment is the agent borrowing?
Your best people’s — or the vendor’s defaults? One brand judgment across all agents, or each in its own voice?
The gearbox
Not “does the AI work?” — can it be trusted?
Clear without false certainty. Fair, explainable, reversible.
The driver
Does leadership own machine judgment?
Who signs the risk appetite — and who is accountable, by name?
When the judgment layer fails, it fails in public
None of these failed on model quality. Every one failed on the human judgment around the AI — a Steering, Gearbox or Driver gap, the three controls the AIPS judgment layer reads.
62% of large US companies now build or deploy AI agents; the share building controls into them fell to 30% from 43%. KPMG, Sep 2026
42% of companies expect to give AI agents real decision-making authority by 2030 — only 5% have the controls in place today. BCG Applied AI Index, Sep 2026
Air Canada — chatbot invented a refund policy; the airline called the bot “a separate legal entity”. The tribunal made it pay.
Gearbox · DriverCommonwealth Bank — cut 45 call-centre roles for an AI voice-bot. Call volumes rose; the bank reversed the cuts and called it an “error”.
DriverGerman clinic chatbot — gave its doctors specialist titles that do not exist. The Higher Regional Court of Hamm held the clinic liable for the bot’s claims.
Gearbox · DriverWoolworths “Olive” — customers reported the AI phone assistant insisting it was a real person. Woolworths confirmed the lines were scripted and removed them.
SteeringDecision 3 · Will I come back?
Ten reasons customers quit AI quietly
They don’t complain — they just don’t return. Most of the time nobody asks them why. Grouped by what customers want: effortless to have, certain it works, and trust in the outcome.
How it works
Three questions, one factor, one number
Asked by the AI itself, the moment the conversation ends. No new survey, no panel. Alternatively, the three questions ride on your existing NPS or CSAT survey.
Recommend
“Based on how this company uses AI to serve you — would you recommend it?”
Trust
“Do you trust what the AI decided — and how it got you there?”
Come back
“What should this company improve before AI helps you again? Tick all that apply — or none.”
× Judgment
Does the company’s own judgment deserve the score? Read from evidence, never self-reported. It can only lower the number.
= One number
−100 to +100
Score = 10 − ticks. Comparable across journeys, business units and companies; trendable quarter by quarter. Plus a ranked fix list.
Independent — produced by nobody who sells you AI; the work that follows goes to your AI provider. Complementary to NPS: NPS reads brand advocacy; AIPS isolates the AI signal.
What you get
Three readings from one instrument
Where to put the next tranche of AI capital, what customers want fixed, and whether your own judgment is inside the agents.
1 · CEO prioritization matrix
Every AI initiative on one grid — value at stake against trust in the AI. The quadrant is the capital verdict.
2 · Trust heat map
What customers want fixed before the AI helps them again — three categories, read per journey.
3 · Judgment heat map
Steering, Gearbox, Driver — read from evidence per journey. Weak controls discount the score.
Illustrative positions.
The matrix says where the next tranche goes; the heat maps say what to fix to earn it — an independent brief for whoever builds your AI.
The Pioneer Sprint
Four weeks, no fee, results before the premiere
For a handful of companies with AI agents live in front of customers. One per industry.
Week 1
Pick 2 or 3 high-value AI journeys. Switch the questions on in your agent.
Weeks 2–3
Customers answer. We interview a few people and read the evidence for the judgment factor.
Week 4
Score, heat maps, fix list. One hour with the executive sponsor.
What you give
2–3 hours of one executive sponsor’s time, a contact to switch the capture on, and the agent’s operational owner to discuss improvement actions.
What you get
Your AIPS — brand and per journey — plus the trust heat map, the judgment read and a customer-ranked fix list.
Ask about an open slot
No fee for pioneers. Results private; recognition at the world premiere on November 5 only with your written approval.
Learn more about a Sprint