Skip to content

AI for Media & Advertising

The attribution broke. The measurement discipline is what is left.

We engineer AI for performance teams, media planners, and agencies: creative production that keeps up with the refresh cycle, audience and bidding that still work after ATT and consent mode, media-mix modelling that reads next to platform attribution, and brand-safety guardrails that hold on open exchanges.

Discuss your media and measurement workflowSee how the system works

The operational reality

You spend across Meta, Google, TikTok, and CTV. The cycle is faster than the report.

iOS cuts platform attribution. Consent cuts the rest. The CFO still wants one number.

Measurement is the discipline. The spend proves it works.

Felt by heads of performance, media planners, brand marketing leads, and the agency teams that absorb what weak measurement and slow creative leave behind.

  1. 01

    Brief and plan

    A quarterly plan sets channel split, audience strategy, and the creative concepts that will carry each campaign across Meta, Google, TikTok, and CTV.

    Friction · The plan is built against last quarter's measurement model; the model is already a quarter behind the platforms.

  2. 02

    Creative production

    Agencies and in-house studios produce variants: static, video, UGC, aspect-ratio cuts. Variants ship to the channel-specific creative platforms.

    Friction · Weekly refresh cycles collide with monthly production timelines; the best-performing variant is a week late and burnt out on arrival.

  3. 03

    Audience, bidding, and measurement

    Audiences, bids, and budgets are set in Meta Ads Manager, Google Ads, TikTok Ads, DV360, and The Trade Desk. Measurement runs through platform attribution, aggregated attribution (SKAN, GA4 modelled conversions), and increasingly a media-mix model.

    Friction · Platform numbers overstate; MMM lags by weeks. The daily operator is squeezed between two systems that disagree.

  4. 04

    Pacing and brand safety

    Spend is paced against targets; brand-safety filters run on open exchanges and video. Negative keywords and placements are curated weekly.

    Friction · Open-exchange placements drift into MFA sites and low-quality video; nobody has time to clean the inventory list.

  5. 05

    Reporting and iteration

    Weekly readouts combine platform numbers, aggregated attribution, and MMM commentary. Decisions on creative, audience, and channel split flow into next week.

    Friction · The readout is a slide deck rebuilt from four dashboards. The decisions it should drive arrive after the week has moved on.

Swipe the workflow

Where the work breaks

The friction is in the places nobody budgets for.

  • 01

    Creative velocity cannot keep up with the refresh cycle.

    How we fix it · A system we build catches this before it costs you.

    Impact · The best-performing variant is a week late; CPAs climb while the team waits on production rather than on measurement.

  • 02

    Attribution reads one number on-platform and another in MMM.

    How we fix it · A system we build catches this before it costs you.

    Impact · Daily operators cannot act with confidence; strategy meetings relitigate the measurement model instead of the plan.

  • 03

    Open-exchange inventory drifts into MFA and low-quality video.

    How we fix it · A system we build catches this before it costs you.

    Impact · Spend lands where no human saw it; brand exposure quietly degrades even as reported impressions grow.

  • 04

    Consent mode cuts the signal the old audience tools relied on.

    How we fix it · A system we build catches this before it costs you.

    Impact · Lookalike seeds narrow, event streams miss conversions, and the system that used to tune itself needs a human in the loop again.

  • 05

    Weekly readouts are a slide deck rebuilt from four dashboards.

    How we fix it · A system we build catches this before it costs you.

    Impact · The decisions the data should drive arrive after the week it was supposed to inform.

What we would build

Systems, not features. Each with its own job.

  • 01

    Creative production that matches the refresh cycle

    Performance refresh cycles are weekly; production cycles are not.

    Who benefits · Heads of performance, creative directors, and the agency and in-house teams who ship variants to Meta, Google, and TikTok.

    Inputs
    The brand brief, the campaign brief, past winners and their performance, source assets, and channel-spec requirements.
    What the system does
    The system drafts variant copy and generates on-brand variants across required aspect ratios. Each variant is tagged with the hypothesis it tests, so the measurement later knows what it meant.
    Output
    A continuously refreshed variant library, each with the brief, the hypothesis, and the channel-spec compliance already done.
    Benefit
    The best variant is live inside the refresh cycle, not after it. Creative velocity stops being the ceiling.
    Human stays here
    A creative director approves every variant before it ships. The system refuses variants that score below the brand brief, even when a test shows a lift.
  • 02

    Audience and bidding under consent mode

    Platform auto-bidding degrades without clean event streams; consent mode and ATT cut the signals it was tuned on.

    Who benefits · Heads of performance, paid-media leads, and ad ops.

    Inputs
    Server-side event streams through Conversions API and Google's Measurement Protocol, your CRM and lifecycle data, and the campaign brief.
    What the system does
    The system builds and refreshes consented, server-side event streams and first-party audience seeds. It tunes bidding against your own target metric rather than the platform's default.
    Output
    A clean event stream running to each platform, audience segments refreshed on real engagement, and bidding tuned to the number the business actually cares about.
    Benefit
    The platform tools work again, on signals you can audit. Your CPA target is the one being optimised toward, not a proxy for it.
    Human stays here
    Consent is honoured at the event level. The system refuses to stream events the user did not consent to, and audits run against your consent logs.
  • 03

    Media-mix modelling next to platform attribution

    Platform-attributed revenue overstates; MMM is slow and arrives late. Teams relitigate measurement every week.

    Who benefits · Heads of performance, finance partners, and the planning team.

    Inputs
    Platform spend and reported conversions, server-side event streams, promo and pricing calendars, macro data where it matters.
    What the system does
    A rolling MMM reads next to platform attribution and reconciles the two with a documented bridge. It is refreshed on a weekly cycle, not quarterly.
    Output
    A single weekly readout with platform numbers, MMM contribution, and the bridge; next week's recommended channel split comes with it.
    Benefit
    Decisions get made against one number the team agrees on, inside the week they are supposed to inform.
    Human stays here
    Planners sign off on the recommended split. The system shows its intervals and the assumptions behind each contribution.
  • 04

    Brand safety and inventory curation on open exchanges

    Spend drifts into MFA sites and low-quality video; the curation list is older than the campaign.

    Who benefits · Ad ops, programmatic buyers, and brand marketing.

    Inputs
    Delivery logs from DV360, The Trade Desk, YouTube and CTV; your brand-safety and suitability criteria; MFA and known-low-quality lists.
    What the system does
    The system monitors delivery, flags placements drifting outside policy, and drafts the updated allow and deny lists. Decisions are proposed; buyers decide.
    Output
    A weekly curation review with the handful of placements most worth acting on and the specific list changes to make.
    Benefit
    Spend lands where humans see it; brand exposure stays consistent with the brief rather than the exchange.
    Human stays here
    List changes are reviewed and approved by ad ops. The system does not block placements silently; everything is logged.
  • 05

    Weekly readouts that drive the next week

    Reporting is a slide deck rebuilt from four dashboards; the decisions it should drive arrive after the week moves on.

    Who benefits · Heads of performance, agency leads, and brand clients.

    Inputs
    Platform APIs, MMM output, event streams, creative performance, and the campaign brief.
    What the system does
    A weekly brief is generated from the same data sources, with the three decisions worth making this week and the one metric to watch.
    Output
    A readable Monday brief, with platform numbers, MMM contribution, creative winners and losers, and the recommended next steps.
    Benefit
    The team spends the week acting, not rebuilding the deck. Clients get one source of truth.
    Human stays here
    Every brief is reviewed and sent by the account lead. Nothing is auto-published to a client.

Show the system

What it could look like in your hands.

A design concept, not a shipped product. Interfaces like this get drawn out before any code lands, so the system reads back the way the people using it work.

performance / campaign dashboard
As of Mon 09:04
Spend today
$42,180

+3.4% vs plan

Impressions, 7d
18.4M

+6.1% WoW

Blended CPA
$38.20

-$2.80 WoW

Conversions, 7d
9,842

+2.9% WoW

Spend and conversions by channel, last 7 days

Bar is spend. Dot is conversions. Scaled so a channel that converts above its share lands above the bar.

  • Spend
  • Conversions
059,000118,000$ / 7dMeta3,900 convGoogle3,100 convTikTok1,250 convYouTube820 convCTV320 convDV360452 conv
Synthetic data. This is a design concept of how a weekly performance view across Meta, Google, TikTok, and CTV could present spend, conversions, and creative-variant performance in one place.

End-to-end workflow

From the trigger to the outcome.

A representative weekly cycle, from the Monday brief to the next variant live. Automated stages run on the data as it lands; human stages are explicit and never skipped.

  1. 01

    Signals land

    Platform APIs, server-side event streams, creative performance, and brand-safety delivery logs are pulled into one data boundary each night.

    Integration
  2. 02

    Monday brief generated

    A weekly brief lands with platform numbers, MMM contribution, creative winners and losers, and the three decisions worth making this week.

    Automated
  3. 03

    Account lead reviews

    The brief is reviewed, edited, and sent. Nothing publishes to a client or to a platform without the account lead.

    Human in loop
  4. 04

    Creative variants drafted

    New variants are drafted against the week's hypotheses in required aspect ratios, each tagged so the measurement knows what it meant.

    Automated
  5. 05

    Brand and brief check

    Variants are scored against the brand brief; any below threshold are routed back for a rewrite rather than published.

    Guardrail
  6. 06

    Creative director approves

    Approved variants ship to the channels. Rejected ones carry the reason into the next drafting cycle.

    Human in loop
  7. 07

    Bidding and audiences refresh

    Server-side event streams refresh audience seeds and the bidding target; curation lists update on brand-safety signals.

    Integration

How exceptions are handled

  • If consent is revoked on an event, the system refuses to stream it to the platforms and the audience refresh skips that record.
  • If a creative variant scores below the brand brief, it is routed back even when the test shows a lift.
  • If platform-attributed numbers and MMM diverge beyond a documented band, the brief flags the divergence rather than picking one side.
  • If delivery lands on an MFA or disallowed placement, the next refresh of the deny list surfaces it for a buyer to approve.

How it is engineered

Built on your stack, not ours.

Design choices that make this run against the real platforms, not a demo. We build to the APIs the ad stack already exposes and to the measurement discipline teams actually need now.

01

Platform integration

Meta Marketing API, Google Ads API, TikTok Marketing API, DV360 and The Trade Desk, plus server-side event streams through Conversions API and Measurement Protocol. We write back only through the APIs each platform supports.

02

Measurement layer

Platform-attributed numbers, aggregated attribution, and a rolling media-mix model, with a documented bridge between them. The MMM refreshes on a weekly cycle, not quarterly.

03

Creative pipeline

A versioned brand brief, structured prompts, and channel-spec compliance checks. Every variant carries the brief, the hypothesis, and the channel-spec validation metadata.

04

Consent and data boundary

Consent is honoured at the event level and audited against your consent logs. First-party data stays in your tenant; model inference runs with no retention and no training on your data.

05

Brand-safety curation

Delivery logs are monitored against your suitability criteria and known-low-quality and MFA lists. Allow and deny lists are versioned and reviewed; nothing is blocked silently.

06

Observability and evaluation

Every brief, variant, bid change, and curation list change is logged with model version, inputs, and the human approver. An evaluation harness watches brief quality and MMM drift.

What the business will measure

Numbers the operator already watches.

We agree on metrics and baselines before building. These are the numbers a performance or planning lead usually watches when deciding whether a system is actually working.

  • 01

    Blended CPA against MMM, not platform

    Baseline today

    Measured today from platform APIs and your conversions endpoint, with a reconciled MMM contribution per channel.

    How it moves

    Bidding is tuned to the number the business cares about, not the platform's default; MMM gives one agreed denominator.

  • 02

    ROAS by channel, bridged

    Baseline today

    Platform-reported ROAS and MMM-attributed ROAS, with the bridge between them documented each week.

    How it moves

    The weekly brief publishes both and the gap; decisions are made against the bridge rather than against either number alone.

  • 03

    Creative velocity (new variants per week)

    Baseline today

    Creative production throughput by week, baselined from the current intake and ship cadence.

    How it moves

    Drafting and channel-spec compliance are automated inside the brand brief; the creative director's time shifts to approval, not production bottlenecks.

  • 04

    Incremental lift measured against a holdout

    Baseline today

    Where feasible, holdout tests on paid or lifecycle campaigns, with the incremental lift reported alongside platform numbers.

    How it moves

    Holdout design is wired into the brief; results read against the holdout rather than against platform-attributed revenue.

  • 05

    Share of impressions on allowed inventory

    Baseline today

    Delivery share on allowed vs MFA / disallowed placements, weighted by spend.

    How it moves

    Curation lists refresh weekly on delivery signals; buyers approve the changes rather than the system blocking silently.

Human in the loop

Oversight, consent, and the measurement bridge

Where the humans stay in control.

The system is designed around what performance teams, their clients, and the privacy regime already expect. Nothing is autonomous that should not be.

  • 01

    Consent at the event level

    Server-side event streams honour consent per event. Revocations propagate on the same cycle; audits run against your consent logs.

  • 02

    Measurement bridge documented

    Platform-attributed numbers and MMM contribution are reconciled with a documented bridge each week; the gap is reported, not hidden.

  • 03

    Human approval on every send

    Creative variants, bid changes, and client briefs are reviewed and approved by the right human before anything ships.

  • 04

    Brand brief as ground truth

    Every drafted variant is scored against the brand brief. Variants below threshold are routed back rather than published.

  • 05

    Brand-safety transparency

    Allow and deny lists are versioned and auditable. The system does not block placements silently; every change carries a reason and an approver.

  • 06

    No training on your data

    Your event streams, creative performance, and client data are not used to train base models. Any fine-tuning happens in your tenant, on your data only.

Proof

Work we have shipped, and the shape of what we build.

Real engagements where we can speak to them, and representative use cases where we cannot.

How the build runs

Scoped tightly, shipped in phases.

Fixed price per phase, approved before the phase starts. You own the code, the models, the data pipelines, and the deployment.

  1. 01

    Discovery

    Two weeks. We sit in a weekly performance readout and a creative review, map the real workflow against your platforms and measurement, and agree on the metrics we will move.

  2. 02

    Prototype on live campaigns

    Four to six weeks. One brand or client, one measurement bridge, one creative pipeline. The team uses it alongside the current week; we measure against the baseline.

  3. 03

    Production rollout

    Eight to twelve weeks per workflow. Platform APIs, consented event streams, MMM, creative pipeline, brand-safety curation, SSO, audit trails, and the evaluation harness.

  4. 04

    Handover

    You get the full source, the prompts, the MMM models, the runbooks, and a two-week walkthrough with your performance, creative, and engineering teams.

  5. 05

    Optional ongoing support

    A fixed-term support agreement for platform API changes, measurement regime shifts, and MMM recalibration. Never a retainer dependency.

Common questions

Answers before the first call.

01Do you work directly with the platform APIs or through a vendor?
Directly. Meta Marketing API, Google Ads API, TikTok Marketing API, DV360, The Trade Desk, Conversions API, and Measurement Protocol. We are not a partner of any of them; we build you a system that is compatible with the stack you run.
02How does this handle iOS tracking and cookie deprecation?
The measurement layer assumes the signal is lower resolution. Server-side event streams run through Conversions API and Measurement Protocol, aggregated attribution is read next to a rolling MMM, and the bridge between platform numbers and MMM is published each week rather than argued about.
03What is your position on media-mix modelling?
MMM is back, but on a weekly cycle, not a quarterly one. Platform attribution gives directional signal; MMM gives contribution. Both read in the same brief with a documented bridge, so the team decides against one number instead of two.
04Will the creative variants still look like the brand?
The brand brief is the ground truth. Every drafted variant is scored against it; variants below threshold are routed back rather than published. A creative director approves every variant before it ships.
05How do you handle brand safety on open exchanges?
Delivery is monitored against your suitability criteria and known-low-quality and MFA lists. The system drafts allow and deny list changes each week; your buyers approve. Nothing is blocked silently, and every change is logged.
See the other details on the FAQ page

Next step

A thirty-minute conversation. Bring one brand or client where the platform numbers and the MMM disagree today, and we will walk through what the first phase would look like, where the measurements come from, and what has to be true for your account lead and finance partner to approve it.

Discuss your media and measurement workflow

Helpful to have on the call

Helpful to have on the call: the platforms you buy on, one campaign or client with a measurement or creative-velocity pain point, and the person who owns the number you would like to move.

Explore

A different problem in every sector, one engineering standard underneath.

Featured · AI for healthcare

Healthcare

Why is your clinician still charting at midnight?

Your clinicians are charting until midnight and your front desk is drowning in prior auths. We build AI that writes clinical notes automatically, triages patients to the right care faster, and finds any record with a simple question, so your people spend their hours on patients instead of paperwork.

See AI for healthcare
stallwart / healthcare

Live run

Healthcare

Running
as of nowhealthy
See every industry we build for

An AI-First Engineering Company

Get a price in one call.

One 30-minute call. You leave with an architecture, a fixed price, and a start date. No charge, and no slide deck.

Booking now for December 2026

Book a callSee how it works