AI vs Manual Google Ads Management: 6 Months of Testing, Here's What I Learned

Methodology: 12 e-commerce accounts, average budget $9,000/month, varied industries (fashion, home, sports, beauty). On each account, I compared:

- **Bidding**: Smart Bidding (tROAS / tCPA) vs enhanced manual CPC - **Campaigns**: Performance Max vs classic Search + separate Shopping - **Targeting**: AI Audience Targeting vs manual targeting - **Ads**: RSA automatic optimization vs manual pinning of best elements

Test duration per configuration: minimum 6 weeks (learning phase + period for significant data).

Disclaimer: these results apply to mid-market e-commerce accounts. Your results may vary based on industry, account maturity, and data volume.

**Verdict: AI > Manual for accounts with 50+ conversions/month**

On 8 accounts with more than 50 monthly conversions, Smart Bidding (tROAS) outperformed manual CPC on 7 of them. Average performance: +22% ROAS, -8% CPA at constant budget.

The critical condition: data volume. Below 30-40 conversions/month, Smart Bidding oscillates without managing to calibrate. I saw underspend phases of 40-60% of budget on data-poor accounts.

**What I learned:** don't switch to Smart Bidding until the account reaches 30 conversions/month. Use "Maximize Conversions" without a target to accumulate data, then add the tCPA or tROAS target.

**Verdict: PMax > Search+Shopping on some accounts, catastrophic on others**

This was the most surprising result. On 6 accounts where I tested Performance Max against Search + Standard Shopping combined:

- 3 accounts: PMax 30-50% higher ROAS - 2 accounts: similar performance (+/- 10%) - 1 account: catastrophic PMax (-45% ROAS, CPA x2.8)

The catastrophic account had a 15,000-product catalog with high margin disparity. PMax optimized for click volume, not margin. Without strong signals on high-value conversions, it spent heavily on low-margin products.

**What I learned:** PMax requires clear conversion value segments (different values by product/category). Without that, it's a black box that can destroy your profitability.

**Verdict: Manual > AI for niche audiences under 10,000 people**

For campaigns targeting very specific audiences (B2B niche, technical industries, restricted geographies), manual targeting keeps the advantage. Google's audience AI needs volume to function: below a certain threshold, it expands too broadly and degrades lead quality.

Real test: B2B campaign targeting HR directors at SMEs. With Optimized Targeting enabled, lead qualification rate dropped from 68% to 41%. Back to manual targeting: rate returned to 65%.

**What I learned:** disable "Optimized Targeting" on niche B2B campaigns. Leave it active for broad consumer e-commerce campaigns with audiences over 100,000 people.

**Verdict: Manual Mix + AI > Full Automatic for RSAs**

Google recommends letting AI test all headline and description combinations. In theory, this is optimal. In practice, AI can assemble incoherent combinations that hurt brand perception.

Real example: a premium brand positioned on "artisanal craftsmanship" saw Google combine "Unbeatable Price" (promotional headline) with "Handcrafted Since 1985" (brand headline). Excellent CTR, but product return rate +18%: customers attracted by price weren't the target audience.

**What I learned:** pin at least 1 headline in position 1 to guarantee entry message consistency. Leave positions 2 and 3 free for automatic optimization.

After 6 months, here are the tasks where manual management remains essential:

**1. Portfolio strategy**: AI optimizes campaign by campaign. It doesn't perceive the overall strategy (growth vs profitability, planned seasonality, product launches).

**2. Lead quality judgment**: a $15 CPL can be excellent or disastrous depending on lead quality. AI can't qualify without integrated CRM data.

**3. Differentiating creativity**: the best ads on my accounts are always written by a human who deeply knows the brand. AI produces efficient mediocrity, not excellence.

**4. Reacting to events**: reputation crisis, aggressive competitor action, stock shortage. AI reacts with a 1-2 week lag (learning cycle). A human reacts in 1 hour.

**My conclusion:** AI is a performance lever, not a replacement. The best accounts I manage combine Smart Bidding + manual strategy + human creatives. That combination outperforms, not either alone.