# How to analyze costs and profitability per product with AI > Guide to knowing what makes you money: well-allocated costs, conversational analysis on real data and the decisions that follow. - Canonical: https://serchai.com/en/guides/ai-profitability-analysis/ - Site: Serchai (https://serchai.com) — AI tools comparator - Language: en - Updated: 2026-07-26 --- ## Tools you will use - [Holded](https://serchai.com/en/reviews/holded/) — Invoicing, accounting and SMB management in the cloud, with AI in the flow. - [Claude](https://serchai.com/en/reviews/claude/) — Anthropic's AI assistant for writing, analysing and thinking through documents. - [Gamma](https://serchai.com/en/reviews/gamma/) — Turns a text or a topic into presentations and documents that look good. ## The steps, in short 1. **Tag revenue and costs by business line** — Without tagged data there is no analysis: the per-product or per-service category is the base. 2. **Allocate costs with a simple, stable criterion** — Directs to their line and shared costs by a declared rule: simple beats perfect. 3. **Interrogate the data with the assistant** — Which line earns, who subsidizes whom and what happens if you touch prices: in plain language. 4. **Turn the analysis into two decisions** — Reprice or cut: the analysis that touches neither prices nor catalog was curiosity. > **TLDR:** Most SMBs know their total result and not which line generates it and which consumes it: profitability analysis attacks exactly that blindness. The system: revenue and costs tagged by line in Holded, shared costs allocated by a simple declared rule, the conversational interrogation of the data with the assistant and the close in decisions: reprice what loses or cut it. The usual surprise: the star product that makes no money. This guide is for anyone selling several things (products, services, lines) and pricing by intuition: the business that invoices well and earns so-so, without knowing where the difference goes. The answer is almost never "a little everywhere": there are usually lines subsidizing others, and seeing it changes the decisions. The usual limits: this analysis orients management. Official accounting allocation criteria, with your [accountant](https://serchai.com/en/guides/ai-accountant-collaboration/). ## 1. Tag revenue and costs by business line The analysis starts at the data, and the requirement is tagging: every income and every cost with its line (product, service, channel, whatever cuts your business meaningfully). In [Holded](https://serchai.com/en/reviews/holded/) that means categories and tags applied with discipline, from about $8 a month, with the [digitization](https://serchai.com/en/guides/ai-receipt-digitization/) flow delivering every document already classified. The right granularity is what you will sustain: three to six well-tagged lines inform more than twenty half-done. And history rules: three tagged months show the pattern, twelve show the seasonality. The cost that escapes most in services is your own and your team's time: without a notion of hours per line (even estimated weekly), the service "doing fine" may be the one devouring the hours you never bill. ## 2. Allocate costs with a simple, stable criterion Direct costs go to their line without debate (product A's materials, project B's hours). The debate is the shared ones (rent, administration, tools), and here the SMB's practical rule: a simple, declared, stable criterion beats the accounting perfection nobody maintains. The usual allocations: proportional to revenue (simplest), proportional to hours spent (fairest in services) or the block mix. Pick one, write it down and keep it, because changing the criterion every quarter turns comparisons into noise. The method's honesty: shared-cost allocation is a convention, and the fine conclusions ("this line earns 2%") are worth less than the thick ones ("this line loses money whichever way you look"). Decisions get made on the thick ones. ## 3. Interrogate the data with the assistant With data tagged and the criterion applied, analysis becomes conversation: [Claude](https://serchai.com/en/reviews/claude/) (from $17 a month, with a permanent free tier) takes the per-line extract and answers the questions that matter: what is each line's real margin? Who subsidizes whom? What happens to the total if line C raises prices 10% and loses 15% of clients? Price simulations are where the assistant pays most: playing the increase scenarios (with explicit client-loss assumptions) gives the decision map intuition never does. The usual rules: calculations verified before deciding, aggregated data without names, and your company's data policy choosing the tool. The uncomfortable question this analysis tends to uncover: the star product (the best seller) with the tiny margin, and the quiet line holding up the house. ## 4. Turn the analysis into two decisions The analysis that ends in no decision was expensive curiosity. The two decision families coming out of here: reprice (the line that loses or barely earns raises prices, with step 3's simulation as the risk map) or cut (the line losing without remedy or strategic function gets discontinued, and the freed hours go to the winners). The conscious third way exists: keeping a losing line for strategy (client entry door, position defense), and the key word is conscious: it stays with the figure in view, not by inertia. The summary for partners or team fits the monthly page of the [financial reports](https://serchai.com/en/guides/ai-financial-reports/), with [Gamma](https://serchai.com/en/reviews/gamma/) formatting it: the lines, their margins and the quarter's decision. The whole sector lives in [AI for accounting and finance](https://serchai.com/en/ai-for/accounting-finance/). ## Common mistakes Analyzing without tagging first. Analysis on totals without lines is reading tea leaves: disciplined tagging is the investment that makes it possible. Perfecting the shared-cost allocation forever. The simple stable criterion that gets maintained informs more than the sophisticated one that gets abandoned. Decisions come from the thick differences. Forgetting the cost of your own time. In services it is the biggest cost and the least counted: the "profitable" line consuming half your hours may not be. Knowing and not deciding. The analysis that finds the losing line and touches neither prices nor catalog only added an informed frustration. ## Frequently asked questions ### How often should the analysis repeat? Quarterly is the useful rhythm: enough for price and catalog changes to show effect, without turning it into obsession. Continuous tagging is what makes repetition cheap. ### Can I trust the price simulations? As scenarios with explicit assumptions, yes: the value is comparing decisions, not predicting the future. Your clients' real elasticity gets discovered by raising, which is why prudent increases go in steps. ### What about the line that loses but "builds image"? Put a number on the argument: what keeping it costs yearly and what evidence supports its attributed contribution. Strategy with the figure in view is strategy. Without it, inertia with a story. ### Can my accountant do this analysis? Firms with controlling services offer it, and this guide leaves you able to hold that conversation with your own data: the best scenario is your disciplined tagging plus their professional criterion.