Fixed Price vs Time and Materials for AI Software Builds
AI changes how software should be priced. Here is how fixed-price, time-and-materials and phased models compare for AI builds, and how to choose the one that protects your budget.
Why pricing models matter more for AI
How a software project is priced shapes how it is run. Fixed-price contracts push vendors to protect scope. Time-and-materials contracts push clients to watch the clock. Both models were designed for a world where engineering effort was the dominant cost and requirements could be written down in advance.
AI projects break some of those assumptions. Engineering is faster because AI assists with architecture, code and testing, but outcomes such as model accuracy are less predictable up front. Choosing the right commercial model is therefore part of managing risk, not just a procurement detail.
Fixed price: when it works
A fixed-price contract sets a single price for a defined scope. It gives buyers budget certainty and shifts delivery risk to the vendor. It works well when the scope is genuinely clear.
- Well-understood workflows, such as a client portal or booking system.
- Replacements for existing systems where the behaviour is already known.
- Clearly bounded integrations between two systems with documented APIs.
The hidden costs of fixed price
Fixed price has a downside: vendors add contingency to cover uncertainty, and every change becomes a negotiation. If the first weeks reveal that users need something different, a fixed-price contract makes the right change expensive. The buyer gets certainty on price but sometimes pays for the wrong product.
Time and materials: when it works
Time and materials charges for the effort spent. It is flexible and suits work where the destination is uncertain, such as research-heavy AI features, experimental products or ongoing improvement of an existing platform.
The hidden costs of time and materials
The risk moves to the buyer. Without strong governance, scope drifts and budgets expand. Time and materials works best with short cycles, visible progress and a clear definition of value for each cycle, which is exactly what AI-accelerated delivery makes possible.
The phased model we recommend
For most AI builds, we use a phased model that combines the strengths of both. Each phase has a fixed price and a defined outcome, and the next phase is scoped using what was learned in the previous one.
- Discovery sprint: a fixed-price, short engagement that validates data, AI feasibility and the first release scope.
- Build phase: a fixed-price first production release focused on one core workflow.
- Run: a predictable monthly fee for hosting, monitoring, model costs and support.
- Change: prioritised improvements delivered in fixed-price increments or a monthly capacity allocation.
This structure gives budget certainty at every step without locking the buyer into a specification written before anyone saw real usage. It is how we price our custom AI software development work, and the detail is on our pricing page.
Why AI makes phasing cheaper
Phasing used to carry overhead: each phase needed planning, handover and re-estimation. With AI assisting architecture drafting, code generation and test coverage, those overheads shrink and phases can be short, often two to six weeks. Our article on building AI software with AI explains why delivery timelines have collapsed.
Budgeting for AI-specific costs
AI software carries running costs that traditional software does not. Model usage is billed per request or per token, retrieval indexes need storage and monitoring requires tooling. These costs should be estimated up front and reviewed monthly.
- Model inference costs, which depend on volume, model choice and prompt length.
- Data storage and indexing for retrieval.
- Monitoring, logging and evaluation tooling.
- Periodic model evaluation and prompt maintenance.
Routing simple requests to cheaper models and reserving premium models for complex tasks can cut inference costs substantially. Our guide to the cost of ownership of AI-built software sets out a full budgeting framework.
Choosing the right model for your project
A simple test helps. If you could write the acceptance criteria today and would be happy if the vendor delivered exactly that, fixed price is reasonable. If you expect to learn important things once users touch the product, phased delivery protects you better. If the work is exploratory or ongoing, time and materials with tight governance makes sense.
Questions to ask any vendor
- What exactly is included in the price, and what triggers a change request?
- Who owns the code, data and models at the end of the engagement?
- How are AI running costs estimated, reported and controlled?
- What happens if a phase shows the original plan was wrong?
Ownership deserves special attention. Some vendors price low and recover margin through platform lock-in. We hand over the codebase and data, as explained in our article on how to choose an AI software development company.
The cheapest contract is the one that lets you change your mind without paying for it twice.
Examples across industries
Pricing structure often follows the industry. Professional services firms usually prefer fixed-price phases that match their own budgeting cycles, as described on our AI software for professional services page. Membership organisations often combine a fixed build with a monthly run fee, which suits the annual renewal cycle described in our AI software for associations & membership work. Automation programmes frequently use monthly capacity, because new workflows are added continuously, as outlined on our AI agents and business process automation page.
Switchpoint Software Design is an AI software development company that prices AI builds so budgets stay predictable while products stay flexible. Measured results are in our AI software case studies. Book a strategy call and we will recommend the commercial structure that fits your project.