Capability 03 · Tuned while it runs
Most campaigns are judged after they end. These are corrected while they run.
AI Campaign Optimization builds the measurement and the models that decide where budget goes: clean signal, incrementality tests, marketing mix modelling and daily reallocation. Proposals arrive with their reasoning; a person approves anything that moves money.
Day 7: retargeting sits at its spend floor, retargeting is cut by the maximum again, and search category asked for 26 percent more, above the limit, so it moved 20 percent and the rest waits for a person to approve.
Illustrative figures in any currency. The dashed tick on each bar is the spend floor.
What we build · 6 parts
Measure it honestly, then let the model act
No model survives a broken measurement layer. Signal quality is fixed first, then experiments calibrate the models, then automation acts within limits you set.
-
Measurement foundations
Server-side tagging, consent mode, conversion APIs with event deduplication and offline conversion import, so platforms and your warehouse agree on what happened.
Clean, consented signal -
Incrementality & geo experiments
Holdout and geo-split tests that measure what advertising actually caused, used to calibrate every model that follows.
Lift · holdouts -
Marketing mix modelling
Bayesian mix models with adstock and saturation curves, calibrated against experiments, producing budget scenarios by channel and market.
MMM · scenarios -
Budget allocation & bidding
Daily or weekly reallocation proposals across channels and campaigns, each with expected effect and confidence, executed through platform APIs after approval.
Proposed · approved · applied -
Creative & audience testing
Structured test plans with sequential testing or multi-armed bandits, fatigue detection and a library of what has already been learned.
Tests · bandits -
Anomaly detection & guardrails
Pacing, cost and conversion anomalies detected within hours, with spend caps, rollback and a kill switch that does not need an agency ticket.
Alerts · caps · rollback
Where it sits in the engine
Spend moves on evidence, not on last week’s dashboard.
The loop one budget decision takes. The highlighted step is where the optimisation model does its work.
- 01 · Signal Conversions in Server-side conversions and offline sales, deduplicated and consented.
- 02 · Model Forecast Expected return per ad set and audience at the next unit of spend.
- 03 · Guard Guardrails Daily shift limits, spend floors, brand safety and frequency caps. Campaign AI
- 04 · Approve Above threshold Moves bigger than the agreed limit wait for a person.
- 05 · Prove Incrementality Geo and holdout tests confirm what the platforms report.
How it runs · 4 stages
Baseline, instrument, model, then automate
Each phase is only worth running because the one before it is trustworthy. Automation comes last, not first. Timings are typical, not promised.
Stage 01 of 4Wk 01–02
Baseline
Current spend, reporting, attribution and data quality reviewed. Gaps between platform numbers and your own records quantified, and the first experiments designed.
Produces
- Signal audit
- Spend baseline
- Experiment plan
Stage 02 of 4Wk 02–05
Instrument
Server-side tagging, consent mode, conversion APIs and warehouse ingestion built, with event deduplication and a single conversion definition.
Produces
- Tracking build
- Warehouse pipeline
- Metric definitions
Stage 03 of 4Wk 05–09
Model
Mix models fitted and calibrated against experiment results, with scenario planning and a written read on what each channel contributes.
Produces
- MMM results
- Experiment readouts
- Budget scenarios
Stage 04 of 4Ongoing
Optimise
Reallocation proposals and creative tests run on a weekly cadence, guardrails monitored, and models refitted as the data grows.
Produces
- Weekly proposals
- Test log
- Monthly performance review
Holdout read-out · what you will see
Test regions against control regions. Then decide.
The read-out a budget decision rests on: matched regions with and without the change, the difference and how sure we are of it.
| Measure | Test regions (8) | Control regions (8) | Difference | 90% interval |
|---|---|---|---|---|
| Orders per 10k people | 41.2 | 37.6 | +9.6% (on track) | +4.1% to +14.8% |
| Cost per incremental order | ₹ 612 | —not applicable | Below ₹ 700 target (on track) | ₹ 520 to ₹ 890 |
| New customers share | 38% | 36% | +2 pts (needs attention) | −1 to +5 pts |
| Returns within 30 days | 6.1% | 6.0% | No change (on track) | −0.8 to +1.0 pts |
What you own
Models, tests and read-outs. In your warehouse.
The measurement model, the test design and the read-outs live in your own data stack, with the code to rerun them.
| # | Deliverable | Format |
|---|---|---|
| 01 | Measurement plan & metric definitions | Doc · sheet |
| 02 | Server-side tagging & conversion APIs | Container · code |
| 03 | Experiment designs & readouts | Plan · report |
| 04 | Marketing mix model & scenario tool | Model · dashboard |
| 05 | Budget allocation proposals | Dashboard · log |
| 06 | Creative & audience test library | Sheet · board |
| 07 | Anomaly alerts & spend guardrails | Alerts · config |
What changes
Spend that proves itself. Three numbers finance will sign.
Optimisation succeeds when lift is measured against a control, cost per outcome falls and budget moves on evidence. They are aims, not guarantees.
-
01
Numbers that agree
One conversion definition across platforms, analytics and the warehouse, so meetings are about decisions rather than whose report is right.
-
02
Budget that follows evidence
Allocation informed by experiments and modelled contribution instead of platform-reported credit.
-
03
Mistakes caught in hours
Pacing and cost anomalies surface the same day, with caps and rollback ready, not at the month-end review.
Cost per acquisition by week, optimised vs fixed split. Lower is better. Shape only: in-flight reallocation against a budget left on its launch split, with the gap confirmed by a holdout test rather than by platform reporting.
Technologies we work with
Warehouse first. Then the models on top.
The analytics, warehouse, modelling and reporting tools we use to measure and move spend. Named as technologies we work with, not partnerships; model providers run inside an approval step, never around it.
- Google Analytics
- Google Tag Manager
- Google BigQuery
- Snowflake
- Airbyte
- Python
- scikit-learn
- Looker
- PostHog
- Mixpanel
- dbt
- Power BI
Frameworks we build to
Measured without tracking people further than they agreed.
Frameworks each model and tag plan is designed against. Frameworks we build to, not certifications we hold.
-
GDPR — General Data Protection Regulation (EU) 2016/679
General Data Protection Regulation (EU) 2016/679
Consent captured per purpose and channel, tracking tags fired only after consent (Consent Mode v2 where Google is used), and erasure requests reaching every connected tool.
-
DPDP Act 2023 — Digital Personal Data Protection Act, 2023
Digital Personal Data Protection Act, 2023
Notices and consent records for India-based customers, with withdrawal honoured in every journey and channel within the same working day.
-
NIST AI RMF 1.0 — AI Risk Management Framework
AI Risk Management Framework
Each model that scores, targets or writes is inventoried with an owner, a measured error rate and a threshold that stops it.
-
EU AI Act — Artificial Intelligence Act (EU) 2024/1689
Artificial Intelligence Act (EU) 2024/1689
AI-generated content labelled where required, targeting and scoring use cases classified by risk, and a person able to override every automated decision.
-
ISO/IEC 27701 — Privacy information management systems
Privacy information management systems
Every marketing tool that holds personal data listed with its purpose, retention and processor terms, and kept current as tools are added.
Services & packages
Know what worked, while there is still time to act on it.
We build the measurement, the experiments and the models that decide where your budget goes, then automate the reallocation within limits you set. Clean signal first, causal evidence second, automation last.
- Categories
- 05
- Services
- 15
- Packages
- 04
How to buy
- 01Pick services. Enquire about one, or add several to a brief.
- 02Choose how to engage. A sprint, a fixed project or an ongoing team.
- 03Send the brief. We reply within one working day.
Start a project
Ways to engage. Same team, same standard.
Three ways in, from a two-minute question to a formal RFQ. Each is read in full by the lead for the work, and anything already in your brief goes with it.
Or book a thirty-minute call-
01
For a first conversation, a press request, or anything that does not need a scope yet.
You get A reply from a lead, not a sales queue
-
02Most useful
Goals, audiences, a budget band and timing. Enough for us to come back with a shape, not only questions.
You get Options and a first scope after one call
-
03
Your pack, your deadlines, and the procurement and security rules the work must meet.
You get Receipt confirmed and a named bid lead
How it is priced
Each package shows how it is priced. Every engagement starts with a written scope and a quote agreed before work begins.
Opens a project brief with this package chosen.
-
In your brief
- Typical length
- 1–3 weeks
- Pricing
- Fixed fee
-
In your brief
- Typical length
- 4–12 weeks
- Pricing
- Fixed price
-
In your brief
- Typical length
- Ongoing · 6-month minimum
- Pricing
- Monthly fee
-
In your brief
- Typical length
- 6–18 months
- Pricing
- Programme fee · by statement of work
Compare what each package includes
| Package | Every engagement includes | Best for |
|---|---|---|
| SprintOne fixed question, answered in one to three weeks. |
|
Discovery, a diagnostic, a prototype or a decision you need to make soon |
| ProjectA defined scope, delivered for a fixed price. |
|
Work you can describe up front: an identity, a system, a set of tools |
| RetainerReserved monthly capacity to run, improve and extend what we built. |
|
Brands and products after launch that need a steady team without hiring one |
| EnterpriseA multi-workstream programme with governance, a dedicated team and SLAs. |
|
Large organisations running change across markets, portfolios or business units |
Questions
What teams ask about campaign ai.
Anything else, ask us directly. A short call is usually quicker than a long email.
01Is this the same as Performance Marketing?
No. Performance Marketing, in Campaign & Content Design, plans and runs the media. This builds the measurement, models and automation that decide where the money should go. It works with our media team or with the agency you already use.
02Attribution, incrementality or mix modelling — which do we need?
All three answer different questions. Attribution shows paths and is useful for operations. Experiments measure true causal lift on a specific channel. Mix modelling covers the whole budget, including channels you cannot track. We use experiments to calibrate the model and treat attribution as a diagnostic.
03Does signal loss from privacy changes break this?
It changes the method, not the outcome. Consent mode, server-side tagging and conversion APIs preserve what people have agreed to share, and mix modelling and geo experiments do not depend on individual tracking at all.
04Will an AI system be allowed to change our budgets?
Only within limits you set. Proposals carry the expected effect, the confidence and the reasoning; a named person approves anything above the threshold you choose; caps, exclusions and rollback are always on. Full autonomy is available for narrow, well-tested decisions once evidence supports it.
05How much data does mix modelling need?
Typically two to three years of weekly history across channels and markets, plus spend, price and seasonality. With less, we start with experiments and lightweight models and build towards it.
Works best with
Better spend needs better creative and better leads.
Creative AI gives the model more to test; Lead Gen turns the traffic into pipeline.
Let’s build what happens next.
Tell us what you’re building. We’ll answer straight.
Three ways to start
Every engagement starts with a written scope and a quote agreed before work begins.
Choose one of the three ways above