What to automate first in an Indian business
Rank candidates on four things: how often it happens, how long it takes, how bad a mistake is, and whether the information is already in a system. Start with high volume, low stakes and clean data — not with the most impressive idea.
Score the candidates, do not argue about them
List the repetitive work your team does. For each one write down four numbers: how many times a week it happens, how many minutes it takes, what it costs when it is wrong, and whether the information needed is already in a system or lives in somebody's head.
The first thing to automate is high on the first two, low on the third, and yes on the fourth. That is usually not the idea anyone was excited about, which is exactly why writing the numbers down beats discussing it.
The usual first wins
Reading a document and putting its contents into a system: invoices, purchase orders, delivery challans. High volume, tedious, and the source is already structured enough to check against.
Classifying and routing what arrives: enquiries, complaints, applications. Getting it to the right desk in seconds instead of after somebody reads it.
Assembling a report somebody currently builds by hand every week out of two systems that do not talk.
Answering the question you get forty times a week, from your own documents, with a person taking over the moment it is not routine.
What to leave alone for now
Anything where being wrong is expensive and hard to reverse — payments going out, pricing commitments to a customer, anything with a legal or tax consequence. Those can be assisted, with a person approving, but they should not run unattended early.
Anything where the knowledge is genuinely in one person's head and has never been written down. Automating that is a documentation project first, and it is worth doing, but it is not a quick win and should not be sold as one.
Keep a person in the loop where it matters
The pattern that works: the system does the work and proposes, a person approves, and the approvals become the record of what good looks like. Over time you widen what it is allowed to do alone, based on evidence rather than optimism.
The measure of success is not how much got automated. It is how many hours came back and how few new errors appeared. Both are countable, and you should agree how you will count them before you start.
What people ask us about this.
Often a normal integration will do, and it will be cheaper and more reliable. If the task is moving structured data between two systems on a rule, that is plumbing. AI earns its place where the input is messy — free text, documents, images, or a judgement that used to need a person to read it.
Measure the before. How many times a week, how many minutes each, how many errors. If nobody wrote that down first, every claim about savings afterwards is a story. It takes an afternoon and it is the difference between knowing and believing.
Most is. That is a finding, not a blocker — but it does change the order. Sometimes the first project is cleaning up one source so that the next three become possible. A good study tells you that in week one rather than month four.
If this is your problem, start here.
Bring the version of this that is happening in your business.
A senior consultant, not a salesperson. If the answer is short we will just answer it, including when the answer is that you do not need us.