CollectivIQ AI platform adds controls for enterprise model spending
CollectivIQ says its platform routes employee AI use by cost and model tier as companies face rising token bills from agents and reasoning models.
By Dominic Okoye · Staff Writer
· 3 min read
Boston-based CollectivIQ is pitching its AI platform as a cost-control layer for companies expanding generative AI beyond pilots, with enterprise pricing Chief Executive John Davie put at about $10 per user per month for customers with at least 100 users. The company says the product matters because token-based AI usage, especially from reasoning models, background jobs and agents, can create bills that managers do not see until after the spending has happened.
CollectivIQ describes the software as an AI consensus platform. Administrators can decide which employees get access to which classes of models based on role, department, business need and budget. The system also supports companywide spending caps, plus daily cost and token limits for individual users.
The company said the controls give visibility into usage at both the worker and organization level and can limit background and agentic activity. CollectivIQ did not disclose revenue or headcount. It said it has about 25 enterprise accounts covering hundreds of users.
What is the CollectivIQ AI platform?
CollectivIQ connects to models from OpenAI, Google, Anthropic, xAI and other suppliers rather than building around a single model provider. Users can send a query to multiple models and receive a combined answer that shows where the systems agree or conflict.
The company is selling that multi-model comparison as a way to reduce reliance on one AI vendor and to spot hallucinations. Davie said model makers are unlikely to check their output against competitors, so CollectivIQ adds its own comparison layer.
That design has an obvious cost problem: asking several models can burn more tokens than asking one. CollectivIQ says its answer is routing. Its new Auto Mode evaluates a prompt and sends it to a low-, medium- or high-cost model tier depending on the task. Routine requests can go to cheaper models, while more subjective or reasoning-heavy work can be sent to higher-cost systems.
Davie said the company initially used the strongest models for every request, but changed that approach as reasoning models became slower and more expensive. Users who choose the highest tier should expect longer response times, he said, while faster models return answers more quickly.
Why AI cost controls are becoming a budget issue
Enterprise AI spending is moving from controlled experiments into everyday workflows, where token usage can be harder to predict. The issue is not limited to chat prompts. Agents and background processes can continue consuming tokens without a clear employee action attached to every request.
Recent examples cited in industry reporting include Uber Technologies using its annual AI coding budget in four months and Swan AI running up a $113,000 monthly bill for a four-person team. Those cases help explain why cost governance is becoming a separate buying category rather than a footnote inside broader AI adoption plans.
CollectivIQ’s pricing model also differs from per-seat premium AI subscriptions. Davie said the company uses model APIs and pays for consumed tokens, then routes requests according to difficulty. He compared the roughly $10 per user per month enterprise price with some individual premium AI subscriptions that cost $40 per month.
The product grew out of Davie’s work at Buyers Edge Platform, the restaurant and food-service procurement technology company he also leads. He said giving every employee a standard enterprise AI license would have cost about $700,000 a year, while limiting staff to a weaker generic tool was not the outcome he wanted.
For its consensus output, CollectivIQ uses what it calls arbiter and judge stages to compare model responses. If most models agree on a number or claim, the system can down-rank an outlier while still showing the user the disagreement. A newer argue mode has models challenge one another’s answers, and users can still inspect the individual model responses.
CollectivIQ said customers can also connect their own models. The company is self-funded and expects to seek outside capital within the next three months, according to Davie. A broader public launch is planned for early August.
This story draws on original reporting from SiliconANGLE.
