What to Automate First: Tally, WhatsApp, Excel

What business processes should a small manufacturer automate first?

The ones where your people retype the same information from one place into another, every day, by rules they could write down. In most Indian SMEs that work sits between the same few tools: TallyPrime for accounts and GST, WhatsApp for orders, approvals and supplier messages, Excel or Google Sheets for planning, stock and reports, a CRM (or a spreadsheet acting as one) for leads, and email for purchase orders, bills and statements.

The first automation should not be the most impressive one. It should be the one your team will trust fastest, because it runs often, is easy to check and does little harm on the day it gets something wrong.

How do we score a process before automating it?

Score each candidate from 1 to 5 on four questions, then multiply the scores. Multiplying, rather than adding, means one very weak answer sinks the candidate, which is what you want for a first project.

• ·Frequency: how often does it happen? 5 is many times a day; 1 is a few times a year.
• ·Rule clarity: could you write the steps on one page? 5 is “always the same steps”; 1 is “it depends on who is asking”.
• ·Checkability: can a person tell quickly whether the result is right? 5 is a total that must match or a field that is filled or empty; 1 is a judgement nobody can verify until much later.
• ·Cost of error: how cheap is a mistake while you learn? 5 is “caught before anything leaves the building”; 1 is a wrong payment, a wrong GST return or a wrong price sent to a customer.
• ·Then sanity-check the winner: does it touch only two or three systems, and does someone own the process today who wants it fixed?

Which processes are good first candidates, and which are bad ones?

Good first candidates are high-volume, rule-based and checkable. Bad ones are rare, ambiguous or costly when wrong, even if they feel painful.

• ·Good: supplier bills received by email, read and entered into TallyPrime as drafts for an accountant to approve.
• ·Good: orders that arrive on WhatsApp or email, captured into the order sheet or ERP with the customer, items and quantities filled in.
• ·Good: the daily sales, collections or production report, assembled from TallyPrime and Excel and sent to the same people each morning.
• ·Good: payment reminders for overdue invoices, drafted from the outstanding list and sent once someone approves the list.
• ·Good: new enquiries from the website or IndiaMART-style portals logged in the CRM and assigned, so none sit unanswered.
• ·Bad first choice: anything that pays money out automatically, files a GST return or changes prices.
• ·Bad first choice: year-end or month-end closing tasks, which happen too rarely to learn from quickly.
• ·Bad first choice: a process nobody can describe the same way twice. Agree the process on paper first.

Is it worth automating if we only have a few hundred invoices a month?

Often, yes, but do the sum with your own numbers rather than anyone's averages. Count how many items arrive in a typical month, time how long one takes by hand (including fixing mistakes and chasing missing details), and multiply. Then ask what else that person could be doing, and what a late or wrong entry costs you today in delayed dispatch, missed credit or a reconciliation at month end.

A few hundred items a month spread across a working month means the task comes up many times a day, which is enough repetition for an automation to settle in. If the honest answer is a couple of hours a month, leave it and pick something more frequent.

Do we need AI, or are simple rules enough?

Use plain rules wherever the information is already structured, and AI only where something has to be read or judged. Moving a confirmed order from a form into a sheet, copying a total into TallyPrime or sending a report at 9 am needs no AI, and rules are cheaper, faster and fully predictable.

AI earns its place in two situations. The first is reading unstructured documents: a supplier bill as a PDF or a phone photo, an order typed loosely in a WhatsApp message, an email with the details buried in the text. The second is a narrow judgement that rules can't capture well, such as which product a customer most likely means, or whether an email is an order, a complaint or a query. In both cases the AI's output should be checked by a rule (does the total match, does the GSTIN exist in your masters) or by a person before it changes anything.

Can processes in TallyPrime be automated?

Yes. TallyPrime can accept data from other software through its XML interface over HTTP, and from Release 7.0 (December 2025) it also supports JSON natively, so vouchers and masters can be created by an automation instead of typed. It can be customised with TDL, Tally's own development language, for checks and actions inside TallyPrime itself.

The catch is where TallyPrime runs. Its interface is served by the TallyPrime application, so it must be running, with the company loaded, on a machine the automation can reach. Many firms run it on one office PC, which works if that PC stays on and the automation runs on the same network. Our guide to TallyPrime integration methods covers the options in detail.

Zapier or custom: what should a first automation run on?

Use a no-code tool such as Zapier, Make or Power Automate when every system involved is a cloud app with a ready-made connector and the rules are simple. Use custom code when one of the systems is TallyPrime on a local PC, when documents have to be read, or when you need proper logging, retries and approvals.

For the stack most Indian SMEs run, the deciding question is usually reach. A cloud-only tool needs a way to talk to TallyPrime on your office network. Automating WhatsApp properly means the WhatsApp Business Platform, whose Cloud API receives incoming messages through webhooks and only allows pre-approved template messages outside the customer service window; the ordinary WhatsApp app on a phone is not built for automation. Whichever route you choose, make sure you own the accounts, can see a log of every run, and can switch it off without breaking anything.

How do we keep a person in control?

Have the automation prepare and a person approve, at least until the numbers show it can be trusted. In practice that means bills arrive in TallyPrime as drafts or in a review list, not as posted vouchers; reminders are sent only after someone ticks the list; and anything the automation is unsure about is flagged rather than guessed.

• ·Name one owner who reviews the exceptions every day.
• ·Make every action reversible, or hold it for approval if it can't be undone.
• ·Log what came in, what the automation did and who approved it.
• ·Send an alert to that owner when a run fails, so it never fails silently.

How do we measure whether the automation is working?

Measure against your own numbers from before you started, not against a vendor's claim. Before switching it on, record a baseline for two to four weeks; then compare the same measures after. If you bring in an outside developer (Seamwire, a Bengaluru software studio, included), agree these measures with them before anything is built.

• ·Volume: items handled per week, and the share handled without a person retyping anything.
• ·Time: minutes per item, and the time from an item arriving to it being in TallyPrime or the ERP.
• ·Accuracy: corrections and rejections at the approval step, and reconciliation differences at month end.
• ·Exceptions: how many items the automation flagged, and why. A falling exception rate is the clearest sign it is learning your business.
• ·Then decide: if the numbers hold for a month or two, loosen the approvals on the easy cases and pick the next process on your score list.