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.
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.
Bid
Every request scored and priced in under 20ms.
Serve
Creative chosen per placement, language and device.
Qualify
Filter funnel captures answers step by step.
Verify
Fraud, duplicate, geo and reachability checks.
Deliver
Pushed to your CRM with the full payload.
What each layer does
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
# 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
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.
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.
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.
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
Replace this list with the integrations you actually support.
// 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": "…" } }
{
"lead_id": "mz_9f31c8a2",
"status": "meeting_booked",
"value": null,
"note": "viewing Saturday"
}
This one endpoint is what makes the optimisation closed-loop.
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
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.
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.
Placeholder service levels — replace with your contracted SLA.
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.