Monthly running costs to plan for
| Cost | What it is | Typical monthly amount |
|---|---|---|
| Model usage | Fees paid to the AI provider per question or document | $20 to $500 for most business tools |
| Search index | Storage that lets the AI find the right passages in your documents | $0 to $150 |
| Hosting | The service that runs your assistant or agent | $20 to $200 |
| Monitoring and tuning | Reviewing answers, updating content, improving accuracy | Included in a care plan from $350 |
Usage costs depend on volume. We estimate them from your expected number of questions or documents before you commit.
Typical price ranges
| Project | Typical range | Typical timeline from approved scope |
|---|---|---|
| AI assistant on your content A chat assistant that answers from your documents, site, or help center. | $4,000 to $12,000 | 1 to 3 weeks |
| AI feature in your product Search, summaries, classification, or drafting built into the software you have. | $8,000 to $18,000 | 2 to 4 weeks |
| AI agent connected to your systems An agent that reads, decides, and takes actions in your tools, with approvals. | $16,000 to $32,000 | 4 to 6 weeks |
| Custom model or AI platform Trained models, evaluation, and the platform to run AI across the business. | $40,000 to $100,000+ | 2 to 3 months |
The timeline starts once the scope is approved and we have what we need from you.
What moves the price
- Amount and quality of your data
- Systems the AI must connect to
- Accuracy and review needs
- Usage volume and running costs
What AI integration means in practice
AI integration means putting an AI model to work inside the tools and processes you already have. For most businesses, that is not building a model from scratch. It is connecting a proven model from a provider such as OpenAI, Anthropic, or Google to your own data, and wrapping it in rules, checks, and screens your team can trust.
The price reflects that work. The model itself is rented. You pay for connecting it to the right information, deciding what it may and may not do, testing its answers, and making it easy to use.
The four types of AI project
An AI assistant answers questions from your own content, such as policies, product manuals, or past support tickets. It uses a technique called retrieval augmented generation (RAG), which finds the relevant passages first and then writes an answer from them, so replies are grounded in your documents rather than guesswork.
An AI feature adds intelligence to a product you already run: summaries of long records, smarter search, automatic tagging, or draft replies a person approves.
An AI agent goes further and takes actions, such as creating a ticket, updating a record, or sending an email, within limits you set. Agents need careful permissions and a person to approve anything that matters.
A custom model or AI platform is trained or tuned on your own data, usually for prediction, classification, or vision. It suits businesses with large, specific datasets and a problem general models handle poorly.
What moves the price
Your data matters most. Clean, well-organized documents make an assistant quick to build. Scattered files, scanned PDFs, and inconsistent records take time to prepare.
The systems involved come next. An assistant that reads documents is simpler than an agent that writes to your customer relationship management (CRM) system, your ticketing tool, and your inbox.
Accuracy requirements change the testing effort. A tool that drafts marketing ideas needs light checks. A tool that answers customers about prices or policies needs a test set of real questions, measured accuracy, and a clear route to a person.
Key points
- The state of your data
- The systems the AI reads from and writes to
- How accurate it must be, and who reviews it
- Expected volume, which drives running costs
What is included in the price
Every AI quote from us starts with a short workshop to choose the one use case worth doing first. We then prepare your data, connect the model, and build the screens or integrations your team uses. Before launch we test against a set of real questions and report accuracy, so you know how well it performs. Launch includes usage monitoring, cost alerts, and a review process for answers that need a human.
A typical timeline
An assistant on your content goes live in one to three weeks. An AI feature in your product takes two to four weeks. Agents take four to six weeks, because each action they can take is tested separately. Custom models take two to three months, including data preparation and evaluation.
A worked example
Picture a property management company with a support team answering the same tenant questions every day: rent dates, repair requests, parking rules, and move-out steps. Answers live in a policy folder and old emails.
The first project is an AI assistant on the company's own documents, available to staff first and tenants second. It answers with a link to the source passage, and hands anything about money or disputes to a person. It falls in the $6,000 to $10,000 range and launches in about two weeks, with usage costs of around $60 a month at their volume.
Once it proves reliable, a second project lets it create repair tickets automatically, turning it into a simple agent.
How to keep AI costs under control
Start with one use case that saves measurable time. Measure accuracy before and after launch. Cap usage with alerts so a spike never becomes a surprise bill. Keep a person in the loop for any answer involving money, health, legal matters, or personal data.
How to compare AI quotes
AI quotes vary more than most, because some suppliers price only the connection to the model and leave data preparation, testing, and monitoring out. Ask each supplier how they will prepare your data, how they will measure accuracy, and what happens when the AI does not know the answer.
Ask which model provider they plan to use, why, and how easily it could be swapped later. Providers change their prices and models often, and you should not be locked in. Ask for an estimate of monthly usage costs at your expected volume, written into the quote. Finally, confirm that you own the prompts, the data pipeline, and the code, so another team could run the system if needed.