Using AI to Prepare CMA Reports: A Practical Guide for CAs and Accountants

CMA preparation is one of those engagements where the value a CA adds - financial judgement, ratio analysis, banker relationship context - gets buried under hours of mechanical work. Transposing ITR figures into Excel templates, building five-year projections row by row, checking that the working capital cycle assumption is consistent with the balance sheet. AI does not replace the judgement. It absorbs the mechanical work, which frees you to spend time on the parts that actually determine whether a loan gets sanctioned.

This guide covers a specific workflow: using ChatGPT for drafting and narrative, Excel for the IBA CMA template, and AI-assisted ratio checking before the document reaches the bank.

The Part of CMA Work That AI Handles Well

The IBA CMA format has not changed structurally in years. The six core statements - existing credit limits, operating statement, analysis of balance sheet, comparative statement of current assets and liabilities, maximum permissible bank finance calculation, and funds flow - follow a fixed structure that every bank branch credit officer expects in a specific sequence.

This fixed structure is precisely what AI handles well. Given clean historical financials and a set of assumptions, ChatGPT can build a consistent first draft of all six statements in a fraction of the time it takes to do manually. What it cannot do is verify that the inputs are correct, assess whether the projection assumptions are commercially realistic for that specific industry and market, or make the judgement call on whether the DSCR is strong enough to survive the bank’s internal credit committee - those remain your work.

A practical division: use AI for structure, population, and internal consistency checks. Use your own judgement for assumptions, narrative framing, and the read on whether the proposal will pass scrutiny.

Setting Up the ChatGPT Workflow

Create a dedicated ChatGPT project or a custom GPT if you are on GPT-4o or above. This lets you store a standing system prompt that you reuse across client engagements. A working system prompt:

“You are assisting a practising Chartered Accountant in India with CMA report preparation for bank loan applications. All reports follow the IBA (Indian Banks’ Association) format. All figures are in Indian Rupees. When building projections, flag any assumption that looks inconsistent with the prior two years of actuals. Always calculate and display DSCR for each projected year. Identify any ratio that falls below standard bank thresholds and flag it explicitly.”

With this system prompt saved, each new engagement starts with pasting the client’s two years of actuals and the loan requirement. ChatGPT maintains the context across that session.

Building Projections That Survive Credit Scrutiny

The projections section is where most CMA reports are weak - and where AI assistance has the highest value if used correctly.

Start by prompting ChatGPT to build Year 1 projections before Years 2 and 3. Ask it to justify each line item assumption with reference to the actuals. For example: “Revenue for Year 1 should be explained by the growth rate in the last two years, adjusted for the capacity addition the loan will fund. Show the reasoning.”

This forces an internally consistent narrative rather than percentage growth applied uniformly. When the bank’s credit officer asks “why does revenue jump 35% in Year 2,” there is a documented answer tied to capacity, not an arbitrary multiplier.

Check three ratios before submitting to the bank. Current ratio should be above 1.33 for working capital loans - this is the Tandon Committee threshold that PSU bank credit officers still use as a default floor. DSCR should be above 1.25 for each projected year, not just the average across the tenure. Debt-equity ratio post-loan should be below 3:1 for most term loan assessments at PSU banks; HDFC and ICICI are more flexible but still expect below 4:1.

Use this ChatGPT prompt to run the check: “Given the projected financials above, calculate current ratio, DSCR, and debt-equity ratio for each of the three projected years. Flag any year where any ratio falls below the following thresholds: current ratio 1.33, DSCR 1.25, debt-equity 3.0. For each flag, suggest which input assumption should be adjusted to bring the ratio above the threshold.”

Handling the Narrative Sections Efficiently

The business profile, management background, and project description sections of a CMA are time-consuming to write from scratch for each client. They are also the section most often written generically - which credit officers notice.

A prompt that works well for generating a client-specific narrative quickly:

“Write a 250-word business profile for inclusion in a bank CMA report. Business type: [manufacturing / trading / services]. Operating since: [year]. Legal structure: [proprietorship / partnership / Pvt Ltd]. Industry: [sector]. Key customers: [type - institutional / retail / government]. Loan purpose: [term loan for equipment / working capital limit / both]. Tone: factual, professional, no promotional language. Include one sentence on the promoter’s experience in the industry.”

Edit the output to insert the promoter’s actual background and any specific market context that strengthens the credit case - a government supply contract, a Udyam registration tier, an established relationship with an anchor buyer. These details cannot come from AI; they come from your client intake process.

What AI Should Not Do in This Workflow

Do not use AI to generate the historical financial figures. The ITR, audited P&L, and balance sheet are source documents. Any discrepancy between the CMA and the client’s ITR - even a rounding difference on a sub-head - is flagged during bank scrutiny and reflects on the CA’s certification. AI should receive the historical numbers from you; it should never produce them.

Do not use AI to assess commercial viability of the business. Whether a Rs.50 lakh expansion into a new product line makes sense for a specific client in a specific market is a professional judgement. ChatGPT does not know the client’s competitive position, their local buyer relationships, or whether their sector is facing a regulatory headwind. The CMA can be technically clean and still represent a weak credit case - identifying that is your role.

Finally, all CMA reports require CA certification before submission to any bank for loans above Rs.10 lakh. The workflow above produces a faster, more consistent draft. The certification, and the professional responsibility that comes with it, remains with you.

For context on the common reasons banks reject business loan applications after receiving the CMA, read why business loans get rejected. For a reference on what additional documents banks require alongside the CMA, read business loan documents required.