AI put where it earns its place, not where it demos well.
Most AI projects fail for the same reason most software projects fail: nobody agreed what it was supposed to replace. We start from a task somebody currently does by hand, count what it costs in hours, and only then work out whether a model can take it. If it cannot, we will say so. That is a shorter conversation than the one where you find out in month four.
Where it actually pays
Four uses we have seen return more than they cost. Each one replaces a job somebody is doing by hand today.
Answering the same questions
A support assistant trained on your own documents, prices and policies rather than on the open internet. It answers what it knows and hands over what it does not, which is the difference between one that helps and one that invents an answer and costs you a customer.
Being cited by AI search
Buyers increasingly get an answer from an AI summary instead of a list of links, and never reach the page at all. Being the source that summary quotes is a structural job: clear answers near the top of a page, real structured data, and facts stated in a form a model can lift. It overlaps heavily with SEO and is done alongside it.
Drafting at volume
Product descriptions, category copy and first drafts for a large catalogue, with a human editing rather than writing from nothing. Worth it above a few hundred items and a waste below that, where writing it properly is faster than reviewing what a model produced.
Moving data between systems
Reading an invoice, a form or an email and putting the right fields in the right place. Unglamorous, and usually the highest return of the four because it replaces a task somebody does every day and nobody enjoys.
What we will tell you not to do
The market rewards saying yes to all of this. These are the four we turn down.
A chatbot on a five page site
If a visitor can find the answer by scrolling, a chat window is a worse version of your own page with a chance of being wrong. Fix the page.
Publishing generated articles unedited
Google does not penalise AI-written content for being AI-written. It does drop content that says nothing, and unedited output almost always says nothing. Every post still needs somebody who knows the subject.
Anything that has to be right every time
Quotes, legal wording, medical or financial answers. A model is confident when it is wrong, and that is exactly the wrong property for those jobs.
Replacing a person you still need
These tools take tasks, not roles. An honest project makes a day shorter; it does not make a job disappear.
How a project runs
-
Find the task and price it
Which job, how many hours a week it takes and what those hours cost. Without that number there is no way to tell afterwards whether the project worked.
-
Look at the data it has to run on
Most of these projects fail here rather than at the model. If the inputs are inconsistent, scanned badly, or spread across three systems that disagree with each other, that is the real project, and we will say so before you spend anything on the clever part.
-
Build the smallest useful version
One workflow, running on real data, in front of the people who do the job today. A pilot that works on sample data tells you nothing about the messy inputs you actually have.
-
Put a person in the loop deliberately
Decide up front what the system may do alone, what it has to hand to somebody, and how it behaves when it is unsure. A tool that quietly gets things wrong costs more than the hours it saved.
-
Measure against the same number
Hours before against hours after, plus how often a person had to correct it. If the correction rate is high enough, the saving is imaginary and we stop.
-
Hand it over
Documented, in your accounts, with the prompts and settings in a place your team can change. Nothing that needs us to keep it running.
What it costs
A first workflow runs from ₹40,000 to ₹1,50,000 depending on how many systems it has to touch and how clean your data is. Model usage is billed to your own account, so you can see what it costs and switch providers without asking us. Confirm these figures before this page goes live. Full ranges for every service are on the pricing page.
Which models do you use?
Whichever suits the task and your budget, and we tell you which before we build. Everything we write is kept portable so a model can be swapped when a better or cheaper one appears, which in this field is every few months.
Does our data go into training?
Not on the business tiers we use, which exclude it by contract. We will show you the specific terms for whichever provider ends up in your build rather than asking you to take our word for it.
Will AI write our blog?
It can draft. Publishing a draft is how sites end up with fifty posts that rank for nothing. The useful version is a first draft in half the time with somebody who knows the subject editing it, which is what we do on our own ((blog|/blog/)).
Is this worth it for a small business?
Sometimes. If one person spends six hours a week retyping data between systems, yes. If you are hoping AI will bring customers, no: that is ((search|/seo-services/)) and ((advertising|/digital-marketing/)), and no model changes it.
What if it stops working?
Providers change models and deprecate old ones, which breaks things that were working. That is a real ongoing cost and we say so before you start rather than after. A care plan covers it, or you can take it in house with the documentation we hand over.
Name the task, not the technology
Tell us the job somebody does by hand every week. That is enough for us to say whether this is worth your money.