Ad Campaign Consolidation

Campaign Consolidation

Turn a fragmented multi-campaign account into a consolidation plan: what to merge, what to keep split, in what order, and how to judge the result. The core trade this skill prices is simple and unavoidable: splitting a fixed conversion volume across N campaigns or ad sets divides each branch's learning signal by N.

Consolidation buys signal density for the delivery algorithm; segmentation buys control, relevance, and reporting. Neither side wins by default: price that trade for this specific account instead of preferring one direction.

You recommend; you never execute. The output is a consolidation plan the account owner applies with their own hands and sign-off, never applied changes, even where your harness could reach the account.

This skill activates once fragmentation is already the suspected or confirmed problem. The broader "why is this account underperforming" root-cause work belongs to mbfinotti/advertising-skills@ad-account-diagnostic, which hands its fragmentation verdict to this skill - run the diagnostic first if nobody has established that fragmentation is the problem.

Consolidation is platform-endorsed, not a fringe opinion:

• Evidence for: the major social platform's own (since retired) five-pillar best-practice program listed account simplification as a pillar; the search-ads world has a named method for collapsing single-keyword structures into few intent-themed campaigns (see Platform Notes).
• Evidence against: at least one published large consolidation case came out slightly worse after merging; one source reports that exiting the learning phase alone is worth roughly 5-10%, not a transformation.

Treat consolidation as a hypothesis to test, never a guaranteed win. Do not oversell it in the plan.

Interview

Ask before analysing anything.

• One question per message.
• Offer multiple-choice answers where possible.
• Skip anything already supplied by the user, the data, or a diagnostic handoff.

• Which platform(s)? (search / paid social / short-video / professional-network / several)
• B2B or B2C/ecommerce?
• Monthly spend, and monthly conversion count on the event the account actually optimises to? (This single pair decides most of what follows.)
• Current structure: how many campaigns, ad sets/ad groups, and active ads?
• What can you export, and at what granularity? (Per-branch per-day conversion counts are the working minimum.)
• What is the target CPA or ROAS - and is it derived from unit economics (margin, deal size, close rate) or inherited from a dashboard?
• Which segments exist as separate campaigns/ad sets today, and _why was each one split out_? (The load-bearing question - answers like "for the 2024 launch" or "the previous manager reported that way" versus "different margin per segment" decide the whole plan.)
• Who controls the budget, and is any split externally mandated (client contract, finance, a guaranteed spend per line of business)?
• What reporting does the business need preserved after the merge? (Contractual or operational - not "nice to have".)
• Are any branches currently in the learning phase, and is there an active cost spike or panic driving this request?
• How long is the sales cycle, when is the next peak season, and by what date must the result land? (A hard date promotes the fast, contained moves - prune, then merge - and rules out anything whose payoff needs a learning cycle plus a sales cycle.)
• Do you want a one-off win on this quarter's numbers, or a structure that keeps paying? (One-off promotes prune and re-parent; a compounding mandate promotes merge and, in low-volume B2B, up-funnel.)
• What is the effort ceiling: rebuild hours available, who must sign off, how much political capital the split owners cost you, and how reversible the change must stay? (A low ceiling caps the plan at prune and re-parent; only a high one funds a multi-phase merge.)

Workflow

1. Run the Interview. Collect every answer before touching data.
2. Inventory the current structure. Map every campaign → ad set/ad group → ad, with each branch's daily budget, bid strategy, optimisation event, audience definition, and the stated reason it exists. If you can read the account's exports directly, build this yourself; otherwise emit the exact export steps and a table for the user to fill and return. Tag each split as _deliberate_ (someone can state a live business reason) or _archaeological_ (launch leftovers, a departed manager's reporting habit, an old test nobody closed) - a split nobody can justify is the cheapest merge available.
3. Compute per-branch conversion volume against the platform's learning threshold. Count conversions per branch over a lag-mature window (exclude days still accumulating attributed conversions), on the event the branch actually optimises for - never a downstream event the platform never sees. Compare against the platform's _documented_ threshold (see Platform Notes; re-verify against the live help page, these numbers drift). Where the platform documents no threshold, say so - do not import another platform's number.
4. Run the budget feasibility test - it is independent of the volume test. A branch can be above threshold in theory yet under-budgeted to ever reach it: minimum daily budget ≈ target CPA × exit threshold ÷ threshold period in days. A branch whose budget cannot buy its threshold is fragmented by funding even if its audience is large.
5. Classify every branch using the four states below. Never let "couldn't verify" collapse into a verdict.
6. Test every existing split against the preserve-vs-collapse rules (section below). A split survives only when a rule protects it; the learning cost of keeping it must be named in the plan either way.
7. Apply the Evidence Gate (section below). If it fails, the deliverable is not a merge plan - it is a statement of exactly what extra days, spend, or access would clear it.
8. Produce the target structure. Pick each group's move from Consolidation Moves below - the earliest rung its classification allows, not the most ambitious one.

• For each merged group: the shared objective, offer, optimisation event, bid strategy, and budget level (campaign-level pooling vs per-branch budgets - see Budget Mathematics for when pooling recreates the problem).
• For each preserved split: the rule that protects it and its budget floor.
• Cover the segmentation question explicitly: which segments move into creative variants inside the merged structure (section below), and which keep structural separation.

1. Produce the migration sequence (Migration Mechanics below): phases, what gets built new vs edited, the control, the no-touch window, and the timing relative to peak season.
2. Define the measurement plan and the re-check date (Measuring section below): baseline, judgment window, pass threshold, rollback trigger, and who signs off. If your harness has persistent memory, memorise the inventory, classifications, plan, phase dates, and re-check date so the follow-up run starts from history; otherwise emit a short state block the user can paste into the next session.

Classification States

Assign every branch (campaign or ad set) exactly one state, each with a one-line justification citing the volume test, the budget test, and the preserve rules:

• `merge` - below threshold (or budget-starved) and no preserve rule applies. Name the merge destination.
• `keep` - clears volume and budget on its own, or a preserve rule protects it. Name which.
• `keep but re-parent` - reporting or control must survive, but the bidding signal should pool. The branch stays separately reported under a shared budget and bid strategy (rung 3 below).
• `insufficient evidence` - the data cannot support any of the above. State what would clear it. This is a real state, not a failure to decide.

Consolidation Moves, Ranked by Efficiency