AI project prices you see online range so widely that they don't help anyone plan. The cost of an AI solution is set far less by "AI" than by ordinary project factors: how messy the data is, how many systems it connects to, where it must be hosted, and how right it has to be.
This article is part of AI for Business: 15 Practical Use Cases Across Different Industries — our overview of where AI delivers real value for businesses.
This article covers AI projects in general — automation, document AI, integrations, assistants and agents. If you're costing a chatbot specifically, read how much a custom AI chatbot costs.
Where the money goes
| Cost component | What it covers | What makes it bigger |
|---|---|---|
| Discovery | Problem definition, data review, architecture, prototype | Unclear goals; many stakeholders |
| Data preparation | Collecting, cleaning, OCR, deduplication, structuring | Scanned documents, many versions, inconsistent records |
| AI build | Retrieval pipeline, prompts, extraction logic, evaluation set | Higher accuracy targets; several document types |
| Integrations | Connections to CRM, helpdesk, databases, store, chat tools | Number of systems; write access; poor APIs |
| Interface | Widget, portal, dashboards, admin panel | Custom workflows, roles, reporting |
| Security and hosting setup | Access control, audit logs, private deployment | On-premise or strict data residency |
| Testing and launch | Accuracy review, user testing, staging, rollout | High-stakes domains (legal, finance, health-related) |
How cost varies by type of AI project
| Project type | Typical shape | Relative cost |
|---|---|---|
| Generative AI feature in an existing app | Summaries, drafts or Q&A using an AI API | Lower — often around 3–4 weeks of work in our projects |
| Task automation | Extraction or triage with human review | Lower–Medium, depending on document variety |
| Custom AI assistant | RAG over your documents, interface, admin panel, one or two integrations | Medium — around 6–8 weeks |
| AI agent / multi-step automation | Actions across several systems with approvals and logging | Medium–Higher — every action and failure path must be built and tested |
| Multi-module AI platform | Several document libraries, workflows, roles, dashboards | Higher — around 10–12 weeks, such as a full legal research platform |
Not sure which type you need? The 15 AI use cases guide maps common problems to these project types, and chatbots vs agents explains why agents cost more.
Running costs after launch
- AI model usage — billed per token. As a rough guide, 100–500 queries a day typically costs around $20–$80 a month; 1,000–5,000 queries a day around $100–$400 a month. Heavy document processing and agents use more.
- Hosting — the application, search index or vector database, and storage.
- Support and improvement — fixes, model updates, reviewing what the AI couldn't handle.
- Content upkeep — your team's time keeping documents current.
Caching, routing simple requests to cheaper models, and open-source models on your own servers can all reduce running costs at scale.
Pricing models
- Fixed-price discovery (typically 1–2 weeks): architecture, data readiness, a realistic scope and cost, often a prototype. With us, the fee is credited against the full project if you go ahead.
- Fixed-scope project: a one-time fee to design, build and deploy; you own the result.
- Retainer: monthly engineering for evolving products — features, tuning, infrastructure.
We don't charge per user or per conversation; the only usage-based cost is the AI provider's API, which you pay directly.
Weigh the cost against the value
A budget only makes sense next to the benefit. Estimate the hours the solution saves each month, multiply by the cost of those hours, subtract running costs, and divide the build cost by the result to get a payback period. The full method is in how AI automates repetitive tasks and reduces costs. Faster customer responses and fewer errors count too, even when they're harder to price.
How to budget sensibly
- Start with one use case and a clear success measure.
- Budget for data clean-up — it's the most commonly underestimated line.
- Include subject-expert time for testing.
- Ask every vendor for running costs at your volume.
- Confirm ownership of code and configuration.
- Compare against off-the-shelf options over three years — see custom vs off-the-shelf AI.
For a number specific to your project, send us your use case. We'll reply with a written scope, timeline and running-cost estimate — or suggest a short discovery if it's too early to tell.