Viksya › 03 — ESTIMATE

AI Project Estimation Model

What will this AI project realistically cost — and how confident are you in that number?

Produce P50, P80, and P90 confidence-ranged estimates for AI projects — with non-linearity multipliers, LLM token cost modelling, and CAPEX/OPEX split. Not a flat contingency buffer.

P50/P80/P9011 TabsPERT-BasedLLM Token CostingExcel Workbook
Get Instant Access
AI Project Estimation Model
$97 USD · one-time purchase

Delivered as an Excel workbook. Download immediately after purchase.

Format Excel .xlsx  ·  Pages 11 tabs  ·  Manual 50-page PDF
Get the Estimation Model → ← Back to all tools

■ Instant download  ·  ■ No macros  ·  ■ Excel 2016+

The Problem

Flat contingency buffers are not estimates. They are apologies written in advance.

Most AI project estimates use a single-point figure plus a 20% buffer — which tells a PMO nothing about confidence, risk distribution, or the real cost of AI-specific complexity. When scope shifts, the whole number collapses.

📈
Single-point estimatesOne number with a buffer is not a range — it is false precision. Boards and PMOs need to understand the confidence interval, not just the midpoint.
🤖
AI complexity ignoredStandard project estimation tools have no concept of model iteration cycles, data pipeline complexity, or prompt engineering overhead. These are real cost drivers.
💰
LLM costs invisibleToken-based inference costs are a new and unpredictable line item in AI programmes. Most estimation models do not model them at all.
📝
No CAPEX/OPEX splitFinance and PMOs need to know how the estimate splits between capital and operational expenditure. A single-line estimate cannot answer that question.
How It Works

The scoring framework.

Eleven tabs covering every dimension of AI project cost — from PERT-based ranging to non-linearity multipliers and LLM token cost modelling.

01
PERT EngineOptimistic / Most Likely / Pessimistic inputs per work package — produces P50/P80/P90 ranges
02
Non-Linearity MultiplierAdjusts estimates for AI-specific complexity: data maturity, model novelty, integration burden
03
LLM Token Cost ModelModels inference cost across GPT-4, Claude, Gemini and open-source alternatives by use case volume
04
CAPEX / OPEX SplitAutomatically splits each work package between capital and operational expenditure for finance reporting
05
Team CompositionRole-based effort allocation: data engineers, ML engineers, MLOps, PM, QA, change management
06
Delivery PhasesDiscovery, Alpha, Beta, Live — effort and cost profiled per phase
Key Term
Estimation Model (P50/P80/P90)
A method for projecting cost or effort as a confidence range rather than a single number — P50 is the median case, P80 and P90 the ranges executives use for contingency planning.
What’s Inside

Every tab, explained.

Open the workbook and know exactly where to go and what to enter.

TAB 1
Instructions
How to use PERT inputs and read the confidence range outputs
TAB 2
Work Package Input
Enter tasks with Optimistic / Most Likely / Pessimistic effort estimates
TAB 3
PERT Calculator
Calculates expected effort and standard deviation per work package
TAB 4
Non-Linearity Engine
Apply AI complexity multipliers: data, model, integration, team factors
TAB 5
LLM Token Model
Estimate monthly and annual inference cost by model and volume
TAB 6
CAPEX/OPEX Split
Automated split of total estimate for finance and PMO reporting
TAB 7
P50/P80/P90 Output
Final confidence-ranged estimate with cost and duration bands
TAB 8
Team Composition
Role-level effort allocation and cost-rate modelling
TAB 9
Phase Profile
Cost and effort split across Discovery, Alpha, Beta, Live phases
TAB 10
Executive Summary
One-page estimate summary ready for steering committee
TAB 11
Assumptions Log
Document and track all estimating assumptions for audit trail
P50 Estimate
Most likely cost — 50% confidence
P80 Estimate
Risk-adjusted cost — 80% confidence
P90 Estimate
Conservative cost — 90% confidence
CAPEX/OPEX
Finance-ready expenditure split
Who It’s For

Built for the people who have to defend the number.

Delivery Leads & Engineering Managers

Stop presenting single-point estimates that collapse under scrutiny. Produce confidence-ranged figures that hold up when scope shifts.

Programme & Portfolio Managers

Aggregate P50/P80/P90 estimates across multiple AI workstreams for accurate portfolio-level cost planning and contingency modelling.

CIOs & CTOs

Walk into investment approval meetings with an estimate that includes a confidence range, a CAPEX/OPEX split, and an LLM cost projection — not a guess with a buffer.

Part of the AI Transformation Suite

This tool is included in the AI Transformation Suite bundle alongside the Readiness Assessment and ROI Model.

See the Suite →
Frequently Asked

Questions buyers ask before estimating a project.

What is a P50/P80/P90 estimate?

A method for projecting cost or effort as a confidence range rather than a single number. P50 is the most likely case (50% confidence), P80 and P90 are the more conservative ranges executives use for contingency planning and board sign-off.

Why not use a flat contingency buffer instead?

A single number plus a flat buffer is false precision. It tells a PMO nothing about which assumptions drive the range or how confident the estimate actually is. A confidence-ranged estimate makes the uncertainty visible and defensible instead of hiding it inside one number.

How does this model handle LLM and token costs?

A dedicated tab models monthly and annual inference cost by model choice and usage volume, so token-based inference — a cost line most standard estimation tools ignore entirely — is included in the estimate, not bolted on afterward.

AI Project Estimation Model

Produce P50, P80, and P90 confidence-ranged estimates for AI projects — with non-linearity multipliers, LLM token cost modelling, and CAPEX/OPEX split. Not a flat contingency buffer.

Get the Estimation Model →