Cover Banner

Data extraction using AI: Case Study: Preparing FAR

--x--

opening Slide

      Extraction of data using AI: Case Study: Preparation of Fixed Asset Register

--x--

Factory Slide

Imagine, your client has built a huge factory.

It shall comprise of common units such as—power rooms, utility complexes, connecting corridors

and

at the same time, they have made unique machines in each workshop.

Some work is done for common areas which is to be divided amongst CGUs whereas others are specifically allocated to certain components

--x—

Memo Slide

Imagine, your articles sitting with over 1000+ work completion memos

from over a dozen of vendors,

Trying to understand the technical terms of construction then

1. Recognising
2. Measuring
3. & Classifying them from capital work in progress into standalone Cash Generating Units as per accounting standards.

And at the same time following up the busy engineers for small queries & taunted for common sense.

--x--

DEMONSTRATION  

[To showcase –

1. One Abstract
2. Repetition of same with one more Abstract, Output in a tabulated format

--x—

Responsible CA

By the time GPT performs its work, lets understand the Prompt

- As you can see, I have chosen O4 high mini model, this is useful for higher accuracy.

- We work with very sensitive data, hence, please use temporary mode for chat so that we ensure data confidentiality.

- By defining the user, we are able put GPT in our shoes and think it will think from our point of view.

- write as minimum as possible and avoid multiple prompts otherwise GPT will get confused.

- We need to define our output, the format in which we want it, otherwise these talkative LLMs will eat your ears.

- Perform manual checks, like in this case, although it has successfully performed all the allocation. It has not considered capitalizing blocked ITC as I never mentioned anything about that in the prompt.

--x--

      Points to be considered Slide

1. Prerequisites
2. We need to use professionally made Memos or kept data
3. Giving a detailed prompt about the input & premise on which AI must perform its tasks.
4. Using proper model till MCP is made

--x—

AI as an Extractor

Understanding a single memo in such environment shall take over an hour so this journey of CWIP to FAR takes a quarter for completion.

AI figures out what’s a common area versus a unique asset

1. Measures and tags each cost to the right category
2. Sorts everything into neat component buckets
3. Posts the numbers straight into your capital accounts

What used to take an expert team, and an entire month now wraps up in a single afternoon—letting you spend your time analysing results instead of wrestling with data – provided the client gives the data on time.

In our practices, we generally don’t get data in a standardized format

as not every client is using ERP systems or is maintaining data.

In order to give quality reports, we try to find data from various sources, data is provided -in various formats & nearly 70-80% of our time goes in cleaning & extracting data.

Use AI for Extraction,

& professional Judgement for analysis & understanding the relevance of the extracted data.

--x--

End