TAM/SAM/SOM: how AI analysis is wrong
Fincraft Capital Cookie Notice
COOKIE SETTINGS
The information a cookie contains is generated by the server and can be used by that server whenever a user visits the site which hosts the online services.
Technologies similar to Cookies
Similar technologies are technics or tools, using certain characteristics to identify devices so that visits to a website can be analyzed. It includes, for example, HTML5 local storage, local Shared objects, fingerprinting techniques, tracking pixels and plugins.
For ease of reference, cookies and similar technologies are referred as "Cookies".
Usage of Cookies
Cookies enable us to provide users with certain features and supply us with information about website visits.
We also make use of cookies to make our site more user-friendly, to better understand our existing and prospective customers, including associating a user with different devices and browsers that they may use for online targeting and advertising purposes.
Technologies similar to Cookies
Similar technologies are technics or tools, using certain characteristics to identify devices so that visits to a website can be analyzed. It includes, for example, HTML5 local storage, local Shared objects, fingerprinting techniques, tracking pixels and plugins.
For ease of reference, cookies and similar technologies are referred as "Cookies".
Usage of Cookies
Cookies enable us to provide users with certain features and supply us with information about website visits.
We also make use of cookies to make our site more user-friendly, to better understand our existing and prospective customers, including associating a user with different devices and browsers that they may use for online targeting and advertising purposes.
With the advent of AI, it began to seem as though lengthy and painstaking research could now be delegated entirely to a machine. And in just 15–20 minutes, one could obtain a comprehensive document containing a financial model, a competitor analysis and a five-year forecast. It all looks convincing: the structure, the figures, the charts.
The problem is that plausibility does not necessarily reflect reality, as the experts at Fincraft Capital s.r.o. point out. AI confidently produces an answer, even when analysing data that is incorrect or, at times, irrelevant. And whoever reads the report concludes that they have a good understanding of the situation and the market. Yet this feeling most often turns out to be an illusion.
In this article, experts from Fincraft Capital explain the TAM/SAM/SOM market valuation methodology and, using this as an example, highlight the ‘pitfalls’ of analysing data using AI.
The Triad of Concepts: TAM, SAM and SOM
The TAM/SAM/SOM methodology is a kind of filtering system that analyses the market step by step, progressively narrowing down market shares: from the total market to the addressable segment, and then to the portion you can realistically claim. TAM, SAM and SOM represent three levels of the same question: how much can be earned here?
There are two approaches to calculating these metrics.
‘Top-down’ - starting from the global market, then breaking it down by region, category and segment. Experts at Fincraft Capital note that this estimate is almost always overestimated, as it is based on averaged data and assumptions. ‘Bottom-up’ - starting with a specific customer: customer acquisition cost, average spend, number of orders, and processing capacity. The result is closer to reality. It is precisely the difference between these approaches that most often distinguishes fantasy from a workable scenario.
How to conduct research using AI: a step-by-step guide
AI can indeed be a useful tool — if used correctly. The more specific the input, the more useful the result, according to managers at Fincraft Capital Czech Republic. At the same time, it is important to set the rules from the outset: insist on sources, references and transparent calculation methods.
Step one - establish the context: production or sales, in which region, and using which business model (in-house production, dropshipping, order fulfilment, D2C). Specify the sales channels: your own website, marketplaces, social media. It is important that every detail narrows the analysis down from the abstract notion of ‘a large market’ to the specific ‘niche and its parameters’.
Step two - define the methodology and set the objectives: calculating market size, analysing competitors, evaluating sales channels, and developing a basic financial model.
Step three - the initial results, their assessment and any issues. As a first approximation, the AI generates a document covering the funnel (TAM – SAM – SOM), competitors, channels and the financial model. The first iteration will almost always be based on international reports — Statista, global surveys and English-language sources. For the local market, this is too superficial, according to experts at Fincraft Capital. AI may estimate a country’s population using out-of-date data, show market growth without adjusting for inflation, or use industry-wide averages instead of figures for a specific segment.
Step four - refining the sources and the query. At this stage, it is necessary to examine local data: government statistics, import and export figures, sector reports, actual prices on marketplaces, and data on household expenditure from local studies. Refinements should be made regarding the region, purchasing power and sales channels. Experts at Fincraft Capital s.r.o. note that following this adjustment, the quality of the data improves significantly.
The AI then processes the model. This yields a more accurate calculation, a list of competitors, a channel structure and basic unit economics. But this is not yet the final result.
Only after several iterations - usually five or six — is a document produced that is ready for use. It is not perfect, but it is already sufficiently close to reality.
The problem is that plausibility does not necessarily reflect reality, as the experts at Fincraft Capital s.r.o. point out. AI confidently produces an answer, even when analysing data that is incorrect or, at times, irrelevant. And whoever reads the report concludes that they have a good understanding of the situation and the market. Yet this feeling most often turns out to be an illusion.
In this article, experts from Fincraft Capital explain the TAM/SAM/SOM market valuation methodology and, using this as an example, highlight the ‘pitfalls’ of analysing data using AI.
The Triad of Concepts: TAM, SAM and SOM
The TAM/SAM/SOM methodology is a kind of filtering system that analyses the market step by step, progressively narrowing down market shares: from the total market to the addressable segment, and then to the portion you can realistically claim. TAM, SAM and SOM represent three levels of the same question: how much can be earned here?
- TAM (Total Addressable Market) - the entire market. All the money spent on goods.
- SAM (Serviceable Addressable Market) - the accessible part of the market. The part that is of interest to the researcher: in terms of geography, language and sales channels.
- SOM (Serviceable Obtainable Market) - a realistic revenue forecast, taking into account competitors, budget, team and time. This is the key figure. It forms the basis for the financial model and the decision on whether or not to enter the niche.
There are two approaches to calculating these metrics.
‘Top-down’ - starting from the global market, then breaking it down by region, category and segment. Experts at Fincraft Capital note that this estimate is almost always overestimated, as it is based on averaged data and assumptions. ‘Bottom-up’ - starting with a specific customer: customer acquisition cost, average spend, number of orders, and processing capacity. The result is closer to reality. It is precisely the difference between these approaches that most often distinguishes fantasy from a workable scenario.
How to conduct research using AI: a step-by-step guide
AI can indeed be a useful tool — if used correctly. The more specific the input, the more useful the result, according to managers at Fincraft Capital Czech Republic. At the same time, it is important to set the rules from the outset: insist on sources, references and transparent calculation methods.
Step one - establish the context: production or sales, in which region, and using which business model (in-house production, dropshipping, order fulfilment, D2C). Specify the sales channels: your own website, marketplaces, social media. It is important that every detail narrows the analysis down from the abstract notion of ‘a large market’ to the specific ‘niche and its parameters’.
Step two - define the methodology and set the objectives: calculating market size, analysing competitors, evaluating sales channels, and developing a basic financial model.
Step three - the initial results, their assessment and any issues. As a first approximation, the AI generates a document covering the funnel (TAM – SAM – SOM), competitors, channels and the financial model. The first iteration will almost always be based on international reports — Statista, global surveys and English-language sources. For the local market, this is too superficial, according to experts at Fincraft Capital. AI may estimate a country’s population using out-of-date data, show market growth without adjusting for inflation, or use industry-wide averages instead of figures for a specific segment.
Step four - refining the sources and the query. At this stage, it is necessary to examine local data: government statistics, import and export figures, sector reports, actual prices on marketplaces, and data on household expenditure from local studies. Refinements should be made regarding the region, purchasing power and sales channels. Experts at Fincraft Capital s.r.o. note that following this adjustment, the quality of the data improves significantly.
The AI then processes the model. This yields a more accurate calculation, a list of competitors, a channel structure and basic unit economics. But this is not yet the final result.
Only after several iterations - usually five or six — is a document produced that is ready for use. It is not perfect, but it is already sufficiently close to reality.
Where fantasy begins
Despite their structure and the figures, AI models have recurring weaknesses.
Optimism in forecasts. AI tends to construct ‘rosy’ scenarios. The LTV-to-CAC ratio might appear to be 10 to 1, growth might seem stable and predictable, and profit margins might appear to be above the market average. In theory, this is possible. In practice, it almost never happens at the outset.
Unsubstantiated pricing. This often appears in the model without clear justification. The price is taken as the market average, but without any understanding of why a customer should buy from you specifically.
Inflation and currency distortions. AI may show market growth of 25 per cent in the national currency and present this as a positive sign. But if inflation in the country stands at 15–20 per cent, real growth is significantly more modest, and sometimes non-existent. Currency conversions and inflation adjustments are a weak point for AI, and this work needs to be done manually, warn experts at Fincraft Capital.
Outdated data. Even basic indicators such as population size or income levels may not reflect the current situation.
And finally, the most important factor is the lack of cultural context. AI does not have a feel for the market. It does not understand what is happening with consumer behaviour, which segments are ‘dying out’, where people are cutting back, and where they are prepared to pay a premium. It does not have a feel for price – it simply calculates it.
Despite their structure and the figures, AI models have recurring weaknesses.
Optimism in forecasts. AI tends to construct ‘rosy’ scenarios. The LTV-to-CAC ratio might appear to be 10 to 1, growth might seem stable and predictable, and profit margins might appear to be above the market average. In theory, this is possible. In practice, it almost never happens at the outset.
Unsubstantiated pricing. This often appears in the model without clear justification. The price is taken as the market average, but without any understanding of why a customer should buy from you specifically.
Inflation and currency distortions. AI may show market growth of 25 per cent in the national currency and present this as a positive sign. But if inflation in the country stands at 15–20 per cent, real growth is significantly more modest, and sometimes non-existent. Currency conversions and inflation adjustments are a weak point for AI, and this work needs to be done manually, warn experts at Fincraft Capital.
Outdated data. Even basic indicators such as population size or income levels may not reflect the current situation.
And finally, the most important factor is the lack of cultural context. AI does not have a feel for the market. It does not understand what is happening with consumer behaviour, which segments are ‘dying out’, where people are cutting back, and where they are prepared to pay a premium. It does not have a feel for price – it simply calculates it.
Where reality comes into play
A working model begins where abstract percentages end, note the managers at Fincraft Capital. SOM is not a percentage of SAM. It is a metric that depends on resources: budget, expertise, marketing channels and operational capabilities.
That is why it is so important, first and foremost, to calculate the unit economics: customer acquisition cost, average spend, and profit margins. Then assess your potential: how many customers you can realistically attract. Only then can you attempt to estimate revenue.
Every assumption must be verified. Price should be checked against competitors’ actual offers. Customer acquisition cost should be verified using actual bid rates across advertising channels. Demand should be assessed through user behaviour, not through reports.
And the most important recommendation from the experts at Fincraft Capital is that the model needs to be recalculated as new data becomes available.
Data Interpretation and Quality
AI has made market research faster and more accessible. But alongside this speed, it has brought a new problem - the illusion of understanding.
Nowadays, it is easy to obtain figures. It is far more difficult to interpret them and understand which ones reflect reality.
The quality of the analysis is determined by the quality of the data: actual supplier prices, the actual cost of traffic, and actual conversion rates. Fincraft Capital s.r.o is a platform where online retailers can find verified suppliers, compare terms and conditions, and make decisions based on data rather than assumptions.
A working model begins where abstract percentages end, note the managers at Fincraft Capital. SOM is not a percentage of SAM. It is a metric that depends on resources: budget, expertise, marketing channels and operational capabilities.
That is why it is so important, first and foremost, to calculate the unit economics: customer acquisition cost, average spend, and profit margins. Then assess your potential: how many customers you can realistically attract. Only then can you attempt to estimate revenue.
Every assumption must be verified. Price should be checked against competitors’ actual offers. Customer acquisition cost should be verified using actual bid rates across advertising channels. Demand should be assessed through user behaviour, not through reports.
And the most important recommendation from the experts at Fincraft Capital is that the model needs to be recalculated as new data becomes available.
Data Interpretation and Quality
AI has made market research faster and more accessible. But alongside this speed, it has brought a new problem - the illusion of understanding.
Nowadays, it is easy to obtain figures. It is far more difficult to interpret them and understand which ones reflect reality.
The quality of the analysis is determined by the quality of the data: actual supplier prices, the actual cost of traffic, and actual conversion rates. Fincraft Capital s.r.o is a platform where online retailers can find verified suppliers, compare terms and conditions, and make decisions based on data rather than assumptions.