New: closed-loop CRM optimisation is live for all advertisers. See how it works
Sign up Log in Get a proposal
Technology & AI

Optimised to your sales floor, not to a pixel.

Four systems, one loop: a bidding engine that prices every impression, a scoring model that grades every lead, a verification layer that rejects the bad ones, and a delivery pipeline that puts the good ones in your CRM in under a second — then feeds your outcomes back into the bid.

Architecture

The closed loop

Most performance stacks end at the conversion event. Ours only starts being useful there — because a form completion tells you nothing about whether the phone got answered.

01

Bid

Every request scored and priced in under 20ms.

02

Serve

Creative chosen per placement, language and device.

03

Qualify

Filter funnel captures answers step by step.

04

Verify

Fraud, duplicate, geo and reachability checks.

05

Deliver

Pushed to your CRM with the full payload.

Step 06 is you. Contact outcomes, meeting bookings and closed deals flow back over the same API and become the training signal for tomorrow's bids. Partners who send outcome data typically see cost per closed deal fall materially faster than those who do not.
Systems

What each layer does

Real time

Bidding engine

A bid request arrives with a few dozen fields: publisher, placement, size, device, geo, time, user signals. The model turns that into a probability that this impression leads to a lead your sales team will qualify — then bids the value of that probability against your target cost per lead.

Features in the model

  • Placement and sub-publisher history
  • Creative-to-placement fit
  • Hour, weekday and seasonal effects
  • Device, connection and OS context
  • Geo granularity down to emirate or city
  • Fraud risk and frequency penalties
bidder · pseudocode
# evaluated per bid request, <20ms budget

p_lead   = model_lead.predict(request)
p_verify = model_verify.predict(request)
p_sql    = model_sql.predict(request)

value    = target_cpl × p_lead
           × p_verify × p_sql

bid      = clamp(
             value × pacing_factor,
             floor_price,
             max_bid
           )

if fraud_score > threshold:
    return NO_BID
Every lead

Lead quality scoring

Each delivered lead carries a score from 0 to 100. It blends the qualification answers, the behaviour inside the funnel, the source that produced it and — once your outcome data is flowing — how similar leads have historically performed in your pipeline.

Sales teams use it to sequence the callback list. Our bidder uses it to decide what to buy tomorrow.

85+
Call first — strong intent and fit
60–84
Standard queue
<60
Nurture, or filtered out entirely
Score thresholds are configurable per account. Some partners buy only 80+; others take everything and let the score drive callback priority.
Gatekeeper

Fraud & verification

Nothing leaves the platform without passing a fixed set of checks. The point is not only to protect your budget — invalid leads distort the optimisation model and make every subsequent decision worse.

Technical checks

  • Phone format & carrier validity
  • Email syntax & disposable domains
  • IP, proxy and data-centre detection
  • Device fingerprint consistency

Behavioural checks

  • Impossible completion speed
  • Repeated answer sequences from one source
  • Duplicate across the agreed window
  • Answer bands outside campaign spec

Replacement policy

If something slips through, you flag it and we credit it. The definition of a valid lead is written into the agreement before launch, so there is nothing to argue about later.

Standard return window7 days

Credit processed within1 billing cycle

Typical return rate< 5%

Data handling

Consent captured and stored with every record. Data encrypted in transit and at rest, retention windows configurable per client, and deletion requests honoured across the pipeline. Bring us your DPA and we will work to it.

Under a second

Delivery & API

Real-time HTTPS POST is the default. We will also push into your automation platform, use a native CRM connector, or drop scheduled files if that genuinely suits your process better.

Supported destinations

Custom HTTPS endpoint Webhook Salesforce HubSpot Zoho Bitrix24 Property CRMs Zapier / Make SFTP / CSV Google Sheets

Replace this list with the integrations you actually support.

POST https://api.mediazoo.example/v2/leads
// request headers
Authorization: Bearer <api_key>
Content-Type: application/json

// payload delivered to you
{
  "lead_id": "mz_9f31c8a2",
  "received_at": "2026-03-14T09:12:44Z",
  "campaign": "offplan_marina_q3",
  "creative": "b4_payment_plan_ar",
  "source": "exchange_direct_pub_412",
  "language": "ar",
  "contact": {
    "first_name": "…",
    "phone": "+9715…",
    "email": "…"
  },
  "answers": {
    "budget": "1.5m_3m_aed",
    "purpose": "investment",
    "timeline": "0_3_months",
    "payment": "payment_plan"
  },
  "quality_score": 88,
  "consent": { "given": true, "ts": "…" }
}
POST /v2/outcomes · you → us
{
  "lead_id": "mz_9f31c8a2",
  "status": "meeting_booked",
  "value": null,
  "note": "viewing Saturday"
}

This one endpoint is what makes the optimisation closed-loop.

Always on

Reporting

A live dashboard, not a monthly PDF. Volume, cost, funnel step conversion, creative performance, quality distribution and — where outcome data flows back — cost per meeting and cost per closed deal, attributable to the individual creative that produced it.

Standard views

  • Daily delivery against committed caps
  • Funnel drop-off, step by step
  • Creative league table
  • Quality score distribution

Access

  • Web dashboard, role-based logins
  • Scheduled email summaries
  • Reporting API for your own BI
  • Weekly written read-out from your account lead
Model operations

How the algorithm stays honest

An optimisation model that is never challenged will happily over-fit to a lucky fortnight. These are the guardrails that keep it useful in month six.

Adjust the cadences here to match how your platform actually runs.

Nightly refit

Models retrain against the last rolling window of outcomes, weighted toward recent data.

Held-out validation

A slice of traffic is always excluded from training, so we can see whether the model is actually beating the baseline.

Exploration budget

A fixed share of spend always goes to unproven placements. Without it, the model slowly starves itself of new information.

Human override

Buyers can veto the model. On a launch week, or when a client's sales floor is short-staffed, judgement beats history.

Drift alerts

If predicted lead quality and actual quality diverge beyond tolerance, the campaign flags for review before the budget notices.

0
Delivery pipeline uptime target
<0
Bid response budget
<0
Lead in your CRM after completion
0
Pipeline monitoring & alerting

Placeholder service levels — replace with your contracted SLA.

Technical review

Bring your CTO to the second meeting

We are happy to go through the payload schema, the security model and the integration plan in detail before anyone signs anything.