More and more organisations in Singapore and around the world are looking for agencies that can help them do AI automation projects. I keep getting asked questions on what is the best way to price AI automation projects.
Before starting Scopeyard, I delivered AI projects for customers across healthcare, recruitment and operations. Before that, I spent six years running a profitable product development studio where we charged banks a fairly hefty daily rate. So this article is not theory. It is a practical way to think about pricing AI automation projects without undercharging yourself.
AI automation projects are often difficult to price because it could be as simple as connecting some Zapier workflows together or as complex as developing custom AI agents tied together by Langchain's orchestration layer. In either scenario, it's often just simpler to count the hours you would take to build, multiply this by an hourly rate which could be a modest $25–$50 if you're listed on Upwork to north of $200 per hour if you belong to a reputed consulting firm.
However, I am here to tell you this: don't do this. Don't be lazy with pricing.
1. First, understand what kind of automation it is
Split projects into simple, medium and complex. A good guideline could be this: Simple = one workflow, few tools, low risk. Medium = multiple tools, client data, approvals. Complex = finance, operations, sales, support, agents taking actions, high risk. Developing a sense of this categorisation helps you eventually communicate better with clients and your team.
2. Charge for discovery
Don't guess the price before understanding the workflow. Clients often don't know their own process properly. They may say "Hi I would like to automate the invoicing process using AI" but may not have the full understanding of their own end to end flow. So before giving a big fixed price, do a discovery/audit. Map the workflow, tools, edge cases, data, risks, and expected ROI. Once you understand this you could add the time and effort spent consulting for them, ideating and advising into your pricing. Here is a benchmark that you could use based on multiple sources:
| Project Type | Discovery Fee |
|---|---|
| Small automation | $1,000–$2,500 |
| Medium workflow | $2,500–$7,500 |
| Complex AI system | $7,500–$15,000+ |
3. Price based on business outcome, not the tool
Don't sell "I'll build a Zapier automation." Instead sell "we'll reduce manual lead qualification," "we'll save your team 30 hours/month," "we'll reduce finance admin."
This is the value the AI automation creates for the business when it's working well. This usually comes from time saved, faster turnaround, fewer errors, or increased team capacity.
To estimate this you could simply take the hours saved per month x cost per hour. For example 60 hours saved by the client's analyst who gets paid $50/hour means the annualised business value is 60 × $50 × 12 = $36,000. Now you could charge 10–30% of this premium. If the benefit is indirect, go lower and if it's more obvious and direct then charge the higher side.
4. Adjust for complexity
To gauge this ask questions like "How difficult the automation is to design, build, test, and launch." Look at the number of systems involved, API access, data quality, workflow branches, human approval steps, and testing required.
Here is a quick rule of thumb:
Low — 1–2 tools, simple logic, clean data, low volume
Medium — 3–5 tools, some messy data, multiple workflow paths, moderate testing needed
High — 5+ tools for custom agentic development, human-in-the-loop workflows, poor data quality, custom APIs, many edge cases, high volume.
Depending on what complexity value you've given it, add a multiple to your base pricing. Low complexity could be 1.0 or 1.2x, medium complexity 1.3x–1.7x and high complexity 1.8x–2.5x.
Measuring complexity is important. The worst feeling is often delivering an airplane and charging the price of an engine. You don't feel great, your margins suffer and your team loses motivation. One or two such projects brings morale down more than you know. Multiple such projects can head you towards bankruptcy.
5. Always think about risk. Adjust for it.
A low-risk automation can be cheaper. But if the workflow sends customer replies, updates finance records, qualifies leads, or touches sensitive data, pricing must include testing, guardrails and fallback flows.
Compute a Risk Buffer to account for this. Ask: What does the business lose if this stays manual? What happens if the automation gets it wrong?
| Business Risk | Risk Buffer | When to use |
|---|---|---|
| Low | 5%–10% | Internal workflow, easy to fix if wrong |
| Medium | 10%–20% | CRM updates, customer-facing drafts, operational workflows |
| High | 20%–40% | Finance, legal, HR, compliance, sensitive data, autonomous actions |
Add this risk buffer to your computed price. This accounts for testing, edge cases, safeguards, fallback flows, and extra care.
6. Separate build cost from running cost
Work out how you want to work out the running costs. Usually AI automation projects contain OpenAI/Anthropic/Gemini API cost, Zapier/Make task cost, hosting, vector database, monitoring, WhatsApp/email costs. Decide whether the client pays directly, reimburses, or you bundle in the ongoing maintenance plan with limits.
7. Retainers matter
AI automation is never "build once and disappear." Prompts change, APIs break, clients request improvements, edge cases appear. Serious workflows need maintenance. Initially you could offer them packages where if they sign up for a year or two year long contracts you can offer them discounts. But rule of thumb is the simple lazy equation we spoke about earlier:
Monthly Retainer = Maintenance Hours × Hourly Rate + API and other opex (point 6 above)
And that's really all that there is. So to summarise if I were to put all the seven steps above into one equation to compute pricing then that would be this:
Project Price = (Build Effort × Complexity Multiplier) + Business Value Premium + Risk Buffer + Discovery Fee
Once you start pricing AI automation projects properly, delivery becomes just as important as the price. You need a clear way to break the work into milestones, document what has been approved, show clients what changed and separate internal build tasks from client-facing progress.
Scopeyard helps AI agencies plan AI automation projects, manage milestones, collect client feedback and approvals, and keep client delivery organised. With Scopeyard's MCP server, AI assistants and coding agents can also help create milestones, update cards and keep delivery progress in sync.
I'll just end with this note — AI automation agencies should not compete on who can build the cheapest workflow. They should compete on understanding the business process and delivering something the client can trust.