What people ask before they talk to us.
About NoptiK.
What is NoptiK in one sentence?
NoptiK is a Bayesian predictive analytics platform for digital ad agencies that forecasts campaign outcomes — revenue, conversions, ROAS — with calibrated uncertainty, before campaigns launch and as they run.
How is this different from marketing mix modeling?
MMM is retrospective and slow — typically rebuilt quarterly, designed to attribute past spend to past outcomes. NoptiK is forward-looking and updates daily. Both use Bayesian methods; we differ in three places: the prediction loop is built for forecasting (not attribution), there's a conformal calibration layer that gives distribution-free coverage guarantees, and there's a cross-client hive-mind that learns across an agency's portfolio.
How is this different from last-click or multi-touch attribution?
Last-click attribution is biased and broken in a post-iOS-14 world — it gives the credit to the last touchpoint regardless of causal contribution. Multi-touch attribution makes strong assumptions about how to allocate credit across touchpoints and inherits the underlying tracking gaps. NoptiK doesn't replace attribution — it forecasts outcomes directly, with calibrated intervals you can plan against.
What platforms does NoptiK integrate with?
Meta (Facebook + Instagram) Ads, Google Ads, TikTok Ads, and LinkedIn Ads at launch. We add platforms based on design partner demand. If your portfolio has a platform we don't yet support, tell us and we'll prioritize it.
What kind of agencies is this for?
Digital ad agencies running paid social and paid search campaigns for multiple clients — especially DTC, ecommerce, B2B SaaS, and lead-gen verticals. The hive-mind benefits compound with portfolio breadth: agencies running 10+ clients in adjacent verticals get the most lift from cross-client learning.
How big does an agency need to be?
No minimum. The model is designed for agencies of any size. Smaller agencies benefit more from hive-mind priors (less of their own data); larger agencies benefit more from per-client posterior tightening (more of their own data). Both are useful.
What is the math behind the forecasts?
Bayesian hierarchical models for partial pooling across clients, conformal prediction for distribution-free coverage guarantees, BOCPD for regime-shift detection, Postgres row-level security and differentially-private aggregation for tenant isolation. Full methodology page available.
What does "calibrated uncertainty" actually mean for me?
When NoptiK says there's a 90% chance your campaign's revenue will fall between $X and $Y, that has to be true 90% of the time across many campaigns. We measure this on rolling windows, per-vertical, and publish the calibration plot in-product. If we miss the stated rate, that's a bug we fix — not marketing copy we adjust.
Are you live yet?
Not yet. We're in active development with our first design partners. If you run a digital ad agency and want early access, apply to the design partner program.
What does the design partner program include?
Free during private beta. Hands-on configuration with both founders. Direct influence over the product roadmap — design partners shape what we build first. Early access to new modes and features as they ship.
What does it cost?
Design partners are free during private beta. Pricing for general availability will be a per-tenant SaaS subscription scaled to portfolio size. We will publish pricing publicly when we launch.
What if my agency runs verticals you haven't seen before?
The model handles cold-start verticals via weakly-informative priors and hierarchical pooling from related verticals. Forecast intervals will be wider for new verticals and tighten as your data accumulates. The model is honest about uncertainty rather than confidently wrong on sparse evidence.
What happens if a platform changes its algorithm mid-campaign?
The Bayesian online change-point detector (BOCPD) flags the regime shift. The model widens forecast intervals and re-anchors the prior rather than continuing to forecast off stale data. This is a deliberate design choice: a model that admits the world has changed is more useful than one that pretends it hasn't.
Where your data lives, who owns it, what we don't claim.
Where does my data live?
Every agency tenant gets its own isolated Postgres schema. Row-level security is enforced at the database layer — there is no cross-tenant query path. Data does not leave your isolated schema except through the differentially-private aggregation layer that powers the hive-mind, and that layer mathematically prevents reconstruction of individual records.
What about SOC 2 or ISO 27001?
We are pre-launch and have not completed SOC 2 or ISO 27001 audits. Both are on the roadmap. We will not claim them until we have earned them. If your agency requires either before signing, we'll be transparent about our timeline and provide our internal security posture in detail under NDA.
Who owns the data?
You do. We process your platform feeds to produce forecasts. We do not resell data, we do not share data across tenants in raw form, we do not train external models on your raw data. Your platform credentials are stored encrypted at rest and rotated on standard schedules.
What is the deletion policy?
Standard B2B SaaS terms — request deletion via team@noptik.ai and your tenant schema is dropped within 30 days. Aggregated, differentially-private posterior parameters from your data may persist in the shared model, but no individual record is recoverable from those parameters.
How does the LLM use creative artifacts?
Vision and language models extract structured features (color, emotion, persuasion cues, CTA framing) from creatives. Those features feed the Bayesian model. The LLM never produces the numerical forecast, and the LLM does not retain or learn from your creatives across tenants. Feature extraction is logged with a complete audit trail.