The Journey of a Biscuit

A note to you, Jatin Sir

Thank you for opening your plant to our team on 26th October. We walked the full line with you: the maida and sugar windows, the three dosing floors, the oven, the cooling belt, the packing lanes and the dispatch conveyor at the back. It is a well-run plant. The biscuit comes out the same every time and the lanes keep moving.

But a few things stayed with us after we left.

First, the number of hands still needed to keep the line going. People standing at two or three points along the cooling belt, scanning for bad biscuits, and still some defective biscuits slip through to packing. People at the lanes turning and straightening biscuits so that the packing machines do not jam. People checking packs. People counting bags and cartons.

Second, the oven. It is brought up to heat every day and burns fuel at settings fixed long ago.

Third, the clock. Your plant runs one shift, from 7 am to 7 pm. For the other twelve hours, a plant you have already paid for stands still.

You told us your two biggest worries are labour cost and fuel. So this proposal is built around a simple idea: let AI take over the repetitive checks, and use the same people to run the plant round the clock in three shifts. Same people, same machines, twice the biscuits. Labour cost and fuel per tonne both come down.

We have written it as a story, the story of one biscuit, so that anyone in your team can read it and see where AI fits into your plant.

Warm regards,
Aniruddha Mukherjee
Technical Program Manager, ThirdEye Data

The proposal at a glance

We propose 12 AI use cases that work together as one end-to-end system, from the intake window to the truck. We have ranked them by value and by how easy they are to put in, and grouped them into three phases over about nine months.

Phase 1 is three quick wins: easy to install, low cost, and visible to you within weeks. Once you have seen them work on your own dashboard, we move to the next phase with your confidence and budget.

At full run-rate the plan brings about ₹1.46 Cr a year in labour cost avoided, fuel, giveaway and less recycling, plus 4,500 tonnes of extra output every year, with fewer biscuits going round the recycling loop. It is designed to pay back within 12 to 18 months.

The journey of a biscuit, today

Let us follow one biscuit through your plant. We will call it Biscuit No. 1. Its journey has eight stops, and today it can only make that journey between 7 am and 7 pm.

1. 1
   Raw material intake

   Maida and sugar bags come in through the intake windows. Someone counts the bags by hand and writes them in a register.

2. 2
   Dosing and mixing on three floors

   On the 3rd floor, the sugar machine takes about 75 kg per batch and drops it down. On the 1st floor, the maida machine takes about 270 kg. Both meet in the 2nd-floor blender with the other ingredients and become dough.

3. 3
   Moulding

   The dough goes to the biscuit die, which cuts it into the familiar shape. Biscuit No. 1 is born.

4. 4
   Baking

   Biscuit No. 1 rides into the oven on a running belt and bakes at about 360 °C for about 3.5 minutes. With one shift a day, the oven has to be brought up to heat every morning, burning fuel before the first biscuit is baked.

5. 5
   Cooling, sorting and recycling

   For about 3.5 minutes it cools at room temperature. 6 workers stand at two or three points along the belt, scanning thousands of moving biscuits by eye and picking out the broken, burnt, under-baked and damaged ones. Even after two or three rounds of checking, some defective biscuits are missed and travel on to the packing zone. The biscuits that are picked out are collected and sent for recycling.

6. 6
   Lanes and packing

   The biscuits are lined up into 20–25 lanes. Here 8 more workers turn, flip and straighten biscuits so that they enter the packing machines the right way and do not jam. About 5 packing units pack them into ₹5, ₹10, ₹15 and ₹30 packs, and 4 checkers look at seals, print, codes and weight.

7. 7
   Cartons and storage

   Packs go into boxes and cartons, which are stacked in the store by lot number. The lot tally is kept by hand.

8. 8
   Dispatch

   At the back of the building, the dispatcher puts cartons on the conveyor and they drop into the truck. Cartons are counted by hand against the invoice.

Where the money goes

Your plant loses money in three places: the hands on repetitive work, the hours when the plant stands still, and the fire in the oven.

Leak 1: Labour on repetitive work

This work is tiring. A worker looking at a fast belt for twelve hours starts missing defects after the first two, which is why the plant needs checks at two or three points and still sees defects reach packing; quality then depends on who is on duty that day. Wages go up every year, people leave, and every new person takes weeks to learn the job by watching others.

Assumes ₹16,000 per person per month fully loaded (₹1.92 lakh a year) and about 4,500 tonnes a year from one 12-hour shift, so roughly ₹896 of labour on these jobs in every tonne.

Leak 2: A plant that works only half the day

The building, the oven, the mixers and the packing machines are paid for 24 hours a day, but they work for 12. The simple way to grow is to run more shifts. But doing that the old way means finding, hiring and training 42 more people for the same repetitive jobs. That is why most plants never do it.

Leak 3: Fuel in the oven

The oven is the biggest energy user in the plant. With one shift, it is brought up to temperature every morning, so a part of every day's fuel is spent before a single biscuit is baked. A little extra heat is kept as a safety margin against under-baking. Heat escapes through walls, doors and the chimney. The air-to-fuel ratio drifts over time. And nobody sees fuel per tonne on a live screen; it shows up only in the monthly bill.

If your oven fuel bill is around ₹10 lakh a month today, then at three-shift volume every 1% saved is ₹2.4 lakh a year.

Leak 4: The recycling loop

Broken, under-baked and damaged biscuits are not thrown away; they are sent for recycling. That saves the material, but the loop is not free. Every recycled kilo has already used oven fuel, labour and line time once, and it has to be collected, processed and baked a second time. The more biscuits go round the loop, the less good output the same oven and the same people can make.

The hidden leaks

• Defects that slip into packs. A broken or under-baked biscuit that gets past the manual checks ends up in a pack, which means complaints, returns and questions from the brand owner.
• Pack giveaway. When packs are filled a little above the declared weight to be safe, the extra biscuit goes free to the market, pack after pack.
• Count gaps. Without automatic counting at intake and dispatch, short supply and short shipments are hard to prove.
• Training takes too long. New people learn by watching seniors for weeks, and the plant depends on a few experienced hands.

Enter AI, one step at a time

We do not change your recipe, your oven or your packing machines. We add eyes (cameras and sensors), a brain (AI that runs right on your shop floor) and simple light signals that show your people exactly which biscuit to pick or fix. The machines take over the searching, and your people move to skilled work across three shifts.

How we have prioritised

We ranked every use case on two simple questions: how much value does it bring to your plant, and how easy is it to put in? We start with the quick wins at the top right, prove the savings to you, and then move to the bigger steps.

A day at your plant, after full implementation

It is a Monday in 2027. The plant now runs round the clock in three 8-hour shifts, with the same 21 people on these jobs. Here is how the day goes.

Before and after

What it means in rupees

Labour cost avoided is money you will not have to spend: running three shifts the old way would need 63 people on these jobs. With AI, the same 21 people do it (7 per shift), so 42 hires are never needed. The conservative case allows for 8 people per shift, for example a third picker or a counting helper.

Even in the conservative case, labour and fuel alone bring about ₹94 lakh a year, and the plant makes twice as many biscuits with the same people. These benefits repeat every year and grow as wages and fuel prices go up.

How we build it, end to end

One partner, one responsibility. ThirdEye Data designs, supplies, installs, integrates and supports the full system (cameras, sensors, edge AI boxes, highlight and pointer lights through our hardware partners, and all the software), so you deal with one team for everything.

• Runs on your shop floor. AI decisions are made on rugged edge boxes next to the line, in milliseconds, and keep working even if the internet goes down.
• Works with what you have. We connect to your oven panel, dosing machines, packing units, checkweighers and ERP or Tally. No change to your recipe or main machines.
• Machine data, switched on. In the baseline study we check every machine for built-in sensors and data ports, enable what is already there, and add low-cost sensors only where needed.
• Proof built in. Every reject, every bad pack and every truck count is saved with a photo or video clip.

The roadmap

What about the people?

Nobody needs to lose their job. The people freed by AI are exactly the people who will run the new shifts. Our plan is simple:

• Train the best line workers as AI line operators. They work the highlighter pick zone and the lanes, handle exceptions, and keep the cameras and lights clean and calibrated.
• Use the GenAI Training Buddy for every move. Each person learns a new station in their own language, at their own pace, and is signed off before working it alone.
• Shorter, better shifts. Three 8-hour shifts instead of one 12-hour shift: people stay fresher and more alert.
• Better and safer work. Nobody has to stand for hours searching a fast belt by eye; the AI does the searching.

Why ThirdEye Data

• AI and computer vision specialists, with teams in India and the US.
• We have built this kind of solution before: AI counting of stacked bags in warehouses, counting auto parts loaded into trucks, and tile counting with breakage detection on ceramic lines.
• GenAI in production. We build document and knowledge assistants, the same technology behind Ask-AI and the Training Buddy.
• End-to-end delivery. From site survey and hardware to software, integration, training and support, under one contract.

Commercial approach: start small, grow with confidence

We suggest starting with Phase 1 as a fixed-scope first step, with success measures agreed upfront. The next phases are released only after you have seen Phase 1 results on your own dashboard. Detailed commercials will follow the baseline study, in the model that suits you best:

• Fixed price per phase, with payment linked to the go-live of each phase.
• Monthly subscription, with hardware, software and support included, so there is little upfront spend.
• Savings-linked, where part of our fee is tied to the savings measured on the dashboard against the agreed baseline.

Let us take the first step together

1. 1
   Review meeting

   A one-hour meeting with you and your plant head to walk through this proposal and answer questions.

2. 2
   Baseline and sensor audit (2 weeks)

   Our team on your floor, measuring manpower, output, fuel, rejects and pack weights, and checking machine sensors. You get a baseline report either way.

3. 3
   Phase 1 kick-off

   Automatic Counting, the AI Defect Highlighter and the Executive Dashboard go live within about three months.

What we will need from your side

• One point of contact from the plant, and access for our team during the baseline study
• Last 6 months of fuel bills and daily production figures
• Manpower list by station
• List of machines with any built-in sensors, PLC or data ports
• Existing SOPs or training material, for the GenAI Training Buddy
• Power, network and mounting points at the camera locations

Biscuit No. 1 will taste exactly the same. It will just be made round the clock, by the same people, with less fuel in every tonne.

Assumptions used in this proposal

All values below are working assumptions. They will be replaced by measured values in the baseline study.