How cannibalization happens
Two campaigns in the same account can be eligible for the same search. Google does not let both enter the auction; it picks one, generally the one with the higher Ad Rank rather than the one you would have chosen. Demand you thought was divided by structure is actually being sorted by the system, and it lands in places you did not plan.
It arises most often from broad match spreading beyond its intended set, from an automated campaign type with no keyword list running alongside a search campaign, from separate campaigns per location that all target overlapping areas, or from a brand campaign and a generic campaign both matching a query that contains the brand name.
Why cannibalization matters
The first cost is diagnostic. When one query is served by two campaigns, neither report tells the truth about that query, and comparisons between campaigns become meaningless. You cannot fairly test a landing page or a bid strategy when traffic keeps moving between the things being compared.
The second cost is learning. Automated bidding needs a steady flow of conversions in one place before it can find a pattern. Splitting the same demand across several campaigns gives each of them a trickle, and thin campaigns stay in learning longer and behave more erratically. In a small market like Nepal, where the underlying volume is limited to begin with, this is the more damaging of the two.
Where it goes wrong
The commonest mistake is building structure for reporting convenience — a campaign per service, per city, per product line — without adding the exclusions that keep them apart. Structure alone does not separate demand; negative keywords do.
The other is over-correcting. Merging everything into one campaign removes the overlap but also removes control over budget and geography, which matters when regions perform very differently. And this is not the same thing as keyword cannibalisation in SEO, where two of your own pages compete in the organic results. The mechanism and the fix are unrelated.
What to do about it
Pull the search terms report at account level and look for queries appearing under more than one campaign. That is the evidence, and it is more reliable than reading the structure and assuming. Then decide which campaign should own each block of demand and enforce it: brand terms as negatives in non-brand campaigns, city names as negatives in the campaigns that should not serve them, product terms confined to their own campaign.
Where the split cannot be enforced cleanly — typically with an automated campaign type that exposes no keywords — consider consolidating instead. Fewer campaigns with more data each usually beat a tidy structure that starves every part of it. Revisit campaign structure whenever the budget changes materially, because a structure that made sense at one spend level rarely survives a much larger or smaller one.