Book A Call
  1. Home
  2. Insights
  3. AI app development cost in the US

Cost and budgeting

AI app development cost in the US

Zahid Younis
Co-founder and CTO · · 5 min read
AI app development cost in the US

Adding an AI feature to an app in the US typically costs $20,000 to $80,000 to build, while a full AI product built from scratch usually costs $80,000 to $300,000 or more. Running costs are the part people forget: model and infrastructure bills can range from a few hundred dollars to tens of thousands a month, depending on usage and how carefully the feature is designed.

We build AI products and AI features, from document assistants to the satellite image classification prototype we delivered for the Bavarian government. This guide gives US buyers realistic build and running costs in dollars, and shows where the money goes. Our UK guide, How much does it cost to add AI to an app, covers the same topic in pounds.

Key facts

  • AI chatbot on your own content: $20,000 to $60,000 to build.
  • Retrieval-augmented generation (RAG) assistant over company documents: $40,000 to $120,000.
  • AI agent that takes actions in other systems: $60,000 to $200,000.
  • Custom computer vision model: $50,000 to $250,000.
  • Running costs: $200 to $20,000+ a month; design choices can cut them by 50 to 90%.

How much does an AI feature or product cost to build?

All figures in this guide are market estimates, not quotes. Real prices depend on scope, data quality and supplier.

AI project type What it includes US agency UK studio (in $) Timeline
Chatbot on your content Hosted model, prompt design, guardrails, chat UI $30,000 to $60,000 $20,000 to $40,000 4 to 8 weeks
RAG document assistant Ingestion, embeddings, vector search, citations, access control $60,000 to $120,000 $40,000 to $80,000 8 to 14 weeks
AI agent Tool use, integrations, approvals, monitoring $90,000 to $200,000 $60,000 to $140,000 3 to 6 months
Document processing OCR, extraction, validation, human review queue $50,000 to $120,000 $35,000 to $80,000 8 to 14 weeks
Custom computer vision Data labelling, training, evaluation, deployment $80,000 to $250,000 $50,000 to $170,000 3 to 6 months
Full AI product (SaaS) All of the above plus accounts, billing and admin $150,000 to $400,000+ $100,000 to $300,000+ 5 to 10 months

RAG for business owners: AI that knows your documents explains RAG, AI agents inside apps: use cases that pay back covers AI agents, and AI document processing for UK businesses covers document processing.

Where does the build money go?

  1. Data preparation. Cleaning, structuring and permissioning your documents or data. Often 20 to 40% of the effort.
  2. Evaluation. Building test sets and measuring answer quality, so you know the feature works before customers rely on it.
  3. Guardrails. Preventing wrong, unsafe or off-topic answers and protecting private data (post:D13).
  4. Integration. Connecting the AI to your app, your users’ permissions and your other systems.
  5. User experience. Showing sources, handling uncertainty and letting users correct the AI.

The model itself is usually the smallest part of the build cost.

How much does an AI feature cost to run?

Running costs depend on three numbers: how many requests users make, how much text goes in and out of each request, and which model handles it.

Usage level Example Typical monthly cost
Light Internal assistant for 50 staff $200 to $1,000
Medium Customer-facing chatbot, 20,000 conversations a month $1,000 to $5,000
Heavy AI features used by 100,000+ active users $5,000 to $50,000+

Model prices change often and have fallen sharply over time; always check current pricing pages, such as OpenAI pricing or Anthropic pricing. Vector databases, hosting and monitoring add to the model bill.

Planning an AI feature? We scope it, estimate build and running costs in dollars, and show you how to keep both under control. Book a 30-minute call with a co-founder.

How do you keep AI running costs down?

  • Route by difficulty. Use small, cheap models for simple tasks and larger models only when needed.
  • Cache answers to common questions.
  • Trim the context. Send only the most relevant documents, not everything.
  • Set limits per user and per day.
  • Consider on-device models for simple tasks on modern phones (post:D2).
  • Monitor cost per user from day one, not after the first big invoice.

How to keep AI running costs under control goes deeper on controlling AI running costs, and Choosing an AI model for a business app covers choosing between commercial and open-source models.

A worked example: an AI assistant for a US insurance broker

A broker with 200 staff wants an assistant that answers questions about policies, procedures and carrier rules, with sources, inside its existing web app.

Item Effort Cost at UK studio rates
Discovery, data audit and evaluation plan 3 weeks $10,000 to $14,000
Document ingestion and access control 3 weeks $10,000 to $14,000
RAG pipeline, prompts and guardrails 4 weeks $14,000 to $19,000
Chat interface with citations and feedback 3 weeks $9,000 to $13,000
Testing, evaluation and launch 2 weeks $6,000 to $9,000
Total About 3.5 months $49,000 to $69,000

Estimated running cost: $400 to $1,500 a month at expected usage.

What US rules apply to AI apps?

There is no single federal AI law, but sector rules apply to AI just as to any software: HIPAA for health data (post:G1), fair lending and FCRA rules for credit, FTC rules on deceptive claims, and state laws such as Colorado’s AI Act for high-risk decisions (its start date has been postponed, so check the current status). If you also serve EU users, the EU AI Act applies (post:G6). Rules change quickly, so check the current official text for your sector and state.

Frequently asked questions

How much does it cost to build an AI chatbot?

A business chatbot using a hosted model on your own content usually costs $20,000 to $60,000 to build, depending on integrations, guardrails and design. Running costs typically start at a few hundred dollars a month.

Is it cheaper to use ChatGPT than build a custom AI feature?

Off-the-shelf tools are cheaper to start and fine for internal productivity. A custom feature makes sense when AI needs to work inside your product, with your data and permissions, under your brand. Custom AI vs off-the-shelf ChatGPT tools for your business compares the two.

Do I need my own AI model?

Rarely. Most business AI features use hosted models with your data supplied at query time. Training your own model makes sense for specialist tasks such as computer vision or when data cannot leave your environment.

How long does it take to build an AI feature?

A focused chatbot or assistant takes 4 to 8 weeks; a RAG assistant 8 to 14 weeks; agents and custom models 3 to 6 months.

Why do AI running costs vary so much?

Costs scale with the number of requests, the amount of text processed and the model used. The same feature can cost ten times more with poor design, such as sending whole documents to the largest model for every question.

Official sources and further reading

Planning an app?

Talk it through with one of our co-founders. 30 minutes, no pressure.

Book a call