Instrument
Trust Premium Score
The Trust Premium Score measures how much value an organization earns from AI that people trust, across 15 dimensions in three pillars: risk avoided, performance gained and market value earned. Each dimension is scored from 1 to 5 on what runs today, so the total falls between 15 and 75 and places the organization in one of four bands.
Last verified
Scoring
| Dimension | Score 1 | Score 3 | Score 5 |
|---|---|---|---|
| Pillar 1: Risk avoided (the floor) | |||
| 1.1 Regulatory readiness | No regulatory tracking; AI ships without legal review. | Regulation tracked in key jurisdictions; legal review before deployment. | Compliance infrastructure reused across jurisdictions; the organization helps shape policy. |
| 1.2 Incident preparedness | Customers or the press find AI failures; no post-incident process. | Incident classes, a response playbook and reviews for high-severity events. | Preventive design keeps incidents rare; response is rehearsed and lessons are shared. |
| 1.3 Governance maturity | No governance structure; whoever has access deploys AI. | Named owners, an AI system inventory, risk tiers and deployment gates. | Board-level reporting; governance makes deployment faster, not slower. |
| 1.4 Data protection | Training data provenance unknown; shadow AI reaches uncontrolled data. | Documented provenance, tracked consent, privacy impact assessments for new systems. | Privacy by design in the AI architecture; data governance enables new work. |
| 1.5 Compliance readiness | No documentation of AI systems or decisions; a regulator’s question could not be answered. | Standard model documentation for every production system, with audit trails. | Continuous compliance monitoring; any regulatory inquiry answered within 48 hours. |
| Pillar 2: Performance gained (the engine) | |||
| 2.1 AI adoption rate | AI confined to one team or pilot; shadow AI exceeds sanctioned use. | AI in three to five core functions under coordinated governance. | AI is a core operating capability across the organization. |
| 2.2 Deployment velocity | Projects take 12 months or more; most pilots never ship. | Three to six months to production, with governance checkpoints set up front. | Governance as code; low-risk models reach production in days. |
| 2.3 Model reliability | Production performance unmonitored; users find the failures. | Baselines for every production model, regular monitoring and drift detection. | Automated quality checks; performance data feeds back into the standards. |
| 2.4 Cross-functional trust | Business leaders distrust AI outputs and verify every decision by hand. | Legal, risk and business owners help design and review AI systems. | AI-informed decisions are the default; human override is the exception. |
| 2.5 Innovation velocity | No route for experiments; each new idea waits months for approval. | Light-governance sandboxes with clear criteria for moving to production. | A steady pipeline of experiments, known in the industry for speed and care. |
| Pillar 3: Market value earned (the moat) | |||
| 3.1 Customer trust perception | Customers distrust the organization’s AI; opt-out rates are high. | AI practices explained to customers; customer trust measured quarterly. | Customers share data willingly because they trust its stewardship. |
| 3.2 Brand differentiation | No trust positioning, or claims that practice contradicts (ethics washing). | Published principles, governance documentation or third-party certification. | The organization sets its industry’s standard; competitors benchmark against it. |
| 3.3 Talent attraction | AI talent avoids the organization; turnover runs above average. | Governance is part of the employer brand; candidates ask about it. | A destination employer for AI talent because of its trust record. |
| 3.4 Partner ecosystem | Partners will not share data or co-develop because of governance gaps. | Governance maturity supports standard data-sharing agreements. | Partners join to reach the governed data and AI infrastructure. |
| 3.5 Investor confidence | Investors see AI as unmanaged risk; gaps show up in due diligence. | The board has a defined AI oversight mechanism with regular reporting. | Analysts cite AI governance as a strategic asset of the company. |
Score each dimension from 1 to 5 on evidence of what operates today, not on plans. A policy nobody enforces scores the same as no policy. Add the five scores in each pillar (5 to 25), then add the three pillars for the total (15 to 75). Scores of 2 and 4 sit between the descriptions.
| Total | Band | What it means |
|---|---|---|
| 15-25 | Trust Deficit | AI is a liability. Build the floor first: an AI system inventory, a named owner for each system and an estimate of the Pillar 1 exposure. |
| 26-45 | Trust Neutral | Compliant, with no advantage from trust yet. Fix the weakest dimension in each pillar and start measuring what trust work changes in deployment speed and adoption. |
| 46-60 | Trust Positive | Trust is paying back. Push Pillar 2 and 3 dimensions in the highest-value AI system and use governance maturity to win partnerships. |
| 61-75 | Trust Premium Leader | Trust is a moat. Red-team the trust infrastructure for single points of failure and report the premium to the board with hard numbers. |
Download: Trust Premium Assessment Worksheet
The printable worksheet: every dimension, the score bands and room for your team's evidence.
Enter your email to get instant access. You'll also receive the fortnightly newsletter.
Free. No spam. Unsubscribe anytime.
Why this instrument exists, and what the scores mean for a leadership team: The Trust Premium: Why Trusted AI Is Worth More
Reading the score
Score as a cross-functional team: technology, legal, risk and the business. A single function scoring itself tends to score high.
Then read two things. The band says where the organization stands. The pillar balance says what kind of problem it has. Two organizations can both total 50: one at 22, 15 and 13 has protected itself but captures little value, while one at 12, 20 and 18 captures value on an exposed floor. A Trust Premium Leader scores 20 or more on every pillar, not 25 on one and 12 on another.
After scoring, apply the Family Test. Would you trust this organization’s AI with your family’s financial data (Pillar 1)? Would you rely on it for a decision about your family’s health (Pillar 2)? Would you recommend its AI products to your family (Pillar 3)? If the score says adequate and the answer is no, rescore.
The 90-Day Trust Sprint
A score without a plan is an audit. The sprint turns it into a quarter of work, in the spirit of Minimum Viable Governance: start with the smallest structure that works and mature it through practice.
Weeks 1 to 2: audit
Fill in the worksheet with the cross-functional team. Record the band and the three weakest dimensions. In Trust Deficit, also build the AI system inventory, because nobody can govern what they cannot see.
Weeks 3 to 4: prioritize
Pick three dimensions to move. In Trust Deficit, take them from Pillar 1, since the floor has to hold first. In Trust Neutral, take the weakest dimension in each pillar. In Trust Positive, take the Pillar 2 and 3 dimensions that speed up adoption and set the organization apart.
Weeks 5 to 8: implement
Give each chosen dimension a target one level up, such as from 2 to 3. The rubric is the plan: each level describes what has to be documented, made routine or owned.
- Trust Deficit: finish the AI inventory, name a governance owner for each system, estimate the Pillar 1 exposure, draft the incident playbook and run the Family Test on every production system.
- Trust Neutral: put governance checks into the deployment pipeline, automate checks for low-risk systems, start counting incidents avoided and time saved, and launch AI literacy training.
- Trust Positive: run the trust and adoption loop on the highest-value AI system, use governance maturity to secure a partnership and publish the governance practices.
- Trust Premium Leader: red-team the trust infrastructure, take part in standard-setting and report the premium to the board.
Weeks 9 to 12: measure
Rescore all 15 dimensions and track the change. The aim is the next band, not the highest number, because each band opens different options: leaving Trust Deficit stops the losses, and reaching Trust Positive starts the payback.
| Industry | What raises the bar | Pillar 1 | Pillar 2 | Pillar 3 | Leader total |
|---|---|---|---|---|---|
| Financial services | Credit, fraud and trading models carry AI-specific rules | Above 22 | Above 20 | Above 19 | Above 61 |
| Healthcare | Life-safety stakes and clinical validation | Above 23 | Above 19 | Above 18 | Above 60 |
| Government | Public accountability; failures become political events | Above 22 | Above 18 | Above 17 | Above 57 |
| Consumer technology | Brand sensitivity; trust failures and wins spread fastest | Above 20 | Above 22 | Above 21 | Above 63 |
These thresholds are the framework’s own calibration from each sector’s regulatory pressure and trust sensitivity, not survey data. Use them to read a score in context, not to excuse it.
| Finding | Figure | Source |
|---|---|---|
| What broken trust costs (Pillar 1) | ||
| Maximum EU AI Act fine for a prohibited AI practice; other listed obligations, transparency included, carry up to EUR 15 million or 3%. Fines apply from 2 August 2025 | Up to EUR 35 million or 7% of worldwide annual turnover, whichever is higher | Regulation (EU) 2024/1689, Article 99 |
| Extra breach cost at organizations with high use of unsanctioned (shadow) AI, against low or none | $670,000 per breach ($4.74 million against $4.07 million) | IBM, Cost of a Data Breach Report 2025 |
| Organizations reporting a breach of their AI models or applications, and the share of those without AI access controls | 13%, of which 97% had no AI access controls | IBM, Cost of a Data Breach Report 2025 |
| AI-related incidents reported to the AI Incident Database in 2024 | 233, up 56.4% on 2023 | Stanford HAI, AI Index 2025 |
| AI-related securities class action filings in the US | 7 in 2023, 15 in 2024 | Cornerstone Research, Securities Class Action Filings: 2024 Year in Review |
| Australia’s Robodebt scheme: money refunded to people and debts written off after the automated debt-raising was found unlawful | A$746 million refunded to about 381,000 people; A$1.751 billion in debts written off | Royal Commission into the Robodebt Scheme, Report, 2023 |
| What earned trust returns (Pillar 2) | ||
| Operating profit attributable to AI at organizations spending 10% or more of their AI budget on ethics, against those spending 5% or less (average over two years; a correlation) | 30% higher | IBM Institute for Business Value, The AI ethics trust engine, 2025 (915 executives) |
| Self-reported improvement over 12 months among organizations investing in AI ethics: customer satisfaction and retention, AI ethics incident prevention, AI adoption | +22%, +20%, +19% | IBM Institute for Business Value, The AI ethics trust engine, 2025 |
| US business leaders who say responsible AI boosts ROI and efficiency; who report better customer experience and innovation | 58%; 55% | PwC, 2025 US Responsible AI Survey (310 leaders) |
| What demonstrated trust is worth (Pillar 3) | ||
| Return on equity of companies whose boards are both digitally and AI savvy, against industry average (non-savvy boards: 3.8 points below) | 10.9 percentage points above | MIT CISR, Digitally Savvy Boards: AI Update, 2025 |
| Boards that were both digitally and AI savvy in 2024 | 26% | MIT CISR, Digitally Savvy Boards: AI Update, 2025 |
| People who say they trust AI, across 28 markets; in the US; in China | 49%; 32%; 72% | Edelman Trust Barometer 2025 |
The return and value figures are correlations: organizations that invest in trust may also be better run. Read them as a consistent direction across independent sources, not as proof of cause.
Sources
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), Articles 99 and 113. Official Journal of the European Union, 2024-07-12
- Cost of a Data Breach Report 2025: The AI Oversight Gap. IBM and Ponemon Institute, 2025-07-30
- IBM Report: 13% of organizations reported breaches of AI models or applications, 97% of which reported lacking proper AI access controls. IBM Newsroom, 2025-07-30
- The AI Index 2025 Annual Report, Responsible AI chapter. Stanford HAI, 2025
- Securities Class Action Filings: 2024 Year in Review. Cornerstone Research, 2025-01
- Report of the Royal Commission into the Robodebt Scheme. Commonwealth of Australia, 2023-07-07
- The AI ethics trust engine. IBM Institute for Business Value with the Notre Dame-IBM Technology Ethics Lab, 2025
- 2025 US Responsible AI Survey. PwC, 2025
- Digitally Savvy Boards: AI Update, Research Briefing XXV-3. MIT Center for Information Systems Research, 2025-03-20
- 2025 Edelman Trust Barometer: Insights for the Technology Sector. Edelman, 2025-02
- The AI Trust Imperative. Edelman, 2025-02-13
Ajay's views, from 15 years in the field. Not legal or compliance advice. See full disclaimers →
Published by AI Exponent LLC