AMI

Automotive Market Intelligence
The Business Question

Which features should
go on the next model?

Every production cycle, OEMs decide which features to add, keep, or drop across trims and markets. Get it right and you win share. Get it wrong and you're either over-investing in features nobody values, or missing the ones they expect.

AMI answers this with data: 94,671 market signals, 10 markets, 110+ features, classified by customer importance using the Kano Model.
94,671Market Signals
11,189Companies
1,261Vehicle Models
110+Features Tracked
10Markets
The Framework

The Kano Model

Developed by Professor Noriaki Kano in 1984, this framework classifies product features by how they affect customer satisfaction. AMI applies it quantitatively: adoption rate across the competitive set determines the classification.

Must-Be (Expected)

Customers take these for granted. Missing them causes strong dissatisfaction, but having them doesn't create delight. 90%+ adoption in segment.

Airbags, ABS, Air Conditioning, Power Windows

One-Dimensional (Competitive)

More is better. Satisfaction scales with implementation quality. These are the battleground features. 30–90% adoption.

Fuel Economy, Cargo Space, Towing Capacity, Range

Attractive (Differentiator)

Unexpected features that delight when present. Highest ROI for differentiation. 5–30% adoption.

Head-Up Display, Hands-Free Liftgate, Massage Seats

Indifferent (Low Impact)

Customers don't care either way. Cost reduction candidates. Under 5% adoption.

CD Player, Analog Clock, Cigarette Lighter
Key insight: classifications shift by market. A feature that's "Expected" in the US may still be a "Differentiator" in India or ASEAN.
Actionable Output

Four things AMI tells you

Every feature, for every segment and market, gets a Kano classification, an ROI score, and a clear recommendation. Here's what that looks like in practice:

1. Feature ROI Ranking

Features ranked by return on investment. Kano impact × cost efficiency = asymmetry score. Low-cost Attractive features rank highest — maximum differentiation per dollar.
"Add Hands-Free Liftgate to Tucson SE: low cost, 12% adoption in segment, projected +$4.2M revenue."

2. Gap Alerts

Must-Be features you're missing — competitors all have them, your customers expect them. Regulatory gaps — a feature is Must-Be in one market but absent from your lineup in another.
"Auto Emergency Braking is Must-Be in EU (97% adoption). Your India lineup only offers it at 34%."

3. Cross-Market Insights

Same feature, different classification across markets. Identifies where your global feature strategy needs local adaptation: cost savings in markets where a feature is Indifferent, or first-mover plays in markets where it's still Attractive.

4. Competitive Positioning

Where do your trims sit versus competitors on the Kano map? Which competitors are over/under-investing in each category? Feature-by-feature competitive benchmarking, not just price-spec sheets.
Under the Hood

How AMI classifies features

The Kano classification isn't a survey. AMI calculates it from observed adoption rates across the competitive set. If 95% of SUVs in the US have Blind Spot Monitoring, that's a Must-Be. If only 18% have a Head-Up Display, that's an Attractive feature.

Classification logic: For each feature in a segment and market: 1. Count how many competing trims offer this feature 2. Divide by total trims in segment = penetration rate Thresholds: Penetration ≥ 90%Must-Be (everyone has it) Penetration 30–90%One-Dimensional (the battleground) Penetration 5–30%Attractive (opportunity to differentiate) Penetration < 5%Indifferent (nobody cares yet) These thresholds are calibrated against Kano survey results from J.D. Power and NCBS data. They can be adjusted per segment if needed.
The key question: where does AMI get the data to compute these adoption rates? That's where the data model and the 94K signals come in.
Foundation — The Data Model

A structured map of the automotive industry

To classify features, AMI first needs to know: which vehicles exist, who makes them, what features they have, and how they compete. This requires a precise data model (formally called an ontology) that defines every entity type and relationship.

Entities 6 types

Vehicle Model — RAV4, Model Y, Tucson, etc.
Feature — Lane Keeping Assist, AWD, Panoramic Roof
Company — OEMs, suppliers, regulators
Market Signal — news, filings, announcements
Location — markets and regions
Source — where the signal came from

Relationships 13 types

HAS_FEATURE — RAV4 has Lane Keeping Assist
MADE_BY — RAV4 is made by Toyota
COMPETES_WITH — RAV4 vs CR-V vs Tucson
SUCCEEDS — 2025 RAV4 follows 2024 RAV4
SUPPLIES — Denso supplies Toyota
+ 8 more linking signals to entities
This data model is the lever. Change what counts as an entity or relationship, and the entire downstream analysis — Kano classifications, ROI scores, competitive maps — reshapes.
Foundation — Signal Processing

94K signals become 537K connections

AMI reads every market signal and extracts structured connections. A single news article about Toyota's new hybrid RAV4 creates links between the company, the vehicle, its features, and the competitive set. Multiply by 94,671 signals and you get a dense, quantitative industry map.

Example: how one signal creates connections Signal: "Toyota launches new hybrid RAV4 for 2025" Toyota —MADE_BY→ RAV4 2025 RAV4 2025 —HAS_FEATURE→ Hybrid Powertrain RAV4 2025 —SUCCEEDS→ RAV4 2024 RAV4 2025 —COMPETES_WITH→ CR-V, Tucson, Forester Signal: "NHTSA mandates automatic emergency braking by 2026" NHTSA —AFFECTS→ all US-market vehicles Auto Emergency Braking reclassified as Must-Be in US Each connection carries a weight based on how many signals confirm it. More signals = higher confidence in the relationship.
Foundation — Data Cleaning

Cleaning: from noise to signal

Raw data is messy. "Toyota Motor Corp", "Toyota", and "TM" are the same company. The same news gets reported by 12 outlets. AMI runs a six-step pipeline to merge duplicates, link brand families, and remove noise — so the Kano classifications are based on clean data.

Step 1
Merge Duplicates
1,869 entities merged ("Toyota Motor" = "Toyota")
Step 2
Link Brand Families
Toyota → Lexus, VW → Audi, Hyundai → Genesis
Step 3
Roll Up Data
Sub-brand data flows to parent company
Step 4
Deduplicate Signals
Same story from 12 outlets → one signal
Step 5
Validate Rules
Every vehicle has a maker, every signal has a source
Step 6
Connect Orphans
Link isolated entities to nearest match

Before Cleaning

Entities12,560
Connections102,926
Duplicate companies~1,900

After Cleaning

Entities10,691
Connections78,087
Duplicate companies0
24K noisy connections removed, 6,361 meaningful patterns retained
Validation — How We Know the Data Is Good

Three quality scores

Kano classifications are only as good as the data behind them. AMI continuously measures data quality with three scores, each checking whether the cleaned data faithfully represents the real automotive market.

Completeness
Are all entities connected?

A vehicle with no manufacturer, or a company with no products, means missing data. Missing data means missing features in the Kano calculation — so completeness directly affects classification accuracy.

Score: 0.408 (0 = complete, 1 = many gaps)
Consistency
Do signals about the same entity agree?

If five articles say Toyota is launching a feature but one says they're cancelling it, that's an inconsistency. AMI measures this across all entity-signal pairs. High inconsistency means unreliable penetration rates.

Score: 0.117 (improved 41% after cleaning)
Predictiveness
Do high-urgency domains predict market direction?

Regulatory signals should predict which features become Must-Be. Supply chain disruptions should predict feature availability changes. If these correlations hold, the data captures real cause-and-effect.

Score: 0.582 (moderate — some domains more predictive than others)
Overall quality: 0.369 (7.3% improvement after cleaning). The consistency score improved 41% — meaning the Kano classifications are built on much more reliable signal data.
Validation — Structural Analysis

What the graph structure reveals

Beyond Kano classification, the structure of the industry map itself reveals competitive dynamics. AMI runs structural analysis to find patterns that simple feature counts would miss.

Competitive Clusters

Which OEMs have the most similar feature sets? The data naturally clusters competitors: Toyota-Honda-Hyundai share one pattern, BMW-Mercedes-Audi another. These clusters validate the Kano analysis — "Must-Be" should be consistent within a cluster.

Feature Migration Paths

Features move from Attractive → One-Dimensional → Must-Be over time. AMI tracks this migration across model years. Knowing where a feature is on this path helps predict when it becomes table stakes.

Supply Chain Dependencies

Which suppliers are critical to which features? If a single supplier dominates a Must-Be feature, that's a supply chain risk. AMI identifies these bottlenecks from the SUPPLIES relationships in the graph.

Regulatory Cascades

When NHTSA mandates a feature, it doesn't just affect the US. Euro NCAP and other bodies often follow within 2–3 years. AMI identifies features likely to cascade from Attractive to Must-Be across markets.
The industry map isn't just a database — it's a structure that encodes competitive relationships, supply dependencies, and regulatory pressure. Kano classifications emerge from this structure rather than being assigned manually.
Full Picture

End-to-end: signals to recommendations

From raw news articles to "add this feature to this trim" — here's the complete flow.

Collect
News & Filings
94K signals
Structure
Data Model
6 entity types, 13 relationships
Connect
Build Map
12K entities, 103K links
Clean
Deduplicate
6-step pipeline
Validate
Quality Scores
3 checks, 7.3% improvement
Classify
Kano Analysis
Per market & segment
Recommend
Feature ROI
Ranked actions

Data Coverage

Markets: US, JP, KR, CN, EU, UK, IN, CA, MX, ASEAN
Segments: SUV, Sedan, Pickup, Hatchback, Luxury, EV
Model Years: 2020–2026
Features: 110+ tracked per segment
Sources: News, earnings calls, regulatory filings, NCAP reports

Technical Detail

Databases: 3 SQLite databases (industry map, cleaned map, company directory)
Entity resolution: name matching at 92% similarity threshold
Signal processing: AI-powered extraction (NuExtract) + rule-based validation
Quality scoring: completeness + consistency + predictiveness
1 / 10 to navigate