Glossary

Software that can read, write, listen, look and decide in ways that used to need a person.

AI is a general name for software that handles tasks which once needed human judgement. It can read a message, understand what is being asked, and act on it. In a business it usually means messages, documents and numbers handled by software, with a person still in charge.

For example: A distributor in Coimbatore gets orders on WhatsApp all day. AI reads each message, works out the items and quantities, and enters them into Tally for a person to approve.

An AI that is given a goal and works through the steps itself, checking with a person when it should.

An AI assistant answers when asked. An AI agent goes further: it is given a task, works out the steps, uses your systems to do them, and reports back. It suits jobs with several steps, and it should always have clear limits on what it may do alone.

For example: A machinery dealer in Madurai sends a quotation. An agent waits three days, checks whether the customer replied, sends a polite reminder on WhatsApp, and flags the deal to the sales head if there is still no answer.

Software that talks with customers or staff in plain language and helps them get things done.

An AI assistant reads a question, understands it, and replies the way a well-trained colleague would. It can answer questions, take orders, book appointments and look things up in your systems. A good one knows when to stop and hand the conversation to a person.

For example: A clinic chain in Chennai has an assistant on WhatsApp. It answers questions about timings and fees, books appointments, and passes anything about a medical problem straight to the front desk.

The door through which one piece of software talks to another, without a person retyping anything.

Most software you use has a way for other software to send it data and ask it questions. That doorway is called an API. It is what lets an order from WhatsApp land in Tally, or a new customer in Zoho appear in your billing software, without anyone typing it in again.

For example: A garments exporter in Tiruppur wants shipment updates in one place. Through the courier's API, every tracking change is pulled into a Google Sheet, so the export desk stops checking three websites a day.

Repetitive work done by software on its own, the same way every time, without anyone typing it.

Automation takes a task your team repeats every day and has software do it instead. The task follows a fixed set of steps, so it runs the same way every time and does not get tired at 6 pm. AI adds the ability to handle messy inputs, like a photo of an invoice or a WhatsApp message in mixed Tamil and English.

For example: Picture a clinic chain in Chennai that types every invoice twice, once in the billing software and once in Tally. Typed once and copied across by software, the weekly mismatches stop.

A spare copy of your data kept somewhere safe, so a crash, a mistake or a theft does not wipe you out.

A backup is a copy of your data made regularly and stored away from the original. If a laptop dies, a file is deleted by mistake, or a system is attacked, you restore from the copy and carry on. A backup that has never been tested is only a hope, so we restore from it now and then to prove it works.

For example: A pipes and fittings supplier in Salem keeps all its price lists and quotations in one system. A copy is made every night and kept in a second location. A restore test once a quarter proves the copy is good.

A program that replies to messages, from a simple menu bot to a full AI assistant.

Chatbot is the older and wider word for any program that chats. The simple kind offers a menu of buttons and ready-made replies, and gets stuck the moment a customer types something unexpected. The newer kind is an AI assistant that understands plain language and can actually help, and that is usually what people mean today.

For example: Picture a distributor in Coimbatore whose customers gave up on a menu-style chatbot. An assistant that reads a message like 'send 20 bags of the usual' would bring orders back to WhatsApp.

Computers you rent by the month in someone else's data centre, instead of buying your own. One of those computers is a server.

The cloud is simply someone else's computers, rented over the internet. Each computer you rent is called a server: it runs your system and holds your records, and the organised store of those records (customers, items, invoices) is its database. You pay for what you use, someone else handles the power, cooling and repairs, and your team reaches it from anywhere. For most businesses it is cheaper and safer than a server in the office cupboard, as long as you know which country the data sits in.

For example: A garments exporter in Tiruppur keeps its order-tracking system in a data centre in India. Staff in the factory, the office and the buyer's showroom all see the same numbers, and nobody maintains a server.

One screen that shows the numbers that matter to you today, updated on its own.

A dashboard is a page that pulls live numbers from your systems and shows them in a few charts and totals. Instead of waiting for a report, you open it and see sales, stock, cash and pending work as they stand right now. The best ones show five numbers you act on, not fifty you scroll past.

For example: A machinery dealer in Madurai opens one page each morning: quotations sent, quotations waiting for a follow-up, and orders due this week. It is built from Zoho and Excel and refreshes itself.

The route your data takes from where it is created to where it is used, cleaned up along the way.

A data pipeline is a set of steps that collects data from your systems, tidies it, and delivers it where it is needed. It might pull sales from Tally, stock from Excel and orders from WhatsApp, fix the spellings and codes, and put the result in one place every night. Once it is built, it runs without anyone doing the copying.

For example: Take a distributor in Coimbatore with sales in Tally and stock in three Excel files. A nightly pipeline joins them, so the Monday stock report is ready before anyone reaches the office.

Which country your data physically sits in, and therefore whose laws apply to it.

Data residency is about where your data is stored. A computer in Mumbai means Indian law applies and India's rules on personal data are easier to follow. Some customers, banks and government buyers ask for this in writing, so it is worth knowing the answer before they ask.

For example: A clinic chain in Chennai stores patient records. It chooses to keep all of them on computers in India, and its contracts say so, which keeps it on the right side of India's data protection rules.

Scrambling data so only someone with the right key can read it, even if they steal the file.

Encryption turns readable data into scrambled text that is useless without a key. It protects data while it travels over the internet and while it sits on a disk. If a laptop is stolen or a file is copied, the thief gets gibberish.

For example: A garments exporter in Tiruppur shares buyer price sheets with its agent abroad. The files are encrypted before they leave the office, so a lost phone does not mean a leaked price list.

The limits placed on an AI so it stays on topic, stays polite, and never does what it should not.

Guardrails are the rules an AI must follow no matter what it is asked. They decide what it may talk about, what it must never say, what it may do in your systems, and when it must hand over to a person. Good guardrails are tested by trying to break them before real customers ever see the system.

For example: A clinic chain in Chennai has a WhatsApp assistant that may book appointments and quote fees. It may not give medical advice, and if a patient describes symptoms it replies with a phone number and alerts the front desk.

When an AI gives a confident answer that is simply wrong or made up.

AI that writes text is very good at sounding sure. Sometimes it fills a gap with something that sounds right but is not true: a price that does not exist, a policy you never wrote. The fix is to make it answer only from your own documents, show where each answer came from, and keep a person in the loop for anything that matters.

For example: Picture a pipes and fittings supplier in Salem asking a general AI tool for a product price. It may get a confident number that is wrong. A knowledge assistant built on the supplier's own price list quotes the real price and shows which list it came from.

Connecting two systems so that data typed into one shows up in the other on its own.

Integration means making the software you already use talk to each other. An order entered once appears in billing, stock and the delivery sheet without retyping. It removes the double entry that causes most of the small, expensive mistakes in a business.

For example: Take a distributor in Coimbatore whose WhatsApp orders connect to Tally. Each order becomes a sales voucher as it arrives, and nobody stays back for the evening typing session.

An AI that answers staff questions from your own documents, like a very well-read colleague.

A knowledge assistant reads your price lists, policies, manuals and past quotations, and answers questions from them in plain language. It shows which document the answer came from, so anyone can check. It does not make things up from the wider internet, because it is limited to what you gave it.

For example: Picture a pipes and fittings supplier in Salem that builds quotations by hand from three price lists. With a knowledge assistant, a staff member types 'price for 50 metres of 2 inch PVC, trade rate' and gets the answer with the page it came from.

The kind of AI that reads and writes text, and sits behind most AI assistants.

A large language model is software that has read an enormous amount of text and learned how language works. Give it a question, a message or a document, and it can summarise, translate, answer or draft a reply. On its own it knows nothing about your business; it becomes useful when it is connected to your documents and systems.

For example: A garments exporter in Tiruppur gets buyer emails in English and gives instructions to the floor in Tamil. A language model drafts the Tamil version in seconds and a supervisor checks it before it goes out.

Software that learns patterns from past data instead of following rules someone typed in.

In machine learning, you show software many past examples and it works out the pattern on its own. Show it three years of sales and it learns which items sell in which month. It is the method behind forecasts, fraud checks and most of what people call AI today.

For example: A distributor in Coimbatore gives three years of sales data to a forecasting system. It learns the festival peaks and predicts which items will run short next month, so stock is ordered in time.

The trained piece of AI that does the actual thinking, rented by usage like electricity.

A model is the result of training: a piece of software that has learned to do one kind of task, such as reading text, recognising invoices or predicting sales. You usually rent a model from a provider and pay for what you use. Different models suit different jobs, so a good build often uses more than one and can swap them when a better one appears.

For example: A clinic chain in Chennai uses one model to read scanned lab reports and a different, cheaper one to answer routine WhatsApp questions. The monthly rental is a known number on the bill, and either model can be swapped later without rebuilding.

Software and data kept on computers inside your own building, rather than rented in the cloud.

On-premise means the computers running your systems sit in your office or factory, and you own and maintain them. It gives you full physical control, which some businesses need for legal or contract reasons. It also means you handle power, repairs, backups and security yourself, so it costs more than it looks.

For example: A garments exporter in Tiruppur has a buyer who insists that pattern files never leave the factory. Those files stay on a computer in the office, while the ordinary order tracking runs in the cloud.

The instructions given to an AI, from a one-line question to a full page of rules for how it behaves.

A prompt is what you tell the AI before it answers. It can be a question typed by a customer, or a long set of standing instructions we write: what the business does, how to speak, what never to say, and what to do when unsure. Most of the quality of an AI assistant comes from getting these instructions right and testing them.

For example: A machinery dealer in Madurai has an assistant whose standing instructions say: reply in the customer's language, quote only from the current price list, and pass anything about credit terms to the owner.

Turning a photo or scan of a paper document into text and numbers a computer can use.

OCR stands for optical character recognition. It reads the letters and numbers in a photo or a scanned page and turns them into text you can search, copy and enter into a system. Modern versions handle phone photos, faint print and Tamil as well as English, and AI helps make sense of what it reads, such as which number is the total.

For example: A pipes and fittings supplier in Salem receives delivery challans as WhatsApp photos. Each photo is read, the items and quantities are pulled out, and a purchase entry is drafted in Tally for a person to confirm.

The small piece of text, roughly three-quarters of a word, that AI usage is counted and billed in.

AI models do not read words; they read tokens, which are short chunks of text. In English, 1,000 tokens is about 750 words, and Tamil text uses more tokens per word than English. Model providers charge per token, so the length of your documents and messages is what drives the monthly AI bill.

For example: A knowledge assistant for a clinic chain in Chennai answers 300 questions a day. Each question and answer uses a few hundred tokens, which adds up to a monthly bill of a few thousand rupees.

Teaching software a pattern from many examples, so it can handle new cases it has not seen.

Training is how a model learns: it is shown thousands or millions of examples and adjusts itself until it gets them right. Most businesses never train a model from scratch; they rent one that is already trained and give it their own documents and instructions. Training your team to use the new system is a different thing, and just as important.

For example: A distributor in Coimbatore does not need a model trained on its own data. It rents a ready model, gives it the current price list and a page of instructions, and spends two afternoons training the order desk to use it.

The set of steps a piece of work goes through, from the moment it starts to the moment it is done.

A workflow is the path a task takes through your business: an order arrives, someone checks stock, someone raises an invoice, someone arranges delivery. Writing the steps down is the first thing we do, because you cannot automate what you cannot describe. We then fix one workflow at a time, measure it, and move to the next.

For example: For a machinery dealer in Madurai, the quotation workflow might be: enquiry on WhatsApp, price looked up, quotation typed, sent, followed up after three days. Automate the last two steps and the follow-ups stop being forgotten.