AI Project Estimation Model by Viksya

AI Project Estimation Model is an Excel template (XLSX) available for download on purchase.

Estimating an AI project differs from estimating traditional software delivery, yet most templates still apply a single number plus a flat contingency percentage. Estimates built this way tend to break down during delivery, because AI-specific cost drivers such as data readiness, training iteration cycles and token-based inference spend are not represented in the model.

This workbook produces confidence-ranged estimates. Using three-point PERT (Optimistic / Most Likely / Pessimistic), it calculates P50, P80 and P90 cost and effort ranges, so a base case, a business-case figure and a fixed-price ceiling all come from the same inputs.

WHAT THE MODEL DOES

Confidence-ranged estimates: P50, P80 and P90 cost and effort from three-point PERT, replacing single-point figures with a range.
Non-linearity engine for model development, fine-tuning and training projects: five factors (data readiness, model architecture complexity, training iteration cycles, domain novelty and infrastructure complexity) produce a composite multiplier applied phase by phase across the work breakdown structure.
LLM cost model for integration and consumption projects: token-based projections of daily, monthly and annual API spend across five pre-built use cases, with a call-volume sensitivity table from 0.5x to 5x of baseline volume.
Full cost roll-up: labour, LLM/API costs, infrastructure, overhead and contingency in one summary.

WORKBOOK STRUCTURE

A cover page plus ten working tabs:

1. Instructions: colour-coding convention and step-by-step workflow
2. Settings: currency, role-based daily rate card, LLM pricing tiers, overhead and contingency rates
3. Project Setup: project type, delivery methodology and six configurable phases
4. Effort Estimation: 36-activity work breakdown across six phases and eight roles
5. Non-Linearity: five-factor multiplier engine
6. LLM Costs: token-based cost model with call-volume sensitivity
7. Infrastructure & Licences: compute, MLOps platform and software licence cost sections
8. Risk: twelve-item risk register with probability-weighted exposure and PERT confidence ranges
9. Cost Summary: cost roll-up across all cost categories
10. Dashboard: cost ranges, effort by phase, cost mix and top risk exposure on one screen

WHO IT IS FOR

Engineering leads, AI/ML architects, programme and project managers, PMO teams and delivery heads who need to submit a defensible cost and effort estimate before an AI programme is approved, rather than a single number that will not withstand scrutiny at a steering committee or budget review.

The workbook is methodology-agnostic. Phases can represent sprints, program increments or waterfall stages, and effort is entered uniformly in person-days. Currency and rate cards are set on the Settings tab, so the model can be used in any currency.

FORMAT AND NOTES

– Excel (.xlsx), no macros or VBA
– Works in Excel 2016 and later
– All formulas and inputs are visible and editable

All estimates produced are indicative and depend on the quality of the inputs provided. Outputs should be validated against your organisation’s own assumptions, governance standards and delivery context before use in business cases or procurement decisions.

Published by Viksya.

You must log in to submit a review.