Automate the Boring Bits, Not the Judgement Calls

Most automation goes wrong for the same reason. Somebody buys a tool, then goes looking for work to put into it. The right order is the opposite: find the work first, look at it honestly, and only then ask whether software should touch it.

The dividing line is simpler than most vendors make it sound. Automate the things that happen the same way every time. Leave alone the things where a person has to weigh something up. The hard part is that plenty of tasks look repetitive from the outside and turn out to be judgement in disguise — and a few things that feel like judgement are actually a rule nobody has written down yet.

The three-question test

Before you automate a task, put it through these three questions. It needs a yes to all three.

1. Can you write the rule in one sentence, with no "it depends"?

"When an enquiry form is submitted, send the enquiry to the sales WhatsApp group and create a record with the name, number and product." That is a sentence. No judgement in it.

Now try: "When a customer asks for a discount, decide whether to give it." You cannot finish that sentence without it depends — on the order size, on whether they are a repeat buyer, on how the month is going, on whether the person asking will walk away. That is judgement. Do not automate it. You can automate the surrounding work — pull up their order history, flag that they have bought three times before — but the decision stays with a human.

2. Is every input the rule needs already recorded somewhere?

Automation cannot read a room, a tone of voice, or a conversation that happened on a phone call and was never written down. If the rule needs "whether the customer sounded serious", it fails.

This question is where most small-business automation quietly dies. The rule is clean, but half the information lives in one person's head or one salesperson's personal WhatsApp. Fix the recording first. A task is not ready for automation until the data it needs exists in a system, which is usually why a CRM — a customer record system — comes before automation rather than after it.

3. If it fires wrongly, is the damage small and reversible?

A follow-up email sent to someone who already bought is mildly annoying. An invoice auto-issued at the wrong amount, or a stock level auto-adjusted wrongly, costs real money and takes real time to unwind. Automations fail silently and they fail at three in the morning. The ones you should build first are the ones where being wrong is embarrassing rather than expensive.

What to automate first

In most small Indian businesses, the same handful of tasks clear all three questions comfortably. Start here.

• Getting the enquiry to a human, fast. Form submission goes to a record, a notification, and an owner. No judgement, no missing inputs, and if it double-fires nobody is harmed.
• The acknowledgement. One message confirming you received it and saying when someone will call. This is not a sales message, so nothing rides on getting the tone perfect.
• Reminders to yourself. Quote sent five days ago with no reply, task due today, AMC expiring next month. The automation prompts a person; the person decides.
• Repeat data entry between two systems. Website order into the accounts sheet. Enquiry into the CRM. Anything a human currently does by copying and pasting from one screen into another is pure mechanical work.
• Document generation. Quotes, proforma invoices, delivery challans built from fields that already exist. The template is the rule.
• Reporting. A weekly summary that lands in your inbox without anyone building it. The numbers already exist; assembling them is clerical.

Notice the pattern. Almost all of these move information or prompt a human. Very few of them decide anything. That is what good small-business automation work mostly looks like — plumbing, not brains.

What to leave alone, including things we could sell you

Some of this is sold hard right now. We would not buy it in a small business.

A chatbot that answers anything a customer asks. Where these work is a narrow, well-fenced job: opening hours, order status, a price list, routing to the right person. Where they fail is exactly the conversation that matters, the one where the customer is half-convinced and asks something specific. A confident wrong answer from a bot costs you the sale and you never find out it happened. If you want one, fence it tightly and give an obvious route to a human on every screen.

Auto-posted social content. Scheduling posts a human wrote is fine and sensible. Software generating and publishing posts with nobody reading them first is how a furniture shop ends up posting a festival greeting on the morning of a local tragedy. Social work is judgement wearing a calendar's clothes.

Automatic replies to reviews. A review is a person telling you something in public. A generic reply signals that nobody read it, which is worse than no reply. Automate the alert that a review has landed. Write the reply yourself.

Lead scoring you did not define. A tool that ranks your leads hot and cold using rules it will not show you is a black box making your sales decisions. If you want scoring, write the rules yourself in plain language — spend above a threshold, specific pincode, repeat customer — and keep them visible.

The full suite, on day one. ERPs and all-in-one platforms sold to twelve-person companies usually end with two modules in use and ten paid for. Buy the piece that solves the task you actually listed.

The tasks that look like judgement but are not

There is a third category worth hunting for: work that feels like it needs a person, but only because the rule has never been written down.

Deciding which salesperson gets a new enquiry often feels like judgement. Ask why the allocation happened that way three times in a row, and it is usually "whoever handles that area" or "whoever is free". Both are rules. Same with which enquiries get chased first, which customers get a Diwali message, which quotes get revised. Watch yourself do it, write down what you actually did, and see whether the sentence holds without an "it depends".

This is the highest-value automation you will find, because nobody sells it as a product — it only shows up when you look at your own process.

What nobody tells you about the cost

Every automation you build is something that can break quietly. WhatsApp changes its rules, a payment gateway changes a field name, a form plugin updates, somebody renames a column in a sheet. The automation stops and nothing announces it. Weeks later you notice the enquiries stopped arriving.

So budget for two things people forget. First, a failure alert: if the automation does not run today, somebody hears about it. Second, a manual fallback — a way to do the job by hand for a week without the business stopping. Anything that has no fallback should not be automated at all. This is also why ongoing support matters more for automation than for a static website; a website that breaks is visible, an automation that breaks is not.

A reasonable rule of thumb: if a task takes ten minutes a day, automating it saves you roughly an hour a week. Weigh that against the build and the maintenance. Some small automations are genuinely not worth having.

How to start this week

Take a sheet of paper. For three working days, write down every task that you or your team did more than twice that day. Do not filter, just list.

At the end, run each line through the three questions. One sentence with no "it depends". All inputs already recorded. Small, reversible damage if it misfires. Usually two or three items survive. Build those, one at a time, with a failure alert on each. Leave the rest alone for now — some of them will become automatable later, once the information they need is actually being captured.

If your list keeps stalling at question two because the information is not written down anywhere, that is the real project, and it is a record-keeping or website problem rather than an automation one. Send us your list and we will tell you which of the three questions each item fails — including the ones where the honest answer is that a person should keep doing it.

Questions we get asked