Pilot Pricing by Complexity
Three tiers, one rule: scope, price, and definition of done agreed before we start.
Single-Workflow Agent
- ✔ One workflow, one core integration
- ✔ Tool-use with guardrails
- ✔ Human-in-the-loop approvals
- ✔ Trace logs & basic monitoring
Example: An agent that reads incoming invoices, matches them to purchase orders, and queues payments for approval.
Multi-Step Operations Agent
- ✔ Multi-step workflows across systems
- ✔ 2–4 integrations (CRM, ERP, email…)
- ✔ Memory & state management
- ✔ Monitoring dashboard & alerts
Example: An agent that handles tier-1 support end to end: reads tickets, checks account data, takes action, escalates edge cases.
Multi-Agent System
- ✔ Orchestrated agents for a whole function
- ✔ Complex integrations & data pipelines
- ✔ Approval gates on sensitive steps
- ✔ Full observability & handover
Example: Finance OS for Krish Ventures — agents that run AP/AR, reconciliation, and reporting for the whole finance department.
What Moves the Price Within a Tier
Number of integrations
Each system the agent must read from or act on (ERP, CRM, banking, email) adds connectors, permissions, and testing.
Data messiness
Clean structured data is fast; scanned PDFs, free-form emails, and inconsistent formats need extraction and validation layers.
Risk & approvals
Agents that move money or touch customers need approval gates, audit trails, and stricter testing than read-only agents.
Model hosting
API models are fastest to ship; fully local models (for data-sensitive teams) add setup time — see our private RAG services.
Accuracy bar
"Drafts for a human to review" ships weeks earlier than "acts autonomously" — we help you pick the right bar per step.
Existing documentation
A documented process maps straight into an agent plan; undocumented tribal knowledge needs discovery interviews first.
AI Agents vs RPA vs Chatbots
Same budget, very different capabilities.
| AI Agents | RPA | Chatbots | |
|---|---|---|---|
| Best for | Judgment-heavy workflows: finance ops, support resolution, research | Stable, repetitive clicks and data entry that never change | Answering questions and routing conversations |
| How it works | LLM plans steps, calls tools/APIs, and acts — with approval gates | Scripted bots replay recorded UI steps | Matches intents or generates replies; doesn't act on systems |
| Unstructured data (emails, PDFs) | Yes — core strength | No — needs structured input | Reads it, but can't process it into systems |
| When the process changes | Adapts; re-planning is built in | Breaks; scripts must be rebuilt | Unaffected, but wasn't doing the work anyway |
| Typical build cost | $5k–$25k fixed-scope pilot | $5k–$50k+ plus per-bot licenses | $1k–$15k depending on depth |
| Time to value | 1–8 weeks to a production pilot | Weeks–months, plus license procurement | Days–weeks |
Get a Number, Not a Sales Call
Describe your workflow in a 30-minute call and we'll come back with a fixed scope, a fixed price, and a delivery date.
Book Your Scoping Call