⚡ How We Reduce Costs with Efficient Prompt Engineering

At Blockli Assistant, we’re committed to delivering powerful AI insights without driving up your token usage. One of the key ways we achieve this is through efficient prompt engineering—a practice that ensures every AI interaction is fast, focused, and cost-effective.

Every time the assistant answers a question, it consumes two types of tokens:

• Prompt tokens: The data we send to the AI model

• Completion tokens: The response generated by the AI

If we were to dump raw datasets (like your entire user or activity table) into the AI, it would skyrocket token usage—and your costs. Instead, we’ve built a smarter system that prepares and compresses context before it’s sent to the AI, without compromising accuracy.

Whenever possible, we preprocess data using lightweight scripts before sending it to the LLM (large language model). These scripts compute high-level summaries, group data by trends, and generate metrics that are AI-ready.

For example, if you ask:

“What’s our current user engagement trend over the past 30 days?”

Instead of passing thousands of raw records to the LLM, Blockli Assistant:

✅ Uses a dedicated analytics script to calculate engagement trends

✅ Summarizes key stats like daily activity, growth rates, and drop-offs

✅ Sends only essential, structured insight to the AI for reasoning

The result? You get intelligent, context-rich answers with lower token consumption.

By combining data-aware preprocessing with precision-crafted prompts, we’re able to:

• Reduce unnecessary token usage

• Speed up AI response times

• Increase accuracy by eliminating noise

• Keep your credit consumption optimized

This strategy is built into all major insights—from engagement analysis to version adoption and multi-device trends.

• Data-rich: Informed by precomputed values instead of raw datasets

• Context-aware: Tailored to your unique site, users, and tenant-specific metrics

• Token-efficient: Sends only what the AI needs to reason clearly

Our infrastructure includes:

• Custom analytics scripts (e.g., for engagement, devices, registrations)

• Redis-based precomputed values stored per tenant

• Smart routing logic that determines when data needs to be recomputed

This ensures we only ask the AI to think—not calculate what we already know.

Efficient AI isn’t just about power—it’s about smart design. At Blockli, we’ve engineered every layer of the assistant to work with speed, accuracy, and cost in mind.