The AI Wedge Strategy

TL;DR

A wedge is one specific use case meeting three criteria:

Material economics — changes a unit cost or unit revenue line by 30%+.
6-month proof — measurable outcome within 6 months.
Compounds — winning here creates positions for follow-on bets.

Pick 2–3 wedges. Concentrate resources. Kill what isn’t working. The wedge approach beats “broad AI investment” by 3–5x in observed outcomes for similarly-sized companies.

One use case. Three criteria. The wedge approach beats broad AI investment by 3–5x in companies that execute it.

The “broad AI investment” pattern shows up at most enterprises: small AI initiatives across many functions, modest results across the board, no clear wins. The alternative is the wedge: concentrate on 2–3 specific use cases that meet a high bar, win in those decisively, compound from there. This piece is the wedge frame applied.

What makes a wedge

Criterion 1: Material economics

The use case has to change something material. Specifically: a unit cost or unit revenue line changes by 30% or more.

Examples:

Customer support cost / ticket drops from $12 to $4.
Sales cycle time drops from 60 days to 35 days.
Underwriting cost / policy drops from $200 to $40.
Marketing campaign cost / lead drops from $150 to $50.

Below 30%, the impact gets lost in noise. Above 30%, the impact is visible enough to drive decisions.

Criterion 2: 6-month proof

The wedge has to produce a measurable outcome within 6 months. This forces concrete bets — vague wedges don’t pass the proof test.

The 6-month constraint matters because:

AI capability changes faster than 12-month wedges can keep up.
Decisions need to be made on observed evidence, not projections.
6 months is enough to learn meaningfully but not so long that opportunity cost dominates.

Criterion 3: Compounds

Winning the first wedge creates positions for the next. The wedge isn’t a one-shot; it’s the first move in a sequence.

Examples of compounding:

Customer support wedge wins → operational data flywheel → customer success agent next → onboarding agent after.
Sales productivity wedge wins → CRM enrichment platform → sales coaching agent → pipeline forecasting agent.
Underwriting wedge wins → fraud detection wedge → claims agent next → pricing optimization later.

The compounding criterion separates wedges (strategic) from one-off use cases (tactical).

The wedge selection process

Three steps.

Step 1: Generate candidates

Brainstorm 8–15 candidate wedges. Sources:

Where is your cost or revenue line vulnerable to AI?
Where do you have unique data that creates AI advantage?
Where do customers complain about cost or experience?
Where are competitors moving with AI?

Be inclusive at this stage. Filter later.

Step 2: Apply the three criteria

For each candidate:

Material economics: estimate the unit-line change. >30%? Pass.
6-month proof: can you ship a meaningful pilot in 6 months? Pass.
Compounds: what’s the next 3 things this enables? Specific? Pass.

Most candidates fail at least one criterion. The 2–3 that pass all three are the wedges.

Step 3: Concentrate resources

Pick 2–3 wedges. Allocate disproportionate resources. The mistake is to do “small bets across many” — wedges require concentration to win.

Typical allocation: 60–70% of AI investment goes to the wedges; 30–40% to operational efficiency and infrastructure.

What the execution looks like

The wedge isn’t a pilot; it’s a strategic bet. The execution differs.

1. Senior leadership sponsorship. The wedge needs an exec sponsor with skin in the game (CTO, COO, CRO depending on the wedge).

2. Dedicated team. Not a side project. A team of 5–10 people for 6 months, with full mandate.

3. Production from day one. Not pilot first then production. The bet is to ship to production with measured outcomes.

4. Metric review monthly. Specific outcome metrics reviewed monthly with leadership. Decisions made on data, not projections.

5. Kill or expand at 6 months. At the 6-month mark, the wedge has either produced the material outcome or it hasn’t. Decisive go/kill.

What to do this quarter

Run the wedge generation exercise with senior leadership. 8–15 candidates.
Apply the three criteria. Filter to 2–3 that pass.
Sponsor and resource the wedges. Concrete teams, concrete dollars, concrete sponsors.
Set the 6-month review. On the calendar.

Counter: what about the broad small bets approach?

The broad approach has appeal: distributed risk, learning across many use cases, organizational engagement. The empirical pattern is less appealing — broad approaches typically produce modest outcomes everywhere and decisive wins nowhere.

The exception: companies in the very early stages of AI literacy may benefit from broad exposure for 6–12 months before consolidating to wedges. After that initial phase, concentration wins.

Wedge anti-patterns

Three to avoid.

Anti-pattern 1: The “platform wedge”

“Our wedge is the AI platform.” No — a platform is infrastructure, not a wedge. Wedges are specific use cases that produce outcomes; platforms enable wedges.

Anti-pattern 2: The “everything wedge”

“Our wedge is to make every team AI-enabled.” No — that’s a transformation initiative, not a wedge. Wedges are specific.

Anti-pattern 3: The “competitive parity wedge”

“Our wedge is to match what competitors are doing with AI.” No — wedges should change your specific economics, not match someone else’s. Following isn’t a wedge.

How wedges relate to other strategy artifacts

The wedge fits into the broader frame from AI Strategy for CTOs and CEOs:

The 12-month plan names the wedges explicitly.
The 36-month investment frame allocates capital to wedges.
The quarterly review tracks wedge progress and adjusts.

The wedge is the unit of strategic action. Everything else is supporting structure.

FAQ

Can a company have only one wedge? For smaller companies (under 1000 employees), one wedge is usually right. For larger, 2–3 lets you accept some bet failures.

What if our wedge fails? Kill it; learn from it; pick a new one. The portfolio approach assumes some failures. The discipline is to recognize failure quickly and reallocate.

What if multiple wedges pass the criteria? Pick the two with the highest compounding potential. Save the others for the next cycle.

How does this work for B2B vs. B2C? The principle is the same; the wedges differ. B2B wedges often involve customer-facing capabilities or operations. B2C wedges often involve personalization, support, or product capability.

What about defensive wedges? Yes — preventing competitive disruption can be a valid wedge if the criteria pass. “AI parity in customer support so we don’t lose accounts” is a real wedge if customer churn is on the line.

Working with JAIN on AI wedge strategy? We help executive teams identify and resource the 2–3 wedges that beat the broad-investment alternative. Book a 30-minute call.

Related reading: