The Moloco DSP's Internal Logic: From Bidding to Optimization

In this article, we take an in-depth look at the internal logic of Moloco DSP — from auction handling and bid calculation to budget optimization and ML training — so you can effectively use the platform for CPL/CPA offers.

We provide a structured overview of the platform’s key mechanisms. Our goal is not just to list facts, but to break down the inner workings of Moloco DSP and show you how to quickly master this new channel, achieve high ROI, and scale your campaigns.

Introduction

Moloco DSP is built on deep machine learning and continuous optimization, providing a “ready-to-run” tool for launching profitable campaigns. Right after traffic is launched, Moloco filters the most valuable requests, runs them through 8–10 DNN models, calculates a bid for each auction opportunity, and automatically redistributes the budget by hour and day. As a result, you may see CPI drop by up to 25% and conversions increase by up to 30% within 7–14 days, along with transparent analytics down to individual request level and smooth scalability without “shocking” the algorithms.

Below is a schematic overview of how Moloco DSP works.

Understanding the "backstage" of Moloco DSP is not just theoretical — it's a key to profitable campaigns:

• Without understanding inference and internal auctions, your bid works “blindly,” and you risk overpaying for non-target users.
• Moloco updates models in real time without manual tweaks, and you immediately see changes in CPI and ROI from any targeting adjustment.
• The platform provides log-level data and transparent reports, so you can analyze every request down to the subdomain level.

Key Technology Advantages

Instant Filtering and ML Inference.

Moloco discards low-value bid requests at the reserve price level, saving 10–15% of your budget within the first hour. Each remaining request runs through 8–10 models, predicting conversion likelihood, optimal price, fraud, frequency control, and LTV forecasts.

Automated Budget and Strategy Management.

The Weekly Budget Optimizer redistributes the budget across days and hours, so you always “hit” the peak time slots without manual work. Choose tROAS if your offer has known margins (CPI down to €1.5–2), or Max Sales for CPL campaigns focused on volume.

Predictable Scaling.

From day 7 you can already see stable ROI and gradually increase budget by 10–20% every 2–3 days without disrupting algorithm performance. SDK integration can give up to 20% higher fill rate and reduce latency — especially critical for gambling offers.

Data Collection and Processing

Each bid request passes through a real-time pipeline: the service receives the request from SSP/Ad Exchange, extracts context (time of day, device, session), and merges SDK first-party data with postbacks from MMPs (Adjust, Appsflyer). We previously covered Appsflyer tracker integration in a dedicated article.

ML Models and Predictive Analytics

Moloco uses multiple models at once — up to ten inferences per request with different algorithms for first- and second-price auctions. These models predict conversion and win probabilities, then convert these forecasts into actual bid values, comparing them to the bid floor.

To understand how Moloco DSP makes decisions in real time, it's essential to look at which models are involved in show selection and bid formation. Each bid request runs through 8–10 specialized DNN models, each responsible for a part of the logic:

• Reserve Price Filter — filters out requests with bids below the threshold, preventing wasted resources.
• Pacing Controller — manages budget pacing across the day and week.
• Supply Path Optimizer — chooses the best traffic acquisition route (SSP/ADX), minimizing cost per impression.
• Install Likelihood Model — predicts the chance of app installation by a specific user.
• Purchase/Bet Probability Model — estimates the likelihood of a purchase or bet, especially for gambling offers.
• Retention & Churn Predictor — evaluates if a user will be active in 3–7 days or is a “one-time” lead.
• LTV Estimator — predicts total revenue generated by the user over their lifetime.
• Price Optimization Model — determines optimal bid value, including bidding strategies and competition.
• Fraud Detection Network — detects fraudulent patterns, blocking suspicious impressions.
• Budget Allocation Model — reallocates budget across campaigns, focusing on the most effective ones.

For fast ML inference, Moloco deploys separate TensorFlow-powered VMs communicating over gRPC; all logs are uploaded to BigQuery and GCS for analysis and model retraining.

Thanks to this architecture, the system instantly determines whether to bid, how much to bid, where to show the ad, and how to hit your target ROAS.

With this knowledge, arbitrage specialists can build informed launch strategies, choose suitable creatives, confidently use broad audiences, and track high-quality metrics — not just CPI but also LTV, Retention, and Purchase Rate.

Bidding in Moloco DSP

Bid-to-Value

In Moloco, a bid isn’t just a number you manually set — it’s dynamically calculated based on predictions and the value of a conversion. The formula is:

bid = P(conversion) × Value(conversion)

P(conversion) — the likelihood of conversion. This is predicted by Moloco’s ML model for each specific user at the moment of impression request. It takes into account:

Real-time user behavior

• Campaign history
• Placement, device, time of day, region, etc.
• Even creative quality (based on CTR and engagement)

Value(conversion) — how much you're willing to pay for a conversion. Examples:

• CPI offer = $2
• Registration = $5
• Deposit = $20

Example: If the model predicts a 10% (0.1) chance of conversion and the value of the conversion is $10, Moloco will bid $1.00. If the chance is low (e.g., 0.01), the bid drops to $0.10.

This helps maximize the value of each impression without wasting budget on “random” users.

Bid Shading and Dynamic CPC/CPI

Moloco supports bid shading, automatically adjusting bids to the auction type (first-price or second-price) to reduce overspend while staying competitive.

How it works: If Moloco detects a first-price auction (you pay what you bid), it adjusts the bid using historical win prices.

Real-life gain: If the model wants to bid $2 but knows similar auctions are won at $1.20, it may bid $1.25 — still winning, but saving you $0.75.

Creative and Frequency Considerations

CTR, engagement, and creative performance impact the bid multiplier: high-engagement creatives receive better bids. Frequency capping is dynamically adjusted to avoid banner fatigue.

Prepare creatives in advance

Make quick teasers (images), short dramatic videos (15–30s), and interactive demos (HTML5 playable). Ensure they match Moloco specs (formats, sizes, durations).

Placement Strategy

Start with equal budget distribution across banner, video, playable, and native. After 48 hours, narrow to the top 2 formats with best CPI and conversion.

A/B Testing

Change only one element at a time (text, button color, CTA) to isolate what works. Use the Moloco dashboard to monitor creative-level metrics (CTR, win-rate, CPI).

Dynamic Updates

Refresh creatives every 7–10 days to avoid “banner blindness.” Upload seasonal event or promo materials (e.g. “Live bets”, “Exclusive bonus”).

Budget and Bid Optimization

Weekly Budget Optimizer

Moloco analyzes your weekly spend trajectory and redistributes the budget across time slots with peak performance. This lets you spend more during peak periods (weekends, evenings) and less during off-hours, without exceeding your weekly cap.

Average Daily Budget Optimizer

This flexible mode allows ±50% spend deviation per day, staying within 7× the average daily budget weekly. According to Moloco, this boosts efficiency by 10% over fixed daily budgets.

Pacing and Targeting

Moloco is not just a platform where you set budget and targeting and hit “launch.” Most processes are automated — especially budget pacing and targeting.

Pacing distributes the budget throughout the day and week. Moloco controls how many impressions are delivered when, so your budget isn’t burned early. But it’s more than just “spreading out” spend — Moloco’s algorithms consider:

• When high-conversion users are most active (e.g., more registrations at night)
• Seasonality (e.g., cheaper traffic on Monday vs Saturday)
• Where ROI is highest during the day

Advanced targeting (geo, device, OS, language, connection type) works not as filters, but as ML inputs. So saying “Android users in Mexico” signals Moloco to find the most valuable Android users in Mexico. This improves:

• Model learning speed
• Reduces spend on low-value impressions
• Increases ROI

Example: For a gambling offer in Tier-2 targeting only the Philippines, Moloco may bid on low-end devices. But adding filters like Android 11+, WiFi connection, and mid/high device price tier narrows it to a more conversion-prone audience.

Practical Tips for Arbitrage Specialists

• Generate 500–1000 conversions in the first 3–5 days to trigger aggressive optimization.
• Test campaigns separately by geo and event type to let the algorithm segment users by value.
• Use 3–5 creatives per campaign for dynamic model selection.
• Avoid increasing budgets by more than 30% at once — this disrupts model learning.
• Enable Weekly/Average Daily Budget Optimizer at launch for time-based spend optimization.
• Monitor MMP data quality — postback delays hurt model learning and predictions.

Conclusion

A deep understanding of Moloco DSP’s internal processes, the right strategy, creatives, and placements, and an iterative approach to settings allow arbitrage specialists to master this channel and achieve stable ROI growth from the first week. Use this article as a checklist to make your Moloco DSP campaigns as profitable as possible.

And if you don’t want to spend time manually figuring out all the details — at Rentacc, we offer Moloco DSP agency accounts for rent with full support: from launch to optimization. We’ve already studied the platform’s inner mechanics, tested hypotheses, and built working funnels. All that’s left is for you to launch and scale — backed by our experience and support.