Short answer: Give every deal a company domain, one standard outcome value, a loss reason and a competitor tag, and merge duplicate accounts first, because win/loss analysis counts companies, not deals [[1]](#sources). Fewer than 10 won companies is rated very low confidence [[1]](#sources). The CSV import maps columns such as domain, stage, amount, outcome and loss reason [[2]](#sources).
Who this is for
This guide is for a founder, sales ops lead or RevOps owner who has a pile of closed deals and wants to learn which kinds of companies buy and which do not. The deals may sit in a CRM, a spreadsheet or both.
You can import past deals from a CSV or from your CRM and get a win/loss analysis that proposes ICP hypotheses [[3]](#sources). The analysis is only as good as the file you give it. A column that says "Closed Won" in one row and "won" in the next, or three records for one account, will skew what it counts.
This guide covers the work you do before the import. It does not cover what to sell or where. Outlevo asks which countries you sell into and never proposes markets [[1]](#sources), so that decision stays with you.
If you only have a handful of deals, read how to run a win/loss analysis on few deals alongside this one.
Know the target before you clean. The fewer columns you need to fix, the less can go wrong.
CSV. A CSV import takes up to 200,000 rows and covers companies and their won, lost or open deals [[2]](#sources). It maps your columns to domain, name, country, size, industry and deal fields such as stage, amount, outcome and loss reason [[2]](#sources). Importing uses no credits, and anyone who can add accounts can run it [[2]](#sources).
HubSpot. Outlevo reads company name, domain, owner and lifecycle stage, plus deal pipeline, stage, amount and closed-lost reason [[4]](#sources). It learns the ideal customer from the deals you actually won and lost in HubSpot [[4]](#sources). Those fields should be filled in before you connect. Outlevo writes only to its own properties, never edits your standard fields, and writes nothing until an owner or admin approves the field mapping [[4]](#sources). Connecting and syncing use no credits [[4]](#sources).
Salesforce. Outlevo reads accounts, opportunities and their stages for the ICP and win/loss analysis [[5]](#sources). The connection is a pilot. It needs the Outlevo package installed in your org with Outlevo's help, and there is no AppExchange listing [[5]](#sources).
The cleanup below applies to all three routes. With a CRM connection you fix the records in the CRM. With a CSV you fix the file.
Steps
1. Export everything, then freeze a copy
Pull every closed deal you can, not only the recent or memorable ones. Include won, lost and, if you track them, open deals. The import covers all three [[2]](#sources).
Save the raw export untouched. Do all cleaning in a second file or a copy of the sheet. When a number looks odd later, you can go back to the original and see what changed.
Include these columns if you have them:
- Company name
- Company website domain
- Country, company size and industry
- Deal stage and amount
- Outcome: won, lost or open
- Loss reason
- Competitor
- Close date and owner, for your own checks
Close date and owner are not among the listed import fields [[2]](#sources). They help you spot problems in the sheet, so keep them in your working copy.
2. Standardize the outcome field
Outcome is the column the whole analysis hangs on, so settle it first.
CRM exports often carry several vocabularies at once. You may see "Closed Won," "Won," "closed-won," "Signed," "Customer" and a stage ID with no label. Lost deals may say "Closed Lost," "Dead," "No decision" or "Churned before start."
Do this:
- List every distinct value in the outcome and stage columns. A pivot table or a sort-and-filter does it in a minute.
- Decide which values mean won, which mean lost and which mean still open.
- Write the rule down in one place, such as a second tab, so the next person can follow it.
- Replace every value with one of three: won, lost, open.
- Re-list the distinct values. You should see exactly three.
Watch for the hard cases. "No decision" is a lost deal for your purposes only if the buyer never bought. A deal marked "won" and later cancelled needs a decision, and you should make the same decision for every similar row. Keep the original stage in its own column so you do not lose detail. The import maps stage and outcome as separate fields [[2]](#sources).
Check the amount column while you are here. Mixed currencies, text like "10k" and blank values are common. Make the column plain numbers, in one currency, or leave it blank when you do not know. A guessed amount is worse than an empty cell.
3. Deduplicate accounts before anything else counts them
This step matters more than it looks. The analysis counts companies, not deals, so several deals at one account count once [[1]](#sources). If one real customer appears as "Acme," "Acme Inc." and "ACME Corp," you may end up with three companies where there should be one.
Domain is the safest key. It is one of the fields the CSV import can map [[2]](#sources), and it gives each account a single identifier. Names drift. Domains mostly do not.
A workable pass:
- Add a domain column and fill it for every row. Use the company's main website, written the same way each time: lowercase, with no protocol prefix, no "www" and no trailing slash.
- Strip personal email domains and CRM placeholders. A row whose domain is a free mailbox provider is not a company website.
- Sort by domain and look at every group with more than one row.
- Decide what each group is. Several deals at one real account stay as several deals under one company record. Two company records for one account get merged.
- Check subsidiaries and parent companies. If a parent and a division have different domains, decide once whether they are one account or two, and apply the rule the same way everywhere.
Mind the case where one account has both a lost deal and a won deal, for example you lost the first attempt and won a year later. Do not delete the lost row to tidy things up. Keep both deals under the one account, with accurate outcomes and dates. How the analysis treats that account is its job. Yours is to make sure the record is accurate.
4. Handle missing loss reasons
Almost every export has gaps here. Reps skip the field, pick the first option in the dropdown or type "n/a." The CRM may have a required field that people fill with junk to get past it.
What to do:
- Keep the row. The analysis counts companies [[1]](#sources), so a lost deal with no reason still belongs in the file.
- Leave unknown as unknown. Use one explicit value, such as "Unknown," for blank and junk entries. Do not fill gaps from memory or guess from the deal's size or stage.
- Fill in only from evidence. If the rep's notes, a closing email or a call record state the reason, copy it in and note where it came from. If the reason lives only in someone's head, it is a recollection, not a record. Mark it that way if you use it.
- Merge near-duplicates. "Price," "too expensive," "budget" and "pricing" can share one label if they mean the same thing to you. Keep the groups few and plain. A list of 40 loss reasons with two deals each shows no pattern.
- Keep the free text. If reps typed notes, store them in a separate column. A tidy category is easier to count, and the raw note preserves nuance.
Then count how many lost deals end up as "Unknown." That number is useful by itself. It tells you how much weight the loss-reason findings can bear, and it is a prompt to capture reasons properly from now on.
5. Tag competitors on every deal you can
Outlevo's competitor tracking includes win/loss by competitor [[6]](#sources). That view is only useful if the deals carry a competitor to begin with. So tag them before you import.
Keep it simple:
- Create one competitor column. One name per cell for the main rival in that deal.
- Use one spelling per competitor. Make a short list of canonical names and replace every variant with it.
- Use explicit values for the other cases: "None" when the buyer chose to do nothing or build in-house, and "Unknown" when you cannot tell.
- If several competitors were in play, pick the one that was the final alternative, and put the others in a notes column.
Do not tag a competitor from a hunch. A confident wrong tag skews the by-competitor view more than a blank does.
Competitor is not among the import fields listed for the CSV [[2]](#sources). Keep the competitor tag in your master file, in the same row as the deal it describes.
6. Fill company attributes only where you have them
The import can map country, size and industry [[2]](#sources).
Fill them from your own records where you have them. Leave them blank where you do not. Pick one format for each: a consistent set of size bands, one spelling for each industry, two-letter or full country names but not both. Do not invent values to make the file look complete.
7. Check the sample before you import
Count companies, not rows. After deduplication, count the unique domains with a won outcome and the unique domains with a lost outcome.
Fewer than 10 won companies is rated very low confidence by the analysis [[1]](#sources). That does not mean the import is pointless. It means any pattern should be read as a lead to check, not a finding. The labels exist so you do not mistake a small sample for a firm result.
More won and lost companies give you more to compare. Include everything you can export.
What to check
Run through this list before you import.
Outcome
- The outcome column has exactly three values: won, lost, open.
- Your rule for edge cases is written down and applied the same way throughout.
- Stage is kept in its own column.
Accounts
- Every row has a domain, in one format.
- No two company records share a domain.
- No domain belongs to a free mailbox provider or a CRM placeholder.
- Accounts with several deals still have all their deals, with correct outcomes.
Loss reasons
- Every lost row is still in the file.
- Blank and junk reasons say "Unknown."
- Reasons you filled in came from a record, not a recollection.
- The category list is short enough that each category has real weight.
Competitors
- One column, one spelling per competitor.
- "None" and "Unknown" are used on purpose and kept apart.
Amounts and attributes
- Amounts are plain numbers in one currency, or blank.
- Country, size and industry are in consistent formats, or blank.
Size and format
- Unique won companies counted, and you know whether you are below 10.
- File is a CSV of up to 200,000 rows [[2]](#sources).
- Column headers are clear enough to map without guessing.
Access and privacy
- Remove columns you do not need, such as personal contact details, from the file you import.
- If you connect HubSpot instead, contact import is optional and only happens after an admin turns it on with a stated purpose [[4]](#sources).
Scripted flow
- If you plan to script the cleanup and import, test keys run against a sandbox with 100 fake companies, so you can build the flow without real data or credits [[7]](#sources).
- Metered API calls quote their credit cost first and run only when confirmed [[7]](#sources).
After the import
- Compare the company counts in the analysis to your own count. A mismatch may point to a domain you did not normalize.
- Treat the proposed segments as hypotheses. AI-proposed ICP segments are labelled experimental until a person approves them [[1]](#sources).
Next step
Fix the outcome column and the domains first. Those two changes affect every number that follows. Then work through loss reasons and competitors, and count your unique won companies.
When the file is ready, import it from a CSV or connect your CRM. HubSpot and Salesforce each have a page covering what is read and what is written, and the CSV and Excel page covers the file route. For what the analysis returns, see win/loss analysis software that proposes ICP hypotheses.
If you do not have deal data yet, the free ICP assessment reads up to 5 public pages of your website [[8]](#sources). It returns facts labelled as verified observations or AI inferences, and up to 5 experimental segment hypotheses [[8]](#sources).