Learning Center - Master Analytics Marketplace & MCP Platform

Overview

Why Sell Predictions Instead of Models?

Most data scientists and ML teams face the same dilemma: you've invested months training a high-value model, but the only monetization options — publishing to model hubs, licensing weights, or consulting — either give away your IP or don't scale. Spartera's External API Connector introduces a third path: sell the predictions, not the model. Your model stays on your infrastructure. Buyers get real-time predictions through the Spartera Marketplace. You earn per prediction with zero IP exposure.

Key Points

Model weights and training data never leave your servers

Buyers discover and purchase predictions through the Spartera Marketplace

You set your own price per prediction and keep 100% of it

Works with any framework: TensorFlow, PyTorch, scikit-learn, XGBoost, custom

Works on any cloud: GCP, AWS, Azure, or your own servers

Spartera handles marketplace discovery, authentication, billing, and delivery

Why It Matters

Why Monetize Predictions Instead of Sharing Models

The economics and security of selling inferences

Total IP Protection

Publishing a model to HuggingFace or licensing weights means your IP is out of your hands forever. Competitors can fine-tune, reverse-engineer, or redistribute. With prediction monetization, your model architecture, weights, and training data never leave your infrastructure — buyers only see inputs and outputs.

Recurring Revenue vs. One-Time Sales

Licensing a model is a one-time transaction. Selling predictions creates usage-based recurring revenue that grows as buyers scale. A model licensed for $50K generates $50K. That same model serving 10,000 predictions per month at $2 each generates $240K per year — and grows with demand.

Serve Unlimited Customers from One Deployment

Without Spartera, monetizing a model means managing API keys, billing, contracts, and support for every customer. With the External API Connector, you maintain a single deployment. Spartera handles customer acquisition, authentication, credit billing, and delivery. You serve 10 or 10,000 buyers with zero additional operations.

Continuous Improvement Without Disruption

When you sell predictions via API, you can retrain and update your model at any time without disrupting buyers. They call the same Spartera endpoint and get better results. Try doing that when someone downloaded your model six months ago — they're stuck on the old version forever.

How It Works

End-to-End: From Model to Marketplace

Four phases to your first earning prediction

Train Your Model (Hello World)

Start with a trained model in any ML framework. For this guide, we'll walk through a simple scikit-learn classifier, but the same approach works for TensorFlow, PyTorch, XGBoost, or any framework that can produce predictions programmatically.

Key Points:

Train a model using your framework of choice (scikit-learn, TensorFlow, PyTorch, etc.)

Save/serialize the trained model (pickle, joblib, SavedModel, ONNX)

Validate predictions locally before deploying

Example: A salary prediction model trained on age, education level, years of experience, and gender

Hello World tip: Start with scikit-learn + joblib — simplest path from notebook to production

Deploy as a REST API Endpoint

Wrap your model in a web service that accepts JSON input and returns predictions. You have two main paths: GCP Cloud Run for managed, auto-scaling deployment, or a generic Flask/FastAPI server on any cloud or VPS.

Key Points:

Option A — GCP Cloud Run (recommended): Containerize with Docker, deploy with 'gcloud run deploy', get auto-scaling HTTPS endpoint in minutes

Option B — Generic Flask/FastAPI: Run on any server (AWS EC2, Azure VM, DigitalOcean, on-prem), manage SSL and scaling yourself

Your endpoint must accept POST requests with JSON body

Spartera sends this POST structure: {"asset_id": "...", "testmode": true/false, "function_id": "...", "params": {your input fields}}

Your endpoint must return: {"answer_value": your prediction result}

Secure with an API key passed via HTTP header (recommended), URL parameter, or request body

Create a Connection in Spartera

Register your API endpoint as a data connection in the Spartera Seller Portal. This tells Spartera where your model lives and how to authenticate requests — without ever accessing your model directly.

Key Points:

Select 'External API' as the Storage Engine (engine_id 20)

Enter your API endpoint URL (must be HTTPS)

Select HTTP method: POST (recommended for model inference)

Configure API Key authentication: choose location (HEADER recommended), parameter name (e.g., X-API-Key), and key value

Test the connection — Spartera will ping your endpoint to confirm it's reachable and responding correctly

Your API key is stored securely in GCP Secret Manager — Spartera never exposes it

Create an Asset and Publish to Marketplace

An asset is the marketplace listing that buyers discover and purchase. It defines the model's input parameters (schema), pricing, and metadata. Once published, buyers can get predictions through the marketplace form or via API — with zero knowledge of your backend.

Key Points:

Navigate to Assets and create a new asset linked to your External API connection

Define input schema: the parameters buyers must provide (e.g., age, education_level, years_experience, gender)

Set asset type to CALCULATION (Spartera routes this correctly for External API inference)

Set pricing using Spartera's credit system — you keep 100% of the price you set

Optionally assign a function_id if your single endpoint serves multiple models (e.g., 'predict_salary', 'predict_churn')

Use the Test button to validate end-to-end: Spartera → your API → prediction returned

Publish to Marketplace — buyers can now discover and purchase predictions immediately

Key Benefits

Why Build AI Assets on Spartera

100%

IP Protection

Model weights, architecture, and training data never leave your infrastructure

< 1 Day

Time to Revenue

Go from trained model to earning predictions in hours, not months

Zero

Operations Overhead

Spartera handles discovery, authentication, billing, and customer support

Unlimited

Customer Scale

Serve unlimited marketplace buyers from a single model deployment

FAQs

Common Questions

What ML frameworks are supported?

Any framework that can produce predictions. scikit-learn, TensorFlow, PyTorch, XGBoost, LightGBM, Keras, ONNX, custom code — if it can run behind a REST API and return JSON, it works with Spartera. The External API Connector is framework-agnostic.

Where can I host my model?

Anywhere with an HTTPS endpoint. GCP Cloud Run, AWS Lambda, Azure Functions, AWS EC2, Google Compute Engine, DigitalOcean, Heroku, Render, your own data center — even a Raspberry Pi if it has a public HTTPS URL. Spartera is cloud-agnostic.

Can buyers see my model's code, weights, or training data?

No. Buyers only see the marketplace listing (input parameters, description, pricing) and the prediction output. Your model weights, architecture, training data, and infrastructure are completely invisible. Spartera only sends structured requests and receives the answer_value response.

How does billing work? What does Spartera charge?

You set a credit price per prediction. Buyers purchase credits and spend them on predictions. You keep 100% of the price you set. Spartera earns by selling credits at a markup to buyers. There are no listing fees, monthly charges, or revenue sharing from your set price.

Can I serve multiple models from one API endpoint?

Yes, using function_id routing. Deploy one endpoint, create separate Spartera assets with different function_id values (e.g., 'predict_salary', 'predict_churn'). Your endpoint reads the function_id and dispatches to the correct model. Each asset gets its own marketplace listing and pricing.

What happens if my model endpoint goes down?

Spartera returns an error to the buyer and classifies it as a seller-side issue (5xx). Buyers are not charged credits for failed predictions. We strongly recommend implementing health checks, uptime monitoring, and setting Cloud Run min-instances to 1+ to avoid cold start downtime.

Can I update my model without disrupting buyers?

Yes — this is a key advantage of prediction monetization. Retrain your model, redeploy your endpoint, and buyers automatically get predictions from the updated model. No versioning headaches, no buyer migration, no re-downloads. The Spartera asset and marketplace listing remain unchanged.

Do I need to write any Spartera-specific code?

Minimal. Your endpoint just needs to: (1) parse the Spartera POST body (asset_id, testmode, function_id, params), (2) run your model on the params, and (3) return {"answer_value": result}. That's roughly 10-20 lines on top of your existing model serving code.

What's the difference between an External API asset and a database-backed asset?

Database-backed assets run SQL queries against a connected data warehouse (BigQuery, Snowflake, etc.) and return aggregate analytics. External API assets forward requests to your custom API endpoint and return whatever your model computes. Both appear the same to marketplace buyers — the difference is entirely in the backend processing.

How fast does my API need to respond?

Aim for under 5 seconds. Most buyer requests come through the marketplace web form, where 1-3 second response times feel responsive. For API-to-API integrations, sub-second is ideal. Spartera will timeout and return an error to the buyer after approximately 30 seconds.

Still have questions?

Creating Your First AI Asset

Why Sell Predictions Instead of Models?

Key Points

Why Monetize Predictions Instead of Sharing Models

Total IP Protection

Recurring Revenue vs. One-Time Sales

Serve Unlimited Customers from One Deployment

Continuous Improvement Without Disruption

End-to-End: From Model to Marketplace

Train Your Model (Hello World)

Key Points:

Deploy as a REST API Endpoint

Key Points:

Create a Connection in Spartera

Key Points:

Create an Asset and Publish to Marketplace

Key Points:

Why Build AI Assets on Spartera

IP Protection

Time to Revenue

Operations Overhead

Customer Scale

Common Questions

What ML frameworks are supported?

Where can I host my model?

Can buyers see my model's code, weights, or training data?

How does billing work? What does Spartera charge?

Can I serve multiple models from one API endpoint?

What happens if my model endpoint goes down?

Can I update my model without disrupting buyers?

Do I need to write any Spartera-specific code?

What's the difference between an External API asset and a database-backed asset?

How fast does my API need to respond?

Stay Ahead of the Analytics Revolution

Related Topics

What Are Intelligent APIs?

The Value of Selling Analytics vs. Raw Data

API Design Best Practices for AaaS

Pricing Strategies for Analytics Sellers

Zero Data Movement Architecture

Partnership Models for Data Monetization

Ready to Monetize Your Model?

We use cookies to enhance your experience