A win/loss analysis on a small number of deals is easy to get wrong. Two wins from the same industry start to look like a pattern. One lost deal with a loud reason becomes a rule. The fix is not more cleverness. It is a stricter count, an honest label on the sample, and a habit of writing conclusions as guesses until you have tested them.
This guide shows how to run a B2B win/loss analysis when you have few deals. It ends with a template and a list of checks.
Who this is for
This guide is for founders, first sales hires and small revenue teams who have closed a few deals and lost a few more. You may have a CRM, a spreadsheet, or notes in someone's head. You want to know which kinds of companies to chase next, and you do not have enough history for a statistical model.
It also suits anyone who has been told "we win in fintech" or "we lose to big companies" and wants to check whether the deals support it.
You do not need a tool to do this. A spreadsheet is enough. A later section covers how Outlevo handles the same job if you want to use it.
Steps
Count companies, not deals
The first rule is the one most small teams skip. Count companies, not deals.
A single company can produce several deals. It might have a pilot, then an expansion, then a renewal. It might have a lost deal one year and a won deal the next. If you count each deal, one enthusiastic customer looks like several. Your sample looks bigger than it is, and one company's traits get weighted several times over.
Outlevo's win/loss analysis follows the same rule: it counts companies, not deals, and labels small samples. Under 10 won companies is very low confidence [[1]](#sources). Use that as your own yardstick. If you have fewer than 10 won companies, say so in the first line of your write-up, and read everything below it as a set of leads, not findings.
1. Write down the question
Pick one question before you open the data. Good ones are narrow:
- Which kinds of companies have we closed, and which have we lost?
- Where do deals stall, and what do the lost ones have in common?
- Which competitors show up in lost deals?
A vague question ("what are we doing wrong?") invites you to find a story in any data. A narrow one tells you which columns to fill in.
2. List the deals, then collapse them to companies
Export or copy every closed deal in the period you care about. Include open deals in a separate list, but keep them out of the comparison. They have no outcome yet.
Then collapse the list so that each company appears once. You need a rule for companies with mixed outcomes. For example, a company that lost once and later won counts as won, with a note. Another sensible rule is to count the company by its most recent closed deal. Pick one rule, write it at the top of the sheet, and apply it to every row. The exact rule matters less than using it consistently.
3. Label each company won, lost or open
Use plain labels. Keep "lost" for companies that said no or went silent after a real conversation. Companies you never reached do not belong in the lost column. They are a different problem, and mixing them in will blur the result.
4. Add attributes, each with a source and a date
For each company, add the attributes you suspect matter. Common ones are industry, size, country, the stage the deal reached, the amount, and the loss reason.
Write down where each fact came from and when you read it. A company's size from its own website last month is better than a number from memory. Outlevo applies a similar bar to its own records: a Readiness Card needs at least three observations, each with a source and a date [[2]](#sources). You can borrow the habit. If you cannot back an attribute with a source, leave the cell empty instead of guessing.
Country is the one to handle with care. It tells you where past customers are. It does not tell you where to sell next. Outlevo asks which countries a customer sells into and never proposes markets [[1]](#sources). Treat your own sheet the same way: record country as a fact about the past, and make market decisions separately.
5. Compare won and lost, one attribute at a time
Take one attribute and split the companies by it. Look at industry first, then size, then deal stage reached. For each, ask two questions:
- Do won companies cluster in one value?
- Do lost companies cluster in the same value, or in a different one?
If both won and lost companies sit in the same segment, that segment does not separate the two groups. It may still be a good market, but this data cannot say why you win there. If won companies cluster in one segment and lost companies are spread out, that is a lead worth noting.
Resist combining attributes early. With few companies, every extra filter leaves you with a handful of rows, and a handful of rows will always look like a pattern.
6. Read the loss reasons with suspicion
Read each loss reason as one person's account of the deal, written after a loss. It is useful, but it is one view. "Price" may mean the buyer did not see enough value. "No budget" may mean the purchase was not a priority.
Group the reasons into a few plain categories and count companies in each. Note which reasons appear alongside which segments. Then, if you can, add what you know from the conversations. Do not treat the typed reason as the full answer.
If you lose deals to named competitors, keep a column for the competitor. Even in a spreadsheet, a competitor column is cheap to fill in.
7. Label the confidence
Write the sample size and a confidence label at the top of the summary, before any conclusion. For example: "Won companies: [count]. Confidence: [label]." Use the under-10 line from the start of this guide as your anchor. Above that line, do not invent your own tiers. Say how many companies you have and let the reader judge.
The point of the label is to stop the summary from travelling without its warning. A slide that says "enterprise buyers don't convert" will be repeated for months. A slide that says "enterprise buyers did not convert in this small sample" is harder to misuse.
8. Turn findings into hypotheses
Rewrite each observation as a testable statement about who to contact. A finding says what happened. A hypothesis says what you will try and what you expect.
- Finding: "Most won companies were in logistics."
- Hypothesis: "Logistics companies are a better fit than the average account we contact. We will contact more of them and compare reply and meeting rates against our other segments."
Outlevo takes the same stance. Guesses are labelled as hypotheses, never presented as findings [[2]](#sources), and AI-proposed ICP segments stay experimental until a person approves them [[1]](#sources). Your own notes deserve the same label. Mark each segment "experimental" until a test has run.
9. Design a small test for each hypothesis
For each hypothesis, decide what you will do, how long you will run it, and what result would make you drop it. Keep the test small. Contact a modest list in the segment, keep everything else about your outreach unchanged, and compare the outcome with a control group of accounts outside the segment.
Write the "drop it" condition before you start. Otherwise any result will look like support.
When the test finishes, add the new won and lost companies to the sheet and run the comparison again. The sample grows with each round.
A template you can copy
Set up one row per company with these columns:
| Column | What goes in it |
|---|---|
| Domain | The company's website domain, one row per domain |
| Name | Company name |
| Country | Where the company is based (a fact about the past, not a market recommendation) |
| Size | Employee band or range, with a source |
| Industry | Your own category, kept to a short list |
| Outcome | Won, lost or open |
| Stage reached | Last stage before close |
| Amount | Deal value, if known |
| Loss reason | Rep's reason, grouped into a short list |
| Competitor | Named competitor, if any |
| Source and date | Where the attributes came from and when you read them |
Outlevo's CSV import maps your columns to domain, name, country, size, industry and deal fields such as stage, amount, outcome and loss reason [[3]](#sources). Competitor and Source and date are not among the fields listed for the import, so keep those two for your own use.
Add a header block above the table:
- Period covered
- Won companies, lost companies, open companies
- Your rule for companies with mixed outcomes
- Confidence label
Add a block below it:
- Observations (what the data shows)
- Hypotheses (what you will test)
- Test plan and "drop it" condition for each
That is the whole template. It fits on one sheet.
What to check
Before you share the result, go through this list.
Is every company counted once? Sort by domain and look for duplicates. Subsidiaries and different spellings of the same company are the usual culprits.
Is the sample labelled? The count of won companies and the confidence label should be the first thing a reader sees.
Are the "lost" companies really lost? Remove companies that never had a real conversation, and companies still in play.
Does each attribute have a source and a date? If a cell has neither, fill it in or clear it.
Do the findings survive removing your best customer? Drop the single largest or most memorable win and re-run the comparison. If the pattern disappears, it was one company, not a segment.
Are loss reasons treated as opinions? Look for loss reasons that conflict with what you know from the calls.
Is every conclusion written as a hypothesis? No sentence in the summary should read as settled fact. Each segment is experimental until a test has run.
Does the plan respect the rules for contacting people? If a hypothesis leads to cold email, check the rules that apply to the people you plan to contact before you run the test.
Next step
Open a blank sheet and copy the template columns. Pull your closed deals, collapse them to companies, and write the won and lost counts at the top. Add the confidence label. Then pick the one attribute you most suspect, split the companies by it, and write one hypothesis with a "drop it" condition.
Doing it in Outlevo
If you would rather not maintain the sheet, Outlevo can run the analysis from your past deals. You import past deals from CSV or your CRM, and you get a win/loss analysis that proposes ICP hypotheses [[4]](#sources). The proposed segments are labelled experimental until a person approves them [[1]](#sources).
There are three ways in:
- CSV. An import can hold up to 200,000 rows, and it takes companies with their won, lost or open deals [[3]](#sources). Importing and exporting use no credits [[3]](#sources).
- HubSpot. Outlevo reads your companies (name, domain, owner, lifecycle stage), deals (pipeline, stage, amount, closed-lost reason) and owners [[5]](#sources). It learns your ideal customer from the deals you won and lost [[5]](#sources). Outlevo writes only to its own properties, never edits standard HubSpot fields, and writes nothing until an owner