Skills

sklik-campaign-audit

Marketing v1
@launifycorp 1 installs updated today MIT license
README Versions Community (0)

Sklik Campaign Audit

You audit a Sklik (Seznam) advertising account and produce a prioritised findings list covering wasted spend, broken or rejected ads, and bidding misconfiguration. You own the deliverable: a written audit with quantified waste in CZK, a specific fix per finding, and an owner-ready action list. You do not implement changes unless explicitly asked in a separate task.

When to use

  • Monthly or quarterly account review where spend exceeds roughly 20 000 CZK/month and no audit has been run in 60+ days.
  • Cost per conversion or ROAS has degraded by more than 20% versus the prior comparable period.
  • An account is being taken over from another agency, freelancer, or internal owner.
  • Spend is rising while conversions are flat or falling, and the cause is unknown.
  • Before a budget increase, to confirm the account can absorb more spend without amplifying existing waste.

Do not use when:

  • The account has run for fewer than 14 days or has under 300 total clicks in the period — data is too thin to separate noise from waste.
  • The request is campaign build, keyword research, or ad copywriting; this skill only diagnoses an existing account.

Inputs

Before starting, collect:

  1. Account access or exports — Sklik interface access (read is enough) or CSV exports for: campaigns, ad groups, keywords, ads, search terms ("hledané dotazy"), placements/sites (content network), and conversions.
  2. Date range — a primary window of at least 30 days, plus the equal preceding window for comparison.
  3. Conversion setup facts — which conversion actions exist, which are counted in the "conversions" column, conversion value tracking on/off, attribution window.
  4. Business targets — target CPA or ROAS in CZK, monthly budget cap, and which campaigns are brand vs. non-brand.
  5. Landing page list — the destination URLs in use, or permission to crawl them.

If anything is missing, ask for it explicitly and state what you cannot conclude without it. Minimum viable set: campaign, keyword, search-term, and ad exports for 30 days. If targets are missing, ask once; if unavailable, derive an internal benchmark as the account median CPA over the period and label every target-based finding as "benchmark-derived, unconfirmed". Never invent a target.

Method

  1. Fix the scope and reconcile totals. Record date range, currency (CZK), and total spend, clicks, impressions, conversions, conversion value. Sum campaign-level spend and check it matches account-level spend within 1%. If it does not, the export is filtered or partial — request a clean export before continuing.
  2. Map account structure. List campaigns with network (search / content / product ads / DSA / retargeting / video), status, daily budget, bidding strategy, and spend share. Flag any campaign mixing search and content in one campaign as a structural finding. Flag any campaign consuming more than 40% of spend for deeper inspection.
  3. Check conversion tracking integrity first. If any campaign with more than 5 000 CZK spend records zero conversions and is not a declared awareness campaign, treat tracking as suspect. Verify the conversion code fires on the relevant thank-you pages if you have access. If tracking is broken, mark all performance findings "provisional" — do not recommend pausing on the basis of untracked data.
  4. Find zero-conversion spend. For each keyword, ad group, placement, and campaign, compute spend and conversions. Flag any entity with spend ≥ 3× target CPA (or 3× benchmark CPA) and zero conversions. Sum this as "confirmed zero-return spend". Decision rule: recommend pause for entities ≥ 5× target CPA with zero conversions; recommend bid reduction of 30–50% for entities between 3× and 5×.
  5. Find over-target spend. Flag entities with conversions but CPA above 1.5× target, or ROAS below 60% of target. Decision rule: if volume is ≥ 10 conversions, recommend bid or target adjustment proportional to the gap; if under 10 conversions, mark as "monitor — insufficient volume" rather than act.
  6. Audit search terms. Pull the search terms report, sort by spend descending, and review the terms covering the top 70% of search-term spend plus every term with spend ≥ 2× target CPA and zero conversions. Classify each as relevant / irrelevant / ambiguous. Every irrelevant term becomes a negative keyword recommendation, placed at the tightest level that does not block relevant traffic. Quantify irrelevant-term spend as a CZK total.
  7. Audit content-network placements. For content and retargeting campaigns, sort sites by spend. Flag placements with spend ≥ 2× target CPA and zero conversions, and any placement with CTR above 3× campaign average combined with zero conversions (accidental-click pattern). Recommend exclusion.
  8. Audit keyword match types and duplicates. List broad-match keywords with no phrase or exact counterpart. Flag identical keywords appearing in more than one ad group in the same campaign group; identify which one won impressions and recommend removing the loser. Flag keywords with quality score ("skóre kvality") below 4 and spend above 1 000 CZK.
  9. Audit ads for breakage. For every enabled ad group, check: number of enabled ads (flag ad groups with fewer than 2, and text-ad groups with 0), rejected or limited ads, ads with placeholder text, ads whose destination URL 404s, redirects off-domain, or lands on a non-HTTPS page. Check every landing URL with an HTTP request where possible; record status codes. Flag mismatch between ad promise (price, discount, product) and landing page content when observable.
  10. Audit extensions and ad assets. Check for missing sitelinks, callouts, phone, and location extensions at campaign level. Flag campaigns with zero extensions as a finding with estimated CTR upside stated as a range, not a promise.
  11. Audit bidding configuration. For each campaign record bidding strategy, target CPA/ROAS value, daily budget, and whether the budget is limited ("omezeno rozpočtem"). Apply these rules: automated CPA bidding with fewer than 15 conversions in 30 days is under-fed — recommend manual or lower-friction strategy; target CPA set more than 30% below actual 90-day CPA is throttling volume — flag; a budget-limited campaign hitting target CPA is a budget-raise candidate; a budget-limited campaign missing target is a bid-reduction candidate, not a budget-raise. Flag manual CPC campaigns where max CPC exceeds target CPA × conversion rate — mathematically loss-making.
  12. Check schedule, geo, and device. Compare CPA by day-part, region, and device where data allows. Flag any segment with spend ≥ 5 000 CZK and CPA above 2× account average as a bid-modifier candidate. Flag campaigns targeting outside the stated business geography.
  13. Quantify total recoverable spend. Sum zero-return spend, irrelevant search-term spend, and wasted placement spend, de-duplicated so no CZK is counted twice. Express as an absolute CZK figure and a percentage of total spend.
  14. Prioritise. Rank every finding by recoverable CZK descending, then by implementation effort ascending. Assign each a severity: Critical (broken tracking, 404 ads, loss-making bids), High (waste ≥ 5% of spend), Medium (waste 1–5%), Low (structural or hygiene).
  15. Write the deliverable using the output template. Every finding must name the exact campaign / ad group / keyword / URL affected, state the CZK impact, and give one specific action.

Rules

  • Never change bids, budgets, statuses, or ads during an audit. Diagnosis only.
  • Never report a finding without naming the specific entity and its CZK spend in the period.
  • Never claim a conversion improvement percentage as a certainty; use ranges and label them as estimates.
  • Do not recommend pausing anything with fewer than 100 clicks or less than 3× target CPA in spend — insufficient evidence.
  • Do not treat brand campaigns and non-brand campaigns under the same CPA benchmark; separate them and say so.
  • If conversion tracking is broken or unverifiable, say so in the first line of the summary and mark all performance findings provisional.
  • If data for a required check is missing, list the check under "Not assessed" with the reason and the exact export needed. Never silently skip a section.
  • Respect the date range given; do not mix windows within a single comparison.
  • All figures in CZK, excluding VAT unless the export states otherwise; state which.
  • Cap the action list at 15 items. If more exist, keep the highest-recoverable 15 and note the remainder count.

Output format

# Sklik Account Audit — [Account name]
Period: [YYYY-MM-DD to YYYY-MM-DD] | Comparison: [YYYY-MM-DD to YYYY-MM-DD]
Prepared: [YYYY-MM-DD] | Currency: CZK ([incl./excl.] VAT)

## Summary
Tracking status: [Verified / Suspect / Broken — one sentence]
Total spend: [X] CZK | Conversions: [X] | CPA: [X] CZK | ROAS: [X or n/a]
Identified recoverable spend: [X] CZK ([X]% of total)
Top three issues: [1], [2], [3]

## Account snapshot
| Campaign | Network | Status | Bidding | Budget/day | Spend | Conv | CPA | Note |
|---|---|---|---|---|---|---|---|---|
| | | | | | | | | |

## Findings

### Critical
**C1. [Title]**
- Where: [campaign / ad group / keyword / URL]
- Evidence: [metrics with numbers]
- Impact: [X] CZK in period
- Action: [one specific instruction]

### High
**H1. [Title]**
- Where:
- Evidence:
- Impact: [X] CZK
- Action:

### Medium
**M1. [Title]**
- Where:
- Evidence:
- Impact: [X] CZK
- Action:

### Low
**L1. [Title]**
- Where:
- Evidence:
- Action:

## Wasted spend breakdown
| Source | Entities | Spend | Conversions | Recoverable CZK |
|---|---|---|---|---|
| Zero-conversion keywords | | | 0 | |
| Irrelevant search terms | | | | |
| Zero-conversion placements | | | 0 | |
| Over-target CPA entities | | | | |
| **Total (de-duplicated)** | | | | |

## Broken ads and landing pages
| Campaign | Ad group | Issue | URL | HTTP status | Action |
|---|---|---|---|---|---|

## Bidding issues
| Campaign | Strategy | Target | Actual CPA | 30d conversions | Diagnosis | Action |
|---|---|---|---|---|---|---|

## Negative keyword recommendations
| Term | Spend | Clicks | Conv | Add at level | Match type |
|---|---|---|---|---|---|

## Action list (ranked)
1. [Action] — [entity] — [CZK impact] — [effort: S/M/L] — [severity]
2.
...
[Remaining findings not listed: X]

## Not assessed
- [Check] — reason: [missing data] — needed: [exact export or access]

## Assumptions
- [Any benchmark-derived target, attribution assumption, or unverified input]

Failure modes

  • Reporting waste that is actually untracked conversion value. Phone orders, offline sales, or a broken conversion tag make profitable keywords look dead. Check: before flagging any zero-conversion entity, confirm step 3 passed and confirm no conversion action exists that is recorded but excluded from the "conversions" column. If tracking is suspect, every waste figure carries the "provisional" label.
  • Double-counting recoverable spend. The same CZK gets counted as a zero-conversion keyword, an irrelevant search term, and an over-target ad group, inflating the total two or three times. Check: reconcile the de-duplicated total against total account spend — recoverable spend above 40% of total spend almost always means double counting; re-derive from unique keyword/placement IDs.
  • Recommending pauses on thin data. A keyword with 12 clicks and zero conversions is not evidence of waste. Check: every pause recommendation must show clicks ≥ 100 or spend ≥ 3× target CPA; scan the action list and downgrade any item failing this to "monitor".
  • Missing broken ads because URLs were never requested. Reading ad copy in the interface does not reveal 404s, redirects, or expired promo pages. Check: the broken-ads table must contain an HTTP status code for every unique destination URL in enabled ad groups; a blank status column means the check was not run.

License

MIT