What to Budget Before You Brief an AI Software Studio
A five-line budget framework for operations teams commissioning custom AI workflows.
What You Need to Know
A realistic AI project budget requires five distinct line items: discovery and data scoping, core engineering, human-in-the-loop review interfaces, infrastructure API costs, and year-one maintenance. Failing to budget for the human review interface is the most common reason operations deployments fail.
At a Glance
- Cost Model
- Usage-based APIs + Custom Build
- Required Budget Lines
- 5 (Discovery, Build, Handoff, APIs, Maintenance)
- Common Omission
- Human review interface
- Ongoing Cost
- Prompt tuning & API tokens
Best For
- Operations directors planning their first custom AI software project
- Commercial leaders trying to estimate automation budgets
- IT leads managing the financial sign-off for operations tooling
Not For
- ×Consumers looking for cheap ChatGPT subscriptions
- ×Teams looking to buy off-the-shelf, per-seat SaaS products
- ×Academics researching the theoretical cost of large language models
Key Takeaways
- An AI workflow budget requires five distinct cost lines, not just a single build fee.
- Discovery costs scale with the messiness and variance of your operational data.
- You must budget for a human review dashboard where staff can approve or reject machine decisions.
- API and infrastructure costs are usage-based and scale directly with your document volume.
- Year-one maintenance is required to tune the system as real-world forms and emails change.
Most operations directors budget for an AI software build the same way they budget for a website: a flat fee for the build and a small margin for hosting. That model breaks within a week of deploying an AI workflow to the operations floor.
When you commission software that reads messy, unstructured data—like a tool that extracts line items from 400 variations of commercial invoices—you are not buying a static product. You are buying a system that makes probabilistic decisions. That requires a fundamentally different approach to funding the project.
Before you write a brief or approach a B2B product studio, your internal spreadsheet needs five distinct cost lines. If you leave out the human review interface or year-one maintenance, your workflow will either stall in development or choke your staff with unmanageable errors within a month.
Here is what drives the cost of each line item, and what you need to budget for.
Line 1: Discovery and Data Scoping
You cannot point an AI at a shared inbox and ask it to "sort the emails." Before any code is written, a studio must audit the actual mess of your inputs.
If you are sorting mixed export attachments, discovery means taking a representative sample of 1,000 historical emails to see what you are actually dealing with. How often do clients combine packing lists and commercial invoices into a single PDF? How many are scans of crumpled paper versus native digital files?
What moves this cost line:
- Input variance: Processing standard HMRC forms is a cheap, known quantity. Extracting data from handwritten supplier delivery notes is highly variable and requires extensive scoping.
- Integration points: Pulling data from a modern API is straightforward. Scraping data from a 15-year-old on-premise ERP system takes major discovery work.
Line 2: The Core Build and Orchestration
This is the line item everyone remembers: the actual engineering. It covers writing the extraction logic, setting up the prompts, routing the data, and connecting your systems.
Software studios now use AI-assisted product delivery to write core code, which drastically compresses the time from discovery to first release. Instead of taking three months to stand up a basic integration, a studio can often deliver a working prototype in weeks. However, faster coding does not mean the logic itself is simple. The build cost is dictated by the complexity of the rules you need the AI to follow.
What moves this cost line:
- Multi-step reasoning: An AI that simply tags an email as "urgent" is a cheap build. An AI that reads an email, checks the referenced order number against your database, determines if the item is in stock, and drafts a holding reply requires complex orchestration.
- Fallback logic: What happens when the AI is only 40% confident? Building the routing logic that catches and redirects uncertain outputs takes engineering time.
Line 3: Human Review and Handoff Interfaces
This is the most dangerous omission in an ops budget. Machine learning models make mistakes. If you do not build a place for your staff to catch those mistakes, the errors will pass straight into your core business systems.
Where you put the human in an AI document workflow defines the safety of the entire project. You are not just paying for a back-end script; you are paying for a front-end dashboard where an operator can hit "Approve," "Edit," or "Reject" on the machine's work.
What moves this cost line:
- Interface complexity: A simple Slack or Microsoft Teams notification asking for a "Yes/No" approval is inexpensive. A side-by-side document viewer that highlights the extracted text on the original PDF takes dedicated UI development.
- Review and hardening: Getting the system ready for live operations requires structured testing. As a baseline, basic review and hardening starts at £199, but scales up depending on how many custom dashboards your team needs to manage the queue.
Line 4: Infrastructure and API Tokens
Unlike traditional software where hosting costs a few pounds a month, AI workflows incur usage-based costs. Every time you send a document to an AI model (like OpenAI or Anthropic) for extraction, you pay for "tokens."
Processing a three-sentence email costs a fraction of a penny. Processing a 40-page technical manual costs significantly more.
What moves this cost line:
- Document volume: Running 50 queries a day versus 5,000 queries a day.
- Document length: Large PDFs consume heavy token counts.
- Model selection: Using a massive, top-tier reasoning model for every tiny task will inflate your API bill unnecessarily. Good engineering involves routing simple tasks to cheaper, faster models.
AI Software Project Cost Calculator
Line 5: Year-One Running Costs
Operations data is not static. A supplier will change the layout of their invoice. A carrier will update their portal. Exhibitors will start asking entirely new questions about a venue.
When the real world changes, your workflow's accuracy will drop. You must budget for ongoing "prompt tuning" and logic adjustments. This is not a failure of the software; it is the reality of applying machine learning to human operations. Treat this like an annual servicing budget for a physical machine.
What moves this cost line:
- Environmental stability: If your industry uses heavily standardised forms, maintenance is low. If you operate in a chaotic sector where formats change monthly, maintenance will be high.
- Internal monitoring: You will need a developer on a retainer or an internal technical lead to monitor error rates and adjust instructions when the AI starts misclassifying data.
Build the Spreadsheet Before You Brief
Do not ask a product studio "how much an AI workflow costs" without defining these five parameters first. Start by mapping your exact daily volume, the number of distinct document types you handle, and how many staff members will need access to the review dashboard.
Once those numbers are in your spreadsheet, you can hold a sensible commercial conversation about what is actually worth building.
Useful tool
Try Samvara's Document Readiness Checklist — Export/import docs by mode.
Quick Comparison
| Budget Component | Traditional Software Build | Custom AI Workflow |
|---|---|---|
| Initial Scoping | Mapping fixed user journeys | Analysing raw data variance and messiness |
| User Interface | Data entry screens | Exception handling and review queues |
| Hosting / Infrastructure | Flat monthly server fee | Variable API token costs based on volume |
| Post-Launch Maintenance | Bug fixes and security patches | Prompt tuning and adjusting to new document formats |
Frequently Asked Questions
Why do API token costs fluctuate each month?
Can we skip the discovery phase to save money?
Why does an AI workflow need ongoing maintenance?
Bottom line
Do not sign off a build budget that lacks a specific line item for human handoffs. Allocate dedicated capital to the dashboard where your staff will catch and correct machine errors, because that interface dictates whether the workflow actually survives contact with your operations floor.
How Samvara researches this guide
We write for exhibition organisers and import/export operators in Australia, the UK, Europe, the US and India. Guides favour specific, verifiable operational advice over generic tips — grounded in systems we have shipped, client workflows, and current industry practice. We revisit articles as tooling and regulations change.
Written by
Shreyansh Doshi, Founder of Samvara
Shreyansh Doshi is the founder of Samvara Technologies, a product studio building operator software and SaaS products for exhibition, import/export, travel and fitness businesses in the UK and Australia. He writes about product delivery, operations systems, and where AI does and does not belong in a real workflow.