LinkedIn Ads Optimization Log (sample)

A representative weekly log from a B2B SaaS account this system runs on. Numbers, dates, and campaign names are illustrative and sanitized from a live account. What is real is the shape: the layer framework, the per-format metric discipline, the insight drill-downs, and the scars. This is what the agent appends every Monday. I approve or strike each recommendation; it executes the approved ones through the LinkedIn Marketing API and writes the result back here.

---

The framework every read runs on

Categorize before you count.

• Layer from the campaign-group name: a Create layer (cold prospecting) and a Capture layer (retargeting and BOFU). Agency or partner spend, if present, stays in its own bucket and never folds into an owned layer.
• Format from the ad-set name: TLA (Thought Leader Ad), Single Image, Document Ad, Video, Connected TV, Conversation Ad, Spotlight.
• The valid metric per format, chosen before judging: TLA and Single Image on true CPC (spend / landing-page clicks); Video on completion rate (2% benchmark); CTV on dwell and completion (never clicks); Conversation Ads on open rate (40%) and CTOR (3%); Document Ads on read-through and true CPC.
• True CPC is computed over clickable formats only. A blended CPC that folds CTV and Video into the click math is the one number that cannot answer a real question.

---

4-Week Comprehensive Analysis

Weeks analyzed: 4 consecutive weeks. Total spend: ~$12,600. Overall true CPC (clickable): ~$73. Leads: 1.

Executive summary

The single most expensive event of the month was a strategy call, not a bad ad: pausing the best-performing format and letting the account run six days on the decision before the change was made.

---

Week 1: Jan 05 - Jan 11

Analysis date: Jan 12. Data: ad-sets export (daily), deduped by (date, campaign, ad set).

Campaign layer categorization

• CREATE (cold): TLA-Ads-Cold, Doc-Ads-Cold, Video-Ads-Cold, SingleImage-Cold(BrandAwareness)
• CAPTURE (RT/BOFU): SingleImage-RT, TLA-Ads-RT, Doc-Ads-RT

Performance summary

Create Layer

By ad set:

Capture Layer

Key insights

INSIGHT 1: Document Ads carry the Create layer. 34 LP clicks at $10.29 CPC, 69% of all Create LP clicks. The multi-page format earns education before the click, so the click that comes is cheaper and more intentional. Everything else in Create is riding on this one format.

INSIGHT 2: Create frequency is 1.6 against a 6-10 awareness benchmark. Budget spread across four campaigns buys reach, not recall. Consolidate onto the two performers or accept this layer as prospecting reach only.

Recommendations

• [ ] Hold Doc-Ads-Cold at $350+; it is the efficiency engine.
• [ ] Pause SingleImage-Cold (0 LP clicks) and shift budget to Doc + TLA.
• [ ] Pull Video-Ads-Cold completion-rate data before judging it (LP clicks are the wrong metric).

Week 1 key metrics

---

Week 3: Jan 19 - Jan 25 (the scar)

Analysis date: Jan 26.

Performance summary

Create Layer

By ad set:

Key insights

INSIGHT 5: Pausing Document Ads spiked the layer CPC 179% in one week.

High-level: Create true CPC went from $48.18 (W2) to $134.58 (W3) the moment Doc Ads were paused.

What drove it: - Doc-Ads-Cold was 59% of LP clicks at ~$11. Removing it forced the layer onto TLA ($50), Video ($175), and Single Image ($117). - Spend fell 21%, results fell 72%. A negative trade.

This is the read-write gap made concrete. The decision to pause was made off-system; the log flagged the CPC collapse on day zero of the next week, but the reversal did not happen for weeks. A correct read is worthless until it is enacted.

INSIGHT 6: First Lead Gen attempt lands at ~$600 CPL on a gift-card incentive. Unsustainable against a $200 target, and a gift-card hook tends to collect incentive-seekers, not buyers. Flagged for a real-offer test.

Recommendations

• [ ] CRITICAL: reactivate Doc-Ads-Cold. Expected recovery: layer CPC back to ~$45.
• [ ] Do not scale the gift-card Lead Gen; test a real offer through the same objective.

Week 3 key metrics

---

Week 8: Feb 23 - Mar 01

Analysis date: Mar 02.

Performance summary

Capture Layer

By ad set:

Key insights

INSIGHT 11: Screenshot social proof beats polished testimonials ~2:1 on CPC in retargeting. The two screenshot creatives run $25-31 CPC; the talking-head and tombstone variants run $50-120. Website Visits objective confirmed more efficient than Brand Awareness for RT.

INSIGHT 12: A high-dwell Engagement campaign has spent $1,050 over three weeks for 0 LP clicks. Strong attention (7.6s dwell), wrong objective, no traffic. Same failure shape as the Doc-Ads scar on the Create side: the objective, not the creative.

INSIGHT 13: 8 weeks, 1 lead, ~$16k spent. Every campaign is on Website Visits or Engagement. No Lead Generation objective exists. The structural fix is one campaign duplication with a Lead Gen objective. It keeps getting written down and not built.

Recommendations

• [ ] Concentrate RT budget on Screenshot-SocialProof-A and -B; pause the tombstone variants.
• [ ] Pause or convert SocialProof-RT (Engagement, 0 LP, $350/wk) to Website Visits.
• [ ] CRITICAL: build one Lead Gen objective campaign. Duplicate the best screenshot creative, change objective, native 3-field form.

---

Week 12: Mar 23 - Mar 29 (the Website-Visits unlock)

Analysis date: Mar 30.

Performance summary

Create Layer

By ad set:

Key insights

INSIGHT 21: A new Website-Visits campaign delivered 24 LP clicks at $41.67, 62% of Create LP clicks. TLA-format creatives inside it beat static images ~12:1 by LP volume. Match the objective to the outcome and let the format that earns attention do the traffic work.

INSIGHT 22: The Spotlight ad holds the best CPC in the account ($6.58) and is chronically underfunded. $13 spent, 2 LP clicks. At $200/week it projects ~30 LP clicks, more than any single ad set. The highest-leverage move available, week after week, is to scale the thing already working.

Recommendations

• [ ] Scale Spotlight-Cold to $200/week. Single highest-leverage action.
• [ ] Scale SingleImage-Cold(Visits); pause the static creatives inside it that show 0 LP.

---

Week 17: Apr 27 - May 03 (leads, finally)

Analysis date: May 04.

Performance summary

Capture Layer

Key insights

INSIGHT 28: First real leads in the log. A Lead Generation objective Conversation Ad went live and collected 2 leads at ~$118 CPL, after 18 tracked weeks at zero. The 18-week structural gap is closed. It validates the mechanism, not the offer, so the next step is running the proven screenshot creative through the same Lead Gen objective and comparing lead quality, not just count.

INSIGHT 29: Capture scaled 3.2x and held ~$35 CPC as budget concentrated on the screenshot social-proof creative. The cleanest scale-up in the account. Penetration finally moved past 30%.

Actions taken

• [x] Built the Lead Gen objective campaign (18 weeks after first flagged).
• [x] Concentrated RT spend on Screenshot-SocialProof-A/-B.
• [ ] Test the screenshot creative through the Lead Gen objective (real-offer vs gift-card).

---

Cross-week trends

Create Layer true CPC

What the log keeps teaching

1. Objective beats creative. The most expensive mistakes were formats stuck on the wrong objective (Engagement when the goal was traffic), not bad ads.
2. The cheapest thing is usually underfunded. The best true CPC in the account sat at near-zero budget for weeks while worse formats absorbed the spend.
3. Match the metric to the format or you will kill your best performer. Judged on LP clicks, the Video and CTV ads look like waste; judged on completion, one of them is the strongest creative in the account.
4. The gap that costs money is between the decision and the click. Every scar here is a correct read that sat unenacted. Closing it, from days to seconds, is why the agent writes to the account instead of only reading it.
5. A recommendation with no owner rots. The log exists so the enacting can be measured, not just the analysis. The week a recommendation goes unexecuted, the log says so.

---

This is a sanitized sample. In a live account, every line is a real object, a real ID, and a real before/after value, and every executed change is written back through the API with its reason attached.