Bear Lumen
Start free

AI economics, explained

Stop guessing your AI margins.

Know which customers cost you money. See what could change.

Bring AI costs and customer revenue together. Trace the expensive features and models, then replay pricing or cost scenarios across your customer base.

Start freeRead the docs

See Bear Lumen in action · 80 seconds

Know what drove the bill. See what changes it.

Follow an illustrative $120,000 AI bill to the customers and feature behind it, then compare the impact of two possible decisions.

Find the loss

Twelve of 70 customers have negative AI margins. Cedarline is −$3,000.

Trace the cause

Research Assistant accounts for $72,000 of the bill; nine of the 12 loss accounts use it heavily.

Compare outcomes

A hypothetical 35% model-cost reduction saves $24,500/month. Targeted repricing adds $9,800/month at fixed demand.

Illustrative scenarios replay recorded usage. Model quality and latency require separate testing; the price case holds demand fixed.

Start free

See what the numbers could look like for your product.

Read the video description

A fictional AI company receives a $120,000 monthly AI bill. A compressed setup shows AI usage tracking, revenue connection, and read-only agent access. The user asks which customers are losing money. Lumen identifies 12 negative-margin customers among 70, with Cedarline lowest at −$3,000. The user asks whether a recently launched Research Assistant feature changed costs. Lumen shows the feature accounts for $72,000 of the bill, including $56,246 from Reasoning Pro, and nine of the 12 loss accounts use the feature heavily. The agent suggests testing a lower-cost model, repricing heavy usage, or packaging access differently. A cost scenario replays recorded model usage at hypothetical rates 35% lower. It reduces modeled AI cost to $95,500, lifts portfolio AI contribution from $80,000 to $104,500, changes Cedarline’s AI margin to +$500, and reduces the underwater count from 12 to five. A separate 20% price increase for nine heavy-use accounts lifts portfolio AI contribution to $89,800 at fixed demand, but leaves 12 accounts underwater and Cedarline at −$1,000. The closing compares the two decisions. The phone edition presents the same 80-second story in a vertical layout.

Customers, models, rates, and figures are illustrative. Cost and revenue scenarios separately replay historical usage. The price scenario holds demand, customer count and AI usage constant; it is not a demand forecast. The current simulator estimates hypothetical rates for recorded models, not a direct model replacement. AI contribution excludes other operating costs. Model quality, token use and latency require separate evaluation. On-screen text carries the story. Audio is instrumental only.

Six providers, auto-detected from the response

OpenAIAnthropicGoogle GeminiAWS BedrockMistralOllama

Cost Intelligence

See what drives AI spend

Attribute each tracked AI response to a customer, feature, and model. Find the workflows behind an outsized bill.

Explore cost intelligence

Margin Intelligence

Find underwater customers

Pair attributed AI cost with revenue and rank the accounts whose margins need attention.

Explore margin intelligence

Pricing Intelligence

Compare options before launch

Replay recorded usage at new product prices or hypothetical model rates. See the effect across the company and its customers.

Explore pricing intelligence

Built for modern AI applications

Token spreadsheets and end-of-month invoices are guesswork. Modern AI products are multi-step workflows, and Bear Lumen meters every request and background task as it happens.

  • Auto-detects OpenAI, Anthropic, Google Gemini, AWS Bedrock, Mistral, and Ollama responses
  • Zero-latency overhead: events queue in memory and flush in background batches
  • Streaming responses pass through unchanged
  • Cost, margin, and insight queries from the same SDK
import OpenAI from 'openai';
import { BearLumen } from '@bearlumen/node-sdk';

const openai = new OpenAI();
const bear = new BearLumen({ apiKey: process.env.BEARLUMEN_API_KEY });

const response = await openai.chat.completions.create({
  model: 'gpt-4o',
  messages: [{ role: 'user', content: 'Hello!' }],
});

// One line. Provider auto-detected, cost attributed.
const result = bear.track(response);

result.model;        // 'gpt-4o'
result.inputTokens;  // 12
result.outputTokens; // 85

Enterprise-grade security by design

Our SDK tracks token counts and metadata only. We do not store, log, or view sensitive prompt payloads or user inputs. Built with SOC 2 compliance standards in mind.

Ready to own your AI unit economics?

See cost, revenue, and margin per customer, feature, and model, so your pricing holds as AI usage scales.

Start free