Grounds the brief in the account’s own data then proves the campaign can be measured at all.
Builds the whole campaign, never half of one atomic, paused, and reversible by construction.
Keeps the live account improving and measures whether each change actually helped.
Brief to live, then better planned, measured, gated, built and improved in one window.
The model researches, drafts and diagnoses. It never commits spend. The router sequences, the gates decide, and one tool carries each mutation.
Paid and organic, split clean one workspace, one governance model.
Over half of accounts run with broken or incomplete conversion tracking.
Six campaign types, one pipeline
Search, Performance Max, Shopping, Demand Gen, App and Video on Google Ads, and Meta Ads on the same brief and the same gates. The same pipeline, whichever surface the objective calls for.
Grounded, not generic
Every decision reads the account’s own data: what has worked before by theme and result, the auction position and the angles rivals own, real volume and cost rather than templates.
The loop that earns autonomy
Each applied change is measured after it lands and the verdict feeds back, so clean work raises the level and harmful work lowers it. The account ceiling is never exceeded, whatever the brief asks for.
Enterprise controls from day one, every workspace isolated, every role scoped, and every action written to an audit trail an operator can read.
Autonomy is earned against outcomes. Approved and declined changes both feed back, and the gates that protect a live account stay in place however high the level climbs.
See the trust modelAn account is connected, a brief goes in, and the first plan comes back with the structure, the rationale on every node, the forecast and the ranked risk flags, plus a measurement verdict on whether the account can support the objective at all.
Request a campaign audit →The live account passes to the optimisation half of the same agent, which observes, diagnoses, proposes, and applies only what has been approved, then measures whether each change helped.
See the loop →What does NIA actually do
It plans a campaign from a brief, grounds it in the account’s data, proves the campaign can be measured, builds and launches it behind a human gate, then runs a continuous optimisation loop over the live account.
Can it spend money on its own
Not at the default level. It builds paused and a person launches. At higher levels it acts within guardrails that the brief cannot override, and every level sits under the account’s own ceiling.
What happens if measurement is broken
A conversion objective does not proceed. The gate returns a remediation spec naming each gap, the fix, the effort and what it costs the campaign to leave it unfixed, so a launch never stalls without a path forward.
Which campaign types are covered
Search, Performance Max, Shopping, Demand Gen, App and Video on Google Ads, and Meta Ads alongside them, all through the same pipeline, the same gates and the same governance.
How does it decide a change is worth making
Every proposed change is tied to the finding that motivates it and must clear a significance threshold for its metric and sample. A failing change is dropped from the set rather than weakened, and nothing is applied to an entity still inside its learning period.
Is anything irreversible
The build is atomic, so a failure creates nothing, and every optimisation change records the prior value of every entity it touches, which is the revert path. A partial build rolls the account back to a clean state rather than leaving a half built campaign able to serve.