Exploration vs Exploitation in Google Ads Portfolio Strategy

The Multi-Armed Bandit Problem (And Why It Matters For Your Ad Spend)

Imagine you're in a casino with 500 slot machines. Each machine has a different, unknown payout rate. You have €1,000 to spend.

What's your strategy?

Option A: Try every machine once, find the best, then play it exclusively
Option B: Play the first profitable machine you find and never try others
Option C: Split your budget - mostly on proven winners, but keep testing new machines

If you chose Option C, congratulations - you understand the explore-exploit tradeoff.

And here's the kicker: This exact problem is what your Google Ads campaigns face every day with your product catalog.

Except instead of 500 slot machines, you have 500 products. And instead of you making the decisions, Google's algorithm is making them for you.

And it's choosing Option B.

What Is the Explore-Exploit Tradeoff?

The explore-exploit tradeoff is a fundamental problem in decision-making under uncertainty. It comes from computer science and machine learning, but it applies to everything from restaurants to hiring to advertising.

Here's the dilemma:

Exploitation: Use your current knowledge to maximize immediate returns (order from your favorite restaurant)
Exploration: Try something new to gain information that might improve future returns (try a new restaurant)

Pure exploitation is safe but has a low ceiling. Pure exploration is risky and inefficient. The optimal strategy is a balance.

In the context of your Google Ads product portfolio:

Exploitation = Scaling proven products: Products that have already demonstrated strong ROAS, high conversion rates, and predictable performance. Safe, profitable, limited upside.
Exploration = Testing new products: Products with limited data, untested audiences, unknown performance. Risky, potentially unprofitable short-term, potentially transformative long-term.

Why Google Ads Over-Exploits

Google's algorithm faces the same explore-exploit tradeoff you do. But it makes different choices than you would - because it has different incentives.

Your Goals:

Long-term revenue growth
Portfolio resilience
Category leadership
New product success
Profit maximization

Google's Algorithm's Goals:

Short-term conversion maximization
ROAS target achievement
Certainty over opportunity
Statistical significance in days, not months
Revenue maximization (for Google, not you)

Notice the misalignment?

Google's algorithm is designed to achieve your stated objectives (hit your ROAS target, maximize conversions) in the shortest time frame possible. This naturally pushes it toward exploitation.

Why test 400 uncertain products when 20 proven products reliably hit your targets?

The Regret Minimization Problem

In machine learning, there's a concept called "regret" - the difference between your actual performance and the performance you could have achieved with perfect information.

Google's algorithm is optimized to minimize regret over short time horizons (days to weeks).

But your business needs to minimize regret over long time horizons (months to years).

Example:

Short-term regret minimization (Google's approach):

Day 1-7: Test 100 products
Day 8-30: Scale the top 10
Day 31+: Optimize those 10 relentlessly
Result: Minimum regret in Days 8-30, maximum ROAS achieved quickly

Long-term regret minimization (Your business needs):

Month 1-2: Test 100 products
Month 3-4: Scale top 20, keep testing next 100
Month 5-6: Discover Product #47 becomes your #1 seller
Month 7-12: Product #47 drives 30% of revenue growth
Result: Higher regret in Months 1-2, but discovered a gem you'd have missed

The algorithm can't see Month 7. It only sees this week's ROAS.

The Exploration Budget Framework

The solution is to manually add exploration back into your strategy. Here's the framework we use with clients:

The 75/25 Split

75% Exploitation Budget:
- Proven products (>1,000 impressions, >0.5% CTR, >1% conversion rate)
- ROAS target: Your normal target (e.g., 400%)
- Bid strategy: Maximize conversion value
- Goal: Reliable, predictable revenue
25% Exploration Budget:
- Unproven products (<1,000 impressions OR new to catalog)
- ROAS target: 50-60% of normal target (e.g., 200-250%)
- Bid strategy: Maximize conversions (gather data faster)
- Goal: Discover new winners, build data

This split isn't arbitrary - it comes from research on optimal exploration rates in multi-armed bandit problems. Studies show 20-30% exploration budget maximizes long-term returns in most scenarios.

Campaign Structure

Implement this as separate campaigns:

Campaign 1: Core Portfolio (Exploitation)

Products: Top 100 by revenue, last 90 days
Budget: 75% of total
ROAS Target: 400%
Bid Strategy: Target ROAS

Campaign 2: Discovery Portfolio (Exploration)

Products: All others (new products, low-impression products)
Budget: 15% of total
ROAS Target: 200%
Bid Strategy: Maximize Conversions

Campaign 3: High-Margin Rescue (Targeted Exploration)

Products: >40% margin, <500 impressions
Budget: 10% of total
ROAS Target: 250%
Bid Strategy: Maximize Conversion Value

Graduation Criteria

Products "graduate" from Exploration to Exploitation campaigns when they hit thresholds:

>2,000 impressions
>50 clicks
>5 conversions
ROAS >300%
Sustained performance for 30+ days

When a product graduates:

Move it to the Core Portfolio campaign
Update product labels in Merchant Center
Let it compete with proven winners

This creates a "portfolio funnel" where products progress from Discovery → Proven → Core.

Real Case Study: The Discovery Payoff

Client: Home goods retailer
Catalog: 724 products
Monthly spend: €180,000

Before Exploration Budget (Pure Exploitation):

Active products: 87 (12% of catalog)
Monthly revenue: €720,000
ROAS: 400%
Revenue growth: Flat for 6 months

Implementation:

Allocated 25% budget to exploration (€45,000/month)
Structured as 2 discovery campaigns + 1 core campaign
Lowered ROAS target on discovery campaigns to 225%

Month 1-2 Results (The Dip):

Blended ROAS: 340% (-15%)
Monthly revenue: €690,000 (-4%)
Active products: 87 → 134 (+54%)
New products tested: 187
CFO was nervous

Month 3-4 Results (The Recovery):

Blended ROAS: 385% (approaching baseline)
Monthly revenue: €740,000 (+3% over baseline)
Active products: 156
Products graduated to core: 23

Month 5-6 Results (The Payoff):

Blended ROAS: 410% (+2.5% over baseline)
Monthly revenue: €865,000 (+20% over baseline)
Active products: 178
Products graduated to core: 41
"Hidden gems" found: 7 products now in top 20

The compound effect: Those 41 new proven products became the growth engine. By Month 12, revenue was €1.1M/month (+53% over baseline) with the same €180K ad spend.

The Psychology of Exploration

The biggest barrier to exploration isn't technical - it's psychological.

Here are the objections we hear:

Objection 1: "We Can't Afford Lower ROAS"

Reality: You can't afford NOT to explore.

Calculate opportunity cost:

If you have 500 products and only 50 are active...
450 products × €2,000 potential monthly revenue per product = €900K hidden opportunity
Even if only 10% of those 450 become winners = €90K/month new revenue

The short-term ROAS dip is an investment, not a cost.

Objection 2: "We Already Tried This, It Didn't Work"

Reality: You probably didn't give it enough time or budget.

Common mistakes:

Testing with <10% budget (not enough to gather meaningful data)
Evaluating after 2-4 weeks (algorithms need 60-90 days)
Same ROAS target on exploration campaigns (defeats the purpose)
No graduation criteria (winners stay in exploration forever)

Objection 3: "Our Products Are Too Different to Compare"

Reality: That's exactly why you need exploration.

If products are similar: exploitation works fine
If products are diverse: exploration is critical (you don't know where winners hide)

Objection 4: "What Do I Tell My CFO?"

Use this framework:

"We're investing 25% of our ad budget in R&D - discovering which products have hidden potential. Based on industry data, we expect ROAS to dip 15-20% in Month 1-2, then recover and exceed baseline by Month 3-4 as we discover new winners. By Month 6, we project 15-25% revenue increase with the same total ad spend."

CFOs understand R&D budgets. Frame exploration as product discovery, not wasted spend.

Advanced Exploration Strategies

Once you've implemented basic exploration, here are advanced tactics:

1. Epsilon-Greedy Strategy

Borrowed from reinforcement learning:

90% of time: Use exploitation budget (proven winners)
10% of time: Use exploration budget (random product sampling)

Implement as:

Week 1-3: Exploitation only
Week 4: Exploration boost (2x normal exploration budget)
Repeat cycle

This periodic exploration pulse prevents algorithm stagnation.

2. Upper Confidence Bound (UCB) Approach

Prioritize products with high uncertainty:

Product Score = Estimated ROAS + (Confidence Interval Width × 0.5)

Products with <100 impressions get high uncertainty bonus
Forces exploration of under-tested products

3. Thompson Sampling

Probabilistic approach:

Each product has a probability distribution of potential ROAS
Sample from each distribution
Allocate budget proportional to sampled values
High-potential, high-uncertainty products get more budget

Hard to implement manually, but you can approximate by weighting margin + uncertainty in product labels.

4. Contextual Bandits

Exploration that considers context:

Season (winter products in winter)
Audience (show running shoes to fitness audiences)
Price point (match product price to audience income level)

Implement as audience-specific exploration campaigns.

Exploration in Different Campaign Types

Standard Shopping

✅ Easy to implement
✅ Full product control
✅ Clear performance visibility

Best for: Large catalogs, frequent testing

Performance Max

⚠️ Harder to implement
⚠️ Limited product control
❌ No product-level visibility

Strategy: Use asset group segmentation + hybrid approach with Standard Shopping

Smart Shopping (Legacy)

❌ Very hard to implement
❌ Black box optimization
❌ Consider migrating

Strategy: Migrate to Standard Shopping + Performance Max hybrid

Measuring Exploration Success

Track these metrics weekly:

Leading Indicators (Week 1-4):

Products with >100 impressions (should increase)
New products getting clicks (should increase)
Portfolio breadth (% catalog with impressions)
Discovery campaign impression share

Lagging Indicators (Month 2-3):

Products graduated to core (should be 3-10/month)
Blended ROAS (should recover to baseline)
Revenue from new products (should grow)
Portfolio concentration ratio (should decrease)

Success Indicators (Month 4-6):

Total revenue >baseline despite same spend
Active product count >2x baseline
Blended ROAS ≥ baseline
Top 20 products include discoveries from exploration

Common Pitfalls

1. Exploration Budget Too Small

15-25% minimum. Below that, you're not gathering enough data to matter.

2. Same ROAS Target Everywhere

Exploration needs 40-50% lower ROAS target or it just becomes exploitation.

3. No Graduation Process

Winners stuck in exploration campaigns can't reach their potential.

4. Giving Up Too Early

60-90 days minimum before evaluating. Month 1-2 will look bad - that's expected.

5. Exploration Without Strategy

Random testing ≠ strategic exploration. Prioritize by margin, stock, category gaps.

The Compound Effect of Continuous Exploration

Here's why exploration matters long-term:

Year 1 Pure Exploitation:

Q1: 50 active products, €800K revenue
Q4: 52 active products, €840K revenue (+5%)

Year 1 With Exploration:

Q1: 50 active products, €800K revenue
Q2: 73 active products, €780K revenue (dip during discovery)
Q3: 94 active products, €920K revenue (new winners scale)
Q4: 112 active products, €1.1M revenue (+38%)

Year 2 Comparison:

Pure Exploitation: €880K revenue (+5% YoY)
With Exploration: €1.5M revenue (+36% YoY)

The gap widens because:

You're continuously discovering new winners
Your active portfolio keeps growing
Algorithm has more data to optimize
You're less vulnerable to single-product dependency

Your Action Plan

Week 1: Audit current portfolio

Calculate active product %
Measure concentration ratio
Identify exploration opportunity

Week 2: Build business case

Project opportunity cost
Set success metrics
Get CFO buy-in on ROAS dip

Week 3: Implement exploration campaigns

Create 75/25 budget split
Set appropriate ROAS targets
Define graduation criteria

Week 4-12: Monitor and iterate

Track leading indicators weekly
Graduate winners monthly
Resist urge to optimize ROAS

Month 4: Evaluate and scale

Measure success metrics
Adjust exploration % based on results
Share wins with stakeholders

Want help implementing an exploration budget strategy? We offer Exploration Budget Planning that calculates your optimal split and builds the campaign structure for you.

Exploration vs. Exploitation: The Strategy Meta Ads Stole From Your Portfolio