You asked three vendors to quote the same automation. One came back at ₹60,000. One at ₹2.8 lakh. One proposed a retainer and declined to quote a project price at all. Nobody is scamming you. They are pricing three different systems and calling them the same thing.
This is the honest problem with "how much does AI automation cost." The question implies a rate card exists. It does not, in the same way there is no single price for "a vehicle." What determines the number is a set of choices most buyers have not made yet when they ask for the quote.
So this piece does two things. It names the eight variables that actually move the price, and it maps the three architecture tiers that most real work falls into, with ranges drawn from published market data. By the end you should be able to place your own requirement on that map before anyone quotes you.
The eight variables that set the price
Price tracks these, roughly in order of impact.
Integration count. Every system the automation touches is an authentication method, a rate limit, a data format, and a failure mode. Two integrations is an afternoon of plumbing. Six is a project. The jump is not linear because integrations interact - when system C is down, what does the automation do with the record it already pulled from A and B?
Exception handling depth. The happy path is usually 20 percent of the build. The remaining 80 percent is what happens when the API times out, the record is malformed, the customer name has an emoji in it, or two workflows try to update the same order simultaneously. A quote that only covers the happy path will be the cheapest and will also be the one that generates support tickets in month two.
Autonomy level. An automation that files a draft for a human to approve is dramatically cheaper than one that acts without review. Approval steps are not a limitation; they are a cost control. They let you ship with less exhaustive edge-case coverage because a human catches what the system misses.
Reasoning requirement. If the logic is "if X then Y," you want deterministic workflow automation. If the inputs vary and the system has to interpret them, you need a model in the loop, and that changes the economics - non-deterministic systems need more testing, more guardrails, and carry per-token running costs. We go deeper on where that line sits in agent versus workflow automation.
Volume. Volume rarely changes build cost. It changes running cost and architecture. A workflow at 50 executions a day and the same workflow at 50,000 are different systems - queueing, retry backoff, and observability all become mandatory at the higher number.
Data sensitivity. Health, financial, or personally identifying data pulls in access control, audit logging, encryption at rest, and often a self-hosted deployment rather than a managed SaaS. This can double a build, and it is not optional if the data is in scope.
Production monitoring. Somebody has to know when the automation stops working. Without alerting, failures are discovered by customers. This is the line item most commonly missing from cheap quotes and the one most reliably regretted.
Ownership after handover. Whether the vendor maintains it, your team does, or nobody does. "Nobody" is the default when this is not discussed, and it is why a meaningful share of automation spend is written off within a year.
The three tiers most work falls into
Using published 2026 market data rather than our own rate card. Indian agencies serving domestic and global clients report roughly the following bands, per lets-viz's 2026 buyer pricing survey and PrimeFrame's India implementation cost breakdown:
Tier 1 — Single workflow, deterministic
One trigger, one outcome, two or three integrations, rules-based logic, human-visible output. Think: form submission creates a CRM record, notifies a channel, and files a document.
External benchmark: ₹40,000 to ₹80,000 for a scoped build.
What pushes you out of this tier: needing the system to decide something rather than route something.
Tier 2 — Multi-workflow with real integration
Several connected workflows, four to eight integrations, genuine exception handling, some conditional logic, monitoring included. This is where most operational automation actually lives - order routing, lead qualification pipelines, reconciliation between two systems that disagree.
External benchmark: ₹1.5 lakh to ₹3 lakh for the build.
What pushes you out: unpredictable inputs that rules cannot describe, or autonomy without human approval.
Tier 3 — Agentic, with reasoning and tool selection
The system interprets varied inputs, selects tools dynamically, and handles cases nobody enumerated in advance. Requires evaluation harnesses, guardrails, fallback paths, and observability that a deterministic workflow does not.
External benchmark: ₹3 lakh to ₹8 lakh for the build, with meaningful ongoing model costs on top.
Globally the same tier prices at roughly $3,600 to $9,600 per the same surveys, which is worth knowing if you are comparing an Indian vendor against an overseas one.
Where the ongoing cost actually lands
Build price is the number everyone negotiates. Running cost is the number that decides whether the automation survives.
Published 2026 data puts Indian SaaS-tool spend for automation platforms at roughly ₹15,000 to ₹30,000 per month, in-house automation teams at ₹5 lakh to ₹10 lakh per month fully loaded, and project consulting engagements at ₹2 lakh to ₹8 lakh. Global agency retainers sit around $2,500 to $8,000 per month.
The comparison that matters is not build-versus-build. It is two-year total cost of ownership, which means build plus platform plus model usage plus monitoring plus the change requests you will inevitably raise when the business process shifts.
A cheaper build on per-execution pricing can overtake a more expensive build inside twelve months if volume grows. Ask for the two-year number.
The ROI question, handled honestly
Market surveys report payback periods of two to four months and SMB spend of $300 to $2,000 monthly displacing $5,000 to $15,000 monthly of manual labour. Those are real published figures, and they are also selection-biased - the automations that get written up in case studies are the ones that worked.
The version we would stand behind: automation pays back fastest where the process is high-frequency, low-judgment, and currently done by someone expensive. It pays back slowest where the process is rare, requires context, or is already cheap to do manually.
That is why the sequencing question matters more than the budget question, and why we treat it separately in what to automate first.
What to settle before you ask anyone to quote
Three things. Vendors cannot price well without them, and you cannot evaluate quotes without them either.
Name the workflow, not the ambition. "Automate our operations" cannot be priced. "When a COD order is placed to a pincode on our risk list, verify by WhatsApp before dispatch" can be.
Measure the current process. How many times a week does it run, how long does each run take, who does it, and what does a mistake cost? Without this you have no baseline, which means no way to evaluate the quote and no way to prove the outcome afterwards.
Decide who owns it after handover. Vendor-maintained, in-house, or a retained arrangement. Automations with no named owner degrade quietly - an API changes, a credential expires, and nobody notices until a customer does.
If you want the workflow scoped properly before any number gets attached to it, book an automation diagnostic. We will map the process, name the integrations, flag the exception paths, and tell you which tier it genuinely falls into - including when the honest answer is that it should not be automated yet.
Discussing the workflow first and the price second is not a stalling tactic. It is the only sequence that produces a number either side can defend.
