How It Works
Base Hours = Predefined hours per phase based on project type
Complexity Multiplier = Simple (1x), Medium (1.6x), Complex (2.5x)
Feature Hours = Additional hours per feature, scaled by complexity
Total Hours = Sum of all phase hours (Discovery + Development + Testing + Deployment + Documentation)
Total Cost = Total Hours x Hourly Rate
Timeline = Total Hours / (Team Size x 30 productive hours/week)
Maintenance = 20% of Total Build Cost per year
How to Scope an AI Project: A Practical Guide
Scoping an AI project is fundamentally different from scoping traditional software. AI projects carry unique uncertainties around data quality, model performance, and iterative refinement that make fixed-scope estimates challenging. This calculator provides a structured starting point by breaking the project into standard phases and adjusting for project type, complexity, and feature requirements. Use it as a baseline for conversations with clients or stakeholders, not as a rigid contract.
The Five Phases of an AI Project
Every AI project, regardless of type, follows a similar lifecycle. The discovery phase involves understanding the business problem, auditing available data, defining success metrics, and selecting the right approach (fine-tuning vs. prompting vs. RAG vs. custom training). The development phase is the bulk of the work: building the pipeline, engineering prompts or training models, creating the application layer, and integrating with existing systems. The testing phase covers unit tests, integration tests, model evaluation, edge case handling, and user acceptance testing. The deployment phase includes infrastructure setup, CI/CD pipelines, monitoring, and the go-live process. Finally, the documentation phase produces technical documentation, user guides, and runbooks for ongoing maintenance.
Why Complexity Matters More Than You Think
A "simple" chatbot that answers FAQs from a static knowledge base is a fundamentally different project from a "complex" chatbot that handles multi-turn conversations, integrates with CRM and ticketing systems, supports multiple languages, and requires custom model fine-tuning. The complexity multiplier in this calculator reflects the non-linear increase in effort: a complex project is not just twice as hard as a simple one, it is typically 2 to 3 times as hard because of the exponential increase in edge cases, integration points, and testing requirements.
Estimating with Features
Each additional feature adds hours across multiple phases. API integration adds development hours for building the integration layer plus testing hours for ensuring reliability. Custom training requires additional discovery (data collection and preparation), significant development time, and extensive testing to validate model quality. Dashboard features add frontend development and design work. When stacking multiple features, there is often a compounding effect where features interact with each other in ways that add complexity beyond their individual estimates.
The 20% Maintenance Rule
AI systems are not "set it and forget it." Models degrade over time as the real world shifts (a phenomenon called model drift). APIs change pricing and capabilities. User needs evolve. The industry standard is to budget 15-25% of the original build cost annually for ongoing maintenance. This covers monitoring model performance, retraining when accuracy drops, updating integrations when APIs change, fixing bugs that emerge from new edge cases, and making incremental improvements based on user feedback. This calculator uses 20% as a balanced middle-ground estimate.
Setting Client Expectations
The most common source of friction in AI projects is misaligned expectations. Clients often expect AI to be perfect out of the gate, when in reality even the best models have failure modes. Include a buffer of 15-20% in your estimates for iteration and refinement. Communicate clearly that the initial deployment is version 1.0 and that the system will improve over time with real-world data and feedback. Frame the maintenance budget not as an ongoing expense but as the investment that turns a good AI system into a great one.