AI Model Selection
The right model for every extraction. Full visibility into every cost.
- Model selected per extraction
- Real-time token tracking
- Per-user cost analytics
The Problem
Does this sound familiar?
“We couldn't tell if the AI was reading our documents correctly or just hallucinating.”
“We're paying the same processing cost for a 10-page notice as for a 300-page RFP.”
Every document runs through the same model at the same cost regardless of complexity. Simple tenders are over-processed. Complex tenders may be under-powered. Neither is optimal.
“We have no visibility into how much our team's AI extractions are actually costing.”
Extraction runs. Tokens are consumed. Nobody knows which modules, which users, or which document types are driving the cost. Budget allocation is a guess.
“We got a poor extraction result and had no way to try a different approach.”
The document was complex. The extraction missed sections. But the platform offers no model choice — so there is no way to re-run with something more capable. The result stands.
AI Model Selection gives every extraction the right tool for the job — and gives every decision-maker full visibility into what that costs. It doesn't matter whether you're a CTO managing infrastructure costs, a finance head tracking AI spend, a bid manager who needs the best extraction on a critical tender, or an admin managing team-level usage. Every extraction runs on the model that fits the task. Every cost is visible, traceable, and accountable.
How It Works
Step by step
Four steps from document upload to a tracked, costed, optimised extraction.
Step 01
Upload and select your extraction sections as normal
Start the extraction workflow — upload your tender document and select your required sections (PQ, BOQ, Risk, Annexures). The workflow proceeds as normal to the Model Selection step.
Step 02
Select the AI model for this extraction
Choose the model that best fits the document's complexity and extraction task. Each model is described with its strengths, processing depth, speed, and estimated token cost — so the selection is informed, not arbitrary.
Step 03
Extraction runs — token consumption tracked in real time
The selected model processes your document and sections. Token consumption is tracked in real time and recorded against this extraction in your History module — with model name, section extracted, user, timestamp, and cost.
Step 04
Review costs per extraction, per user, per module
Full cost analytics are available in your billing dashboard — broken down by model, extraction type, user, and time period. Re-extraction is available from History at any time — run with a different model, compare outputs, and choose the result that best serves your bid decision.
In Practice
What a real day looks like
Real scenario
A bid team runs extractions on a high volume of tenders per month. Their AI costs have been growing — but nobody knows why. Is it the BOQ extractions? Larger documents? Specific users? The platform gives them no breakdown.
They enable AI Model Selection and configure lighter models for PQ and summary extractions on smaller tenders — reserving the more capable model for BOQ-heavy infrastructure RFPs over 150 pages.
Within the first month, extraction costs drop 34% without any reduction in output quality on the tenders that matter. The billing dashboard shows exactly which document types drove the saving.
On a critical ₹200 crore EPC tender, the team selects the most capable model for BOQ extraction. The output is the most complete they have seen. The pricing team starts from a clean dataset. The bid is sharper. They win.
What You'll Stop Doing
What you'll stop doing
These are the activities AI Model Selection makes unnecessary from the first tender you run through it.
- Running every document through the same model regardless of complexity — over-spending on simple tenders, under-powering complex ones.
- Having no visibility into AI extraction costs across the team — no breakdown by user, module, or document type.
- Getting a poor extraction result with no option to try a different approach on the same document.
- Treating AI usage as a black box — cost incurred but not understood, not tracked, and not optimised.
- Allocating AI budget without data on what is driving consumption across the team.
FAQs
Frequently asked questions
Six questions that address the most common points of hesitation before a team commits to using AI Model Selection on a live tender.
Which AI models are available in Keendex?
Keendex supports multiple AI models across different capability and cost tiers. Available models are displayed at the Model Selection step with descriptions of their strengths, processing depth, and estimated cost per extraction. The model list is updated as new models become available.
Can we set a default model for our team?
Yes. Admins can configure a default model for each extraction type — for example, a lighter model as default for PQ extractions and a more capable model as default for BOQ-heavy documents. Individual users can override the default for any specific extraction.
How is token consumption calculated?
Token consumption is calculated based on the size of the document processed, the extraction sections selected, and the model used. Each extraction records the exact token count and corresponding cost — visible in the History module and billing dashboard.
Can we compare extraction outputs from two different models?
Yes. Re-run any extraction from the History module with a different model. Both versions are stored and can be compared — so your team can select the result that best serves the bid decision.
Can we restrict certain team members from selecting high-cost models to control spend?
Yes. Admins can configure model access by user role — for example, restricting standard users to a defined set of models while allowing senior bid managers or admins to access the full roster including premium-tier models. This keeps extraction costs predictable without removing capability from the team members who need it.
Can costs be tracked per project or client?
Extraction costs are tracked per user and per module. Project-level or client-level cost allocation can be achieved by filtering the billing dashboard by user or by tagging extractions to specific tender references in the History module.
Ready to see AI Model Selection on your live tender?
Book a personalised demo — we run it on a real document from your sector.