The Won-Deal Persona Teardown: 120 Deals to Ranked Buyer Titles
Turn 60 won and 60 lost deals into a ranked target-title list with lift, sample counts, and Wilson intervals, plus the deck personas to drop.
Key takeaways
- Logged contact roles usually hold about one name per deal, while Gong data puts won deals at 8.22 unique email participants against 3.38 on lost deals, so the buyer roster has to be rebuilt from activity records.
- IsPrimary usually marks whoever came in through lead conversion, typically the person who started the deal, so a persona built from primary contacts describes who starts deals, not who signs them.
- A title found in 4 of 20 won deals has a Wilson 95% interval of 8.1% to 41.6%, a 33-point spread that is too wide to rank anything.
- Even with 60 won and 60 lost deals, a 3.0x lift (12 of 60 won vs 4 of 60 lost) leaves the intervals overlapping between 11.8% and 15.9%, so the title stays provisional.
- In Refolk's index, Head of titles make up 60% of UK Revenue Operations leadership against 24% in the US, so seniority has to be judged by company size, not by the words in the title.
- Lift must be computed per deal and against lost deals, because won deals carry about twice the buyer contacts and every title otherwise looks like a winner.
This guide is for founders selling their own product, account executives, SDR leads, and partnerships teams who need to know which job titles actually start, approve, and sign their won deals. It takes one export of 60 won and 60 lost deals through every decision point, including three wrong turns. It ends with a ranked target-title list showing each title group's lift and sample count, plus a drop list of deck personas the evidence does not support.
What does the worked example start from?
The example starts from a 120-row CRM export: 60 Closed Won and 60 Closed Lost deals, each with a create date, close date, and amount. The seller's sales deck names "VP of Revenue Operations" as the buyer. Everything below tests that one line on the deck.
Most top-ranking guides on building a buyer persona from closed won deals say to pull 20 to 30 won deals and look for patterns. That approach skips three problems, and this teardown is built around them:
- Logged contact roles leave out most of the people in a deal. The CRM usually holds one name per deal. The buying group has several.
- Titles change after the deal closes. The CRM shows the person's job now, not the job they had when they signed.
- Twenty won deals cannot separate a pattern from noise. Without lost deals there is no base rate to compare against. Without enough deals, the margin of error is wider than the difference you are trying to see.
A few terms, each defined once:
- Contact role: a CRM record that links one contact to one opportunity, with a role label and a primary flag.
- Buying group: everyone on the buyer side who shaped the decision, whether or not anyone logged them.
- Base rate: the share of lost deals that contain a given title group. It answers the question "how often does this title show up anyway?"
- Lift: the share of won deals that contain a title group, divided by its base rate. A lift of 3.0x means the group shows up three times as often in wins as in losses.
- Wilson interval: a range of likely values for a percentage. Unlike the textbook formula, it stays accurate when the sample is small or the percentage is near 0% or 100%.
To keep the arithmetic checkable, the deal counts used here are the same ones in the interval table further down. Your counts will differ. What carries over to your data is the cut-offs and the decision taken at each fork.
Why do logged contact roles miss most of the buyers?
Logged contact roles miss buyers because converting a lead creates one role per deal and few teams add the rest, while won deals usually involve several people on the buyer side. Treat the logged roles as a starting list, never as the full roster.
The gap shows up from both directions. On the CRM side, one practitioner put the default plainly: "You'll almost always get one Contact Role populated. But in B2B sales, there are usually multiple parties involved in a deal." One consultant's anecdote goes further: "Nearly every client I've had has not used contact roles." Research on sales conversations points the same way:
- Gong Labs looked at 1.8M opportunities and found that 77% were multi-threaded (several buyer-side people were involved). Deals that closed had twice as many buyer contacts as deals that did not.
- An older Gong study of more than 50,000 opportunities found that winning deals had eight points of contact over email and losing deals had three.
- A secondary write-up of Gong data counts 8.22 unique email participants on won deals against 3.38 on lost deals.
- Kaylee Edmondson's 12-month ABM analysis found that closed-won deals touched 6.2 stakeholders on average.
- Gartner's widely cited range for a buying group is 6 to 10 stakeholders (secondary citation).
If logged roles sit near one per deal and email participants near eight, the roster multiplier is about 8x. That number combines two different sources and nobody has verified it. No published audit compares logged roles with the real buying group on the same deals. So step 3 of the procedure measures the multiplier on your own export instead of assuming it.
What the role fields can and cannot tell you
The OpportunityContactRole object has these fields: ContactId, OpportunityId, Role (a picklist), IsPrimary, CreatedDate, and CreatedById. The default role values listed on the related AccountContactRelation object are:
- Business User
- Decision Maker
- Economic Buyer
- Economic Decision Maker
- Evaluator
- Executive Sponsor
- Influencer
- Technical Buyer
- Other
None of these values marks who started the deal or who signed it. Only one contact role per opportunity can carry the Primary flag. Nobody has published how often roles are empty on closed deals, so measure that share yourself in step 2.
Fork 1: is the primary contact the buyer?
The tempting move is to filter to IsPrimary = true and read off the titles. That is the first wrong turn. IsPrimary often marks whoever came in through lead conversion, which is usually the person who started the deal rather than the person who approved or signed it. A persona built from primary contacts describes who downloads your content and books the first call.
At this fork, the example keeps IsPrimary as a column, because it is a useful signal of who started the deal. It also adds a separate signer column, filled in from the order form. Starters and signers are both target titles, but for different stages of outbound, so keep them in separate columns.
Which deals and people belong in the rebuilt roster?
All 120 deals belong in it, including deals with no logged roles. A person belongs if they are on the buyer side, have an email, meeting, or call dated between the opportunity's create date and close date, and pass an engagement filter of two or more interactions or one meeting.
The first trap is in the export itself. In Salesforce's standard "Opportunities with Contact Roles" report, "any Opportunity without a Contact Role doesn't appear." Deals with no roles disappear without any warning, and that inflates the base rate. Build a custom report type that keeps opportunities whether or not they have contact roles.
Next comes the baseline (step 2). Before adding anyone, record three numbers separately for won and lost deals: contact roles per deal, the share of deals with zero roles, and how many roles per deal are marked primary. These are your "before" numbers. The roster multiplier is the rebuilt count divided by the logged count.
Sources disagree on whether the rebuild should come before the baseline. One RevOps tooling vendor argues that contact roles should be required before a deal can close, because once a deal is "Closed Won only: Too late." That is the right fix going forward, but it does not help a look back over deals that already closed. For a teardown, take the baseline first, because that is the only way to measure the gap.
Rebuilding one deal's buyer roster
- Logged rolesStart from every contact role on the opportunity, primary or not
- Activity participantsAdd buyer-side people from emails, meetings, and calls
- Deal windowKeep only activity dated between create date and close date
- Engagement filterKeep people with two or more interactions or one meeting
- Signer checkMark who signed the order form, separately from IsPrimary
The engagement filter matters more than it looks. A webinar attendee or someone CC'd on one email makes the roster look bigger without adding a real buyer. Keeping only people with two or more interactions or one meeting removes most of that noise without dropping real evaluators.
deal_id = CRM opportunity ID outcome = won or lost create_date = opportunity create date close_date = opportunity close date amount = deal amount person_name = buyer-side person in_logged_roles = Y if a contact role exists, else N is_primary = Y or N logged_role = picklist value, blank if none signed_order_form = Y or N emails_in_window = count between create and close meetings_in_window = count between create and close calls_in_window = count between create and close passes_engagement = Y if 2+ interactions or 1+ meeting title_at_close = restored title title_source = profile dates, signature block, invite, or CRM-current title_source_date = date of that source days_from_close = gap between source date and close date flag_over_90_days = Y or N function_group = normalized function level = parsed level level_basis = title words or company size unclassified = Y or N
Use these as column headers. Logged contacts and activity-only people go in the same sheet, told apart by in_logged_roles.
How do you restore each person's title as of the close date?
Use the title source dated closest to the close date, in a fixed order of preference, and record the title, the source, and the source date on every row. Fall back to the current CRM title only when nothing else exists, and flag it when you do.
This is the second fork, and the wrong turn is to use today's titles. Picture someone who was a manager when the deal closed and is a VP now. Using their current title makes VP-level look like the buyer. That is exactly the evidence that would wrongly confirm "VP of Revenue Operations" on the example deck.
How big is the drift? Nobody has published how often a CRM title disagrees with the person's title on a profile record. The closest proxy is how long people stay with one employer. In the US Bureau of Labor Statistics employee tenure survey, median tenure with the same employer was 3.9 years overall, 2.7 years for ages 25 to 34, 3.5 years in the private sector, and 6.2 years in the public sector. Those figures count only changes of employer. Promotions inside the same company are not counted, so title drift is higher than tenure suggests. These are point-in-time figures, so check the latest BLS tenure release before you quote them.
I found no documented procedure for recovering the title someone held on a past date. The order below is working practice and has not been verified:
Title sources by order of preference
- Profile position datesStart and end dates on a public profile show which role covered the close date
- Email signature blockThe signature on a dated email inside the deal window
- Calendar inviteThe title on a meeting invite or attendee record from the deal window
- Current CRM titleFallback only, flagged, because it reflects today and not the close date
Flag any row where the gap between the source date and the close date is more than 90 days. That cut-off is also unverified. If too many rows get flagged to be practical, record the cut-off you chose and the share of rows over it, so the next person can tighten it.
Position dates are the slowest source to collect by hand, and they are where Refolk removes the most work. You can search on role changes directly, for example "people who were promoted into a Head of Revenue Operations role in the last 12 months at London fintech companies". That is the same dated position evidence this step needs, without reading profiles one at a time.
How should titles be grouped into function and level?
Treat function and level as two separate fields. Map function to a coarse group, either an O*NET major group or a short in-house list. Work out level on its own and check it against company size. Record what share of titles you could not classify.
Function: keep it coarse
Classifying jobs from the title alone is unreliable, even for humans:
- O*NET-SOC, the US government's occupation classification, lists 1,016 occupations, and free open-source tools exist that map job titles to its codes.
- One open-source Sentence-BERT coder reports 41.8% accuracy at the exact occupation (276 of 661 titles), rising to 73.4% at the broad 2-digit major group.
- Two human coders classifying 220 job titles agreed on only 55% of the codes.
- One study ran 30,000 titles through an O*NET coding tool, dropped results scored below 70, and kept 12,908 unique titles.
The broad major group, at 73.4%, is the only level stable enough to use without checking by hand. Nobody has published how many titles an open scheme leaves unclassified, so record your own share. A high share means your in-house list is missing a function.
Function labels also overlap more than sales decks suggest. These are counts from Refolk's index of professional profiles for US leadership titles (VP, Director, or Head of). The percentage rows come from the top 25 results for each label:
| Metric | Revenue Ops | Sales Ops |
|---|---|---|
| US total | 1,645 | 2,031 |
| Director share | 48% | 36% |
| Head share | 24% | 20% |
| VP share | 20% | 16% |
| Off-function noise | 8% | 28% |
The Sales Ops search returned "Rev Ops Lead" among its results, and 7 of its 25 results were off-function (VP of Sales variants, a CEO, a VP Accounts). The Revenue Ops search picked up "Head of Revenue Cycle Management", which is a healthcare billing title, and "Head of Sales - Americas". At this fork, the example merges Revenue Ops and Sales Ops leadership into one function group for the lift calculation. Treating them as separate personas would split one buyer group into two thin groups that rarely reach the deal count needed to rank them. Keep the original label as a column so messaging can still tell them apart.
Level: decide by company size, not by the word
SOC codes do not encode seniority. One coding tool assigns 11-2022 (Sales Managers) to the title "sales director". So level has to be worked out separately, and how titles signal level depends on the country:
| Metric | US | UK |
|---|---|---|
| Total matching | 1,645 | 213 |
| Director share | 48% | 28% |
| Head share | 24% | 60% |
| VP share | 20% | 8% |
| Off-target or other | 8% | 4% |
Source: Refolk's index, Revenue Operations leadership titles. Shares come from the top 25 results in each country. The UK Head share includes one variation in capitalisation and one typo ("Hed of Revenue Operations"), which a strict text filter would have missed.
The decision here is to set level using company size wherever "Head of" appears. A Head of Revenue Operations at a small UK company may hold the same seat as a US VP. Before writing any level rules, look at how a target market actually titles the role:
How much evidence does a title group need before it goes on the target list?
A title group goes on the ranked list only when it appears in at least 10 won deals and its won-deal interval sits clear of its lost-deal interval. Lift is the share of won deals containing the group divided by the share of lost deals containing it, and each share gets a Wilson 95% interval.
The cell-count rule echoes a standard statistical rule of thumb: the simple textbook (Wald) interval needs at least 10 cases on each side, written n p̂ ≥ 10 and n(1 - p̂) ≥ 10. A cell here means the number of deals of one outcome, won or lost, that contain the group. The Wilson interval is preferred when the sample is small (under 30) or the percentage is near 0 or 1. The standard reference recommending it over the Wald interval is Brown, Cai and DasGupta (2001). When a count is zero, the rule of three gives a quick upper bound of 3/n.
| Scenario | Point | Wilson 95% interval |
|---|---|---|
| Title in 4 of 20 won | 20% | 8.1% to 41.6% |
| Title in 12 of 60 won | 20% | 11.8% to 31.8% |
| Title in 4 of 60 lost | 6.7% | 2.6% to 15.9% |
| Title in 0 of 60 lost | 0% | at most 5% (rule of three) |
Fork 3: won deals only, twenty of them
The third wrong turn is the one most guides recommend: take 20 won deals, count titles, and rank. Suppose VP of Revenue Operations appears in 4 of 20 won deals and 1 of 20 lost deals. That looks like a 4x lift and seems to confirm the deck. But the won interval alone runs from 8.1% to 41.6%, 33 points wide, and the lost interval overlaps it. And without lost deals at all, there is no lift to calculate. A title that appears in almost every deal would top the list just by being common.
A title that appears in every deal is not a winner. It is furniture.
At 60 and 60: a real lift that still does not separate
With the full export, suppose the merged Director of Revenue and Sales Operations group appears in 12 of 60 won deals (20%, interval 11.8% to 31.8%) and 4 of 60 lost deals (6.7%, interval 2.6% to 15.9%). The lift is 3.0x, and it clears the 10-deal cell rule. But the intervals overlap between 11.8% and 15.9%. The decision is to mark it provisional and not rank it yet. There are two legitimate ways to settle it: add more closed deals, or merge function labels that genuinely overlap. Splitting the group into smaller groups until one looks good is not legitimate.
Now take a group found in 12 of 60 won deals and 0 of 60 lost. The rule of three caps the lost share at 5%, and the won interval starts at 11.8%, so the two separate. Report it as "not seen in lost deals, lost share at most 5%" rather than as infinite lift. The point lift is at least 4x.
Rank, watch, ignore, or drop
The formulas below compute everything for one row of the lift table. Check your sheet against the interval table above before you trust it.
Won share (F2) =B2/C2 Won lower (G2) =(F2+1.9208/C2-1.96*SQRT(F2*(1-F2)/C2+0.9604/C2^2))/(1+3.8416/C2) Won upper (H2) =(F2+1.9208/C2+1.96*SQRT(F2*(1-F2)/C2+0.9604/C2^2))/(1+3.8416/C2) Lost share (I2) =D2/E2 Lost lower (J2) =(I2+1.9208/E2-1.96*SQRT(I2*(1-I2)/E2+0.9604/E2^2))/(1+3.8416/E2) Lost upper (K2) =(I2+1.9208/E2+1.96*SQRT(I2*(1-I2)/E2+0.9604/E2^2))/(1+3.8416/E2) Lift (L2) =IF(D2=0,"lost share at most "&TEXT(3/E2,"0.0%"),F2/I2) Passes (M2) =AND(B2>=10,G2>K2) Self-check: B2=12, C2=60 gives 11.8% to 31.8%. D2=4, E2=60 gives 2.6% to 15.9%.
B2 = won deals containing the group, C2 = total won deals, D2 = lost deals containing it, E2 = total lost deals. Fill down one row per group.
The Passes column compares the won lower bound with the Wilson upper bound for lost deals. When the lost count is zero, that bound is slightly higher than the rule-of-three figure, so the test is a little stricter than the rule-of-three check described above.
The teardown procedure, step by step
Run the seven steps below in order on your own CRM. The data work takes most of an afternoon to a day: about an hour for the export, 30 minutes for the baseline, 2 to 3 hours for the roster rebuild, 2 hours for titles, and an hour for normalization, before you compute lift.
From 120 closed deals to a ranked target-title list
- Export the deal setPull 60 Closed Won and 60 Closed Lost deals with create dates, close dates, and amounts, using a custom report type that keeps deals with no contact roles. Done means exactly 120 rows.
- Baseline the logged rolesCount contact roles per deal, the share of deals with zero roles, and the IsPrimary distribution, separately for won and lost. Done means a contacts-per-deal figure for each.
- Rebuild the roster from activityAdd buyer-side email, meeting, and call participants dated between create and close, keep those with two or more interactions or one meeting, and mark the signer. Done means a measured roster multiplier.
- Restore titles as of the close dateUse the title source dated nearest the close date: profile position dates, then signature blocks, then invites. Done means title, source, and source date on every row, with gaps over 90 days flagged.
- Normalize into function and levelMap function to an O*NET major group or short in-house list, and set level separately, using company size for Head of titles. Done means an unclassified share is recorded and the top 25 titles per group have been checked by hand.
- Compute lift with Wilson intervalsFor each group, divide the share of won deals containing it by the share of lost deals containing it, and add Wilson 95% intervals. Done means every row shows won n, lost n, lift, and both intervals.
- Rank and set the drop listRank groups with at least 10 won deals and separated intervals, and move deck personas that fail to separate onto a drop list with their counts. Done means two lists.
Each step's "done" condition is there so that someone else can audit the step. If you skip writing down a source date or an unclassified share, the final list cannot be checked, and an unchecked persona list ends up getting argued about rather than used.
How does this analysis go wrong?
It goes wrong in eight predictable ways. Most of them inflate the evidence for whatever persona was already on the deck. Each one below has a symptom and a check you can run in minutes.
- Taking the primary contact as the buyer. Symptom: the persona list is full of managers and individual contributors who start deals, and nobody who signs. Check: compare IsPrimary against the person who signed the order form.
- Only looking at deals with roles. Symptom: the base rate looks oddly high and the deal count is short. The standard report type silently drops deals that have no roles. Check: the row count must be 120.
- Using today's titles. Symptom: VP-level titles dominate won deals that a manager actually ran. Median employer tenure is 3.9 years, and promotions add more drift on top of that. Check: the gap between the title source date and the close date is no more than 90 days, an unverified cut-off you should record.
- Counting attendance as buying. Symptom: won rosters swell with webinar attendees and people CC'd once. Check: require two or more interactions or one meeting.
- Keyword matching pulling in the wrong function. Symptom: healthcare revenue-cycle titles or VP of Sales variants land in an operations group. In Refolk's index, 28% of the Sales Ops sample and 8% of the RevOps sample were off-function. Check: review the top 25 titles in each group by hand.
- Treating "Head of" as one seniority level. Symptom: UK buyers look junior next to US buyers. UK results skew heavily to "Head of", 60% against 24% in the US. Check: set level by company size, not by the word.
- Lift on thin cells. Symptom: a title found in 4 of 20 won deals and 1 of 20 lost deals reads as 4x. Check: require 10 or more deals in a cell and intervals that do not overlap.
- Skipping lost deals. Symptom: a title that appears in almost every deal ranks first. Check: calculate lift against the share of lost deals containing that title.
Failure modes 1, 3, and 8 are the three wrong turns from the worked example, and they compound. A persona built from primary contacts, using today's titles, and counted only across won deals can confirm almost any deck. Fix them in that order. Each fix changes the counts that the next one depends on.
There is one false positive that none of these checks catch: a title linked to winning because of deal size or segment, not because that person drives the decision. Gong reports that multi-threading boosts win rates by 130% in deals over $50K. So if large deals both win more and pull in more senior titles, seniority can look predictive when the real driver is deal size. If your amounts vary widely, re-run the lift table within deal-size bands before you act on a senior title.
What should you do with the ranked list, and how do you keep it current?
Turn the ranked list into named people at target accounts, keep the drop list visible to everyone who writes outbound, and rerun the teardown on the next batch of closed deals with contact roles enforced before close.
The deliverable has two parts. The ranked target-title list shows each group's function, level, won n, lost n, lift, and both intervals, sorted by lift and limited to groups that pass. The drop list shows each deck persona that failed, with its counts, so nobody has to rely on memory for why VP of Revenue Operations came off the deck. Provisional groups go in a short third section labelled "watch".
Next, check that the market can actually supply the titles you ranked. Market-level counts are easy to get. Per-account coverage by company-size band has not been published anywhere, so measure it on your own account list before promising pipeline.
Turning a title group into names is where the evidence gets used. Refolk takes the title, function, company size, and region straight from your ranked list as a plain-English query and returns the people behind them, so the persona becomes a call list without a hand-built Boolean search string.
To keep the list current:
- Require contact roles before close from now on, so the next roster rebuild starts from better data.
- Rerun the teardown once another 60 won and 60 lost deals have closed, and compare intervals, not just point lifts.
- Move a watch-list group onto the ranked list only when its intervals separate on new deals.
Before you publish the target-title list
- The export has exactly 120 rows, including deals with zero contact roles.
- Logged contacts per deal are recorded separately for won and lost deals.
- The roster multiplier is a measured number from this export, not the derived 8x.
- Every roster person has two or more interactions or one meeting inside the create-to-close window.
- Every won deal has a signer column filled from the order form, separate from IsPrimary.
- Every row has a title, a title source, and a source date, with gaps over 90 days flagged.
- The unclassified share of titles is recorded.
- The top 25 titles in each function group were checked by hand for off-function noise.
- Head of titles were levelled by company size, not by the word.
- Lift is computed per deal against lost deals, and every ranked row shows won n, lost n, lift, and both intervals.
- No group is ranked with fewer than 10 won deals or with overlapping intervals.
- Every deck persona appears on either the ranked list or the drop list, with its counts.
Questions practitioners ask
How many closed deals do I need to work out which titles buy?
Use at least 60 won and 60 lost deals, and expect even that to separate only strong patterns. With 20 won deals, a title in 4 of them has a Wilson 95% interval of 8.1% to 41.6%, which cannot rank anything. At 60 and 60, a 3.0x lift (12 vs 4 deals) still leaves overlapping intervals. Require at least 10 won deals in a cell before ranking a title, and merge overlapping function labels when cells are thin.
Can I just report on Opportunity Contact Roles to see which titles buy?
No. Contact roles are a starting list, not the buying group. Converting a lead usually creates one role per deal, while Gong data shows won deals with 8.22 unique email participants against 3.38 on lost. The standard Opportunities with Contact Roles report also leaves out deals that have no roles at all. Rebuild the roster from email, meeting, and call participants dated between create and close.
Why do I need lost deals for a buyer persona?
Without lost deals, you have no base rate. A title that shows up in nearly every deal will look like a winning persona even if it appears just as often in losses. Lift divides the share of won deals that contain a title by the share of lost deals that contain it. Only a lift whose Wilson intervals separate tells you the title is actually linked to winning.
What do I do when a title appears in zero lost deals?
Don't report infinite lift. Use the rule of three: with zero hits in 60 lost deals, the true lost share is at most 3/60, or 5%. If the title appears in 12 of 60 won deals, its won interval starts at 11.8%, above that ceiling, so it separates. Report it as not seen in lost deals, with a lost share of at most 5%.
Should Sales Ops and RevOps be separate target personas?
Usually not for scoring. In Refolk's index, a Sales Operations leadership search returned Rev Ops Lead among its results, and 28% of that sample was off-function. Treating the two labels as separate personas splits one buyer group into two thin cells that rarely reach 10 won deals each. Merge them into one function group for lift, and keep the original label as a column for messaging.
How do I find the title someone held when the deal closed?
Use whichever title source is dated closest to the close date. In order: profile position dates, then the signature block on a dated email, then calendar invites. Fall back to the current CRM title only when nothing else exists, and flag it. No documented standard exists for this, so treat the order and the 90-day flag as working practice you can adjust and record.
Try it on the search you came here for
Stop building boolean strings. Just describe the person.
Type one sentence. I plan the search, read GitHub, public LinkedIn and Crunchbase records, and the open web as it is right now, and hand back a ranked list with the reason next to every name.
01Describe them
One plain sentence. Role, city, stack, stage, whatever matters to you.
02I read the web live
GitHub, public LinkedIn and Crunchbase records, the open web. Not a database that went stale last quarter.
03You read the shortlist
Ranked, with the reasoning under every name. Open a profile, ask a follow-up, narrow it down.
- Staff backend engineers in NYC who shipped Rust in production
- Series A fintechs in SF under 50 people, growing headcount this year
- Maintainers of fast-growing Rust web frameworks on GitHub
- No boolean, no filters, no seat to buy. One box.
- Read at search time, so a profile updated yesterday counts today.
- Every step visible as it runs, every name with its reason.
500 free credits on sign-up. No card, no demo call. See real searches.