Ask your books: why did margin fall?
An owner types one question. A good answer is not a paragraph of opinion. It is a bridge from last month's margin to this month's, where every step is a number that opens the entries behind it. Here is that answer, worked by hand.
The question
The owner of a small pump manufacturer opens the October accounts. Gross margin is 42.2 percent. In September it was 44.9 percent. Sales are lower too. The owner types: Why did margin fall in October?
A weak answer says: costs went up and the mix changed. True of every business in every month, and useless. A strong answer says how much each cause moved the margin, in percentage points that add up exactly from 44.9 to 42.2, and lets the owner click into each one. That second answer is a margin bridge, and it is what an AI Controller should produce.
Copper went up on the 2 pumps. That cost 3.1 points. Price rises recovered 0.6, and selling more motors and fewer pumps cost 0.2. 44.9 less 3.1, plus 0.6, less 0.2 is 42.2.
What the books hold
The business sells 3 products. Illustrative figures, in rupees per unit.
| Product | Units Sep | Units Oct | Price Sep | Price Oct | Cost Sep | Cost Oct |
|---|---|---|---|---|---|---|
| Pump 1 HP | 125 | 110 | 18,500 | 18,700 | 9,900 | 10,750 |
| Pump 2 HP | 95 | 85 | 26,800 | 27,200 | 14,600 | 15,550 |
| Motor 0.5 HP | 160 | 175 | 7,200 | 7,240 | 4,300 | 4,300 |
The pumps' unit cost rose by 850 and 950 because the copper winding cost more in October. The motor's cost did not move.
| September | October | |
|---|---|---|
| Sales | 60,10,500 | 56,36,000 |
| Cost of sales | 33,12,500 | 32,56,750 |
| Gross margin | 26,98,000 | 23,79,250 |
| Gross margin percent | 44.9% | 42.2% |
The method: change one thing at a time
The bridge works by starting from September and changing one driver at a time, in a fixed order, recomputing the margin percent after each change. The difference after each step is that driver's effect. Because each step starts where the last one ended, the effects add up exactly to the total change, with no line called other.
October's units, at September's prices and September's costs. Only the quantities have changed, so any movement in margin percent is the mix: selling relatively more of a lower margin product.
October's units at October's prices, still at September's costs. The movement is the price effect.
October's units, October's prices, October's costs. This is the actual October margin. The movement is the cost effect, here entirely copper.
The steps, worked
| Step | Sales | Cost of sales | Margin | Margin % | Effect |
|---|---|---|---|---|---|
| September | 60,10,500 | 33,12,500 | 26,98,000 | 44.9% | |
| 1. October units | 55,73,000 | 30,82,500 | 24,90,500 | 44.7% | Mix: 0.2 points down |
| 2. October prices | 56,36,000 | 30,82,500 | 25,53,500 | 45.3% | Price: 0.6 points up |
| 3. October costs | 56,36,000 | 32,56,750 | 23,79,250 | 42.2% | Copper: 3.1 points down |
44.9 less 0.2, plus 0.6, less 3.1 is 42.2. Percentages are rounded to 1 decimal; the unrounded effects add up exactly.
In rupees the story is just as plain. On October's volumes, the copper increase added 1,74,250 to cost of sales: 110 pumps at 850 and 85 pumps at 950. The price rises added 63,000 to sales. The owner now knows that the price increase recovered barely a third of the copper, and that the next decision is a price, not a cost cut.
What the AI does, and what it must not do
This is where the design of an AI native ERP matters. The arithmetic above must be exact, to the rupee, every time. Language models are not built for exact arithmetic. So a sound Controller splits the job in 2:
- The AI reads the question. It understands that why did margin fall in October means a gross margin bridge, October against September, for the whole company.
- The accounting engine does the numbers. It reads the posted sales invoices and the cost of each unit sold, runs the 3 steps and returns figures that tie to the ledger.
- The answer shows its working. Each step opens the invoices and cost entries behind it. The owner can click Copper and see the 2 products, the purchase bills that carried the new copper price, and the dates.
The AI never computes a figure in the books. It reads the question and words the answer. Every number comes from the engine, and every number opens the entries it came from.
A note on the order of the steps
Change the order of the steps, cost before price for example, and the individual effects shift a little, though the total does not. That is a property of every bridge, not a flaw in this one. What matters is that the order is fixed, stated, and the same every month, so that October's bridge can be compared with November's. A good Controller says which order it used.
The questions that follow
A bridge always leads to the next question, and a Controller should be ready for each:
- How much should we raise the pump price to recover the copper? The engine works out the price that restores 44.9 percent on October's mix.
- Which customers bought the pumps at the old price? The invoices are already in the books.
- Is copper still rising? The latest purchase bills say what the business paid, supplier by supplier.
That is the promise of an AI native ERP in one example: the owner asks in plain words, and the answer is accounting, not opinion.
Beyond the dashboard
Most companies get this answer one of 2 ways. A controller downloads 5 or 6 reports from the ERP into Excel and builds the bridge by hand, which takes an afternoon. Or the company has paid for a set of Power BI or Tableau dashboards, which refresh automatically, but answer only the questions they were designed for. The moment the owner asks something new, such as which customers bought the pumps at the old price, the dashboard has nothing to say.
Controller, the AI inside Niyantrak ERP by Niyantrak Consulting LLP, works the other way: it reads the new question and works out the answer from the live books, with no report designed in advance. Every company deserves a Controller. Niyantrak ERP puts one inside the books, and its cloud edition is coming soon.
Questions and answers
How do you explain why gross margin fell?
Build a margin bridge: start from last period's margin percent and change one driver at a time, in a fixed order (mix, then price, then cost), recomputing the margin after each step. The steps add up exactly to the total change.
What is the difference between a mix effect and a price effect?
The mix effect is the change in margin from selling a different proportion of products at the old prices and costs. The price effect is the change from selling at new prices, before any change in cost.
Can an AI answer why margin fell from the books?
Yes, if it is built correctly. In Niyantrak ERP, Controller reads the question and the accounting engine computes the bridge from the posted entries, and every figure opens the invoices and costs behind it.
Why not use a Power BI dashboard for this?
A dashboard answers only the questions it was built for. A new question needs a new report. An AI Controller inside the ERP answers a new question from the live books without a report being designed first.
Would you like your own month explained like this?
Controller, the AI inside the AI native ERP built by Niyantrak Consulting LLP, answers this kind of question from the books themselves. Until it reaches the cloud, we are happy to walk through your own margin bridge with you.