Fit check: why we score leads before you send
Every lead gets a fit check from 0 to 100 that answers one question: is this company worth an email at all? It is separate from the lead score (fit 0–60 plus intent 0–40), which ranks how ready a lead is. A company can be very ready to buy something you don't sell.
The default send line
| Fit check | What happens |
|---|---|
| 80 and above | Pushed to your sequencer or CRM when you approve, included in the CSV export, and allowed in auto mode. |
| 40–79 | Waits for your review. Approving it asks you to confirm "Send anyway". Left out of the default CSV export. |
| Under 40 | Held back. Same as above: you can still send it anyway, one lead at a time. |
You can change both lines in Settings (0 turns the fit check off). "Export all fits" on the Leads page downloads every lead, with the fit check and reasons in each row.
How the 100 points are made
| Area | Points | What we check |
|---|---|---|
| What they sell | 25 | Does the business look like the kind of company in your ICP? |
| Who they sell to | 15 | Given their customers, would your offer matter to them? |
| Size | 15 | Employee count inside your range (5 if unknown or close). |
| Location | 15 | Country or city in your ICP (5 if unknown). |
| Data check | 15 | Do the name, industry, size, country and description agree? Is the domain a real company and not a free email provider? |
| Contact | 15 | The person's title matches your buyer titles (10) and the email is verified (5) or catch-all (3). |
Some things cap the score whatever else matches: a competitor of yours (max 10), an invalid email that would bounce (max 39), and an AI read that says neither what they sell nor who they sell to fits (max 30). If only one of those two is a clear "no", the lead goes to review (max 59).
The lead page lists every area with its points and a one-line reason.
Where the AI comes in, and what it costs
"What they sell", "who they sell to" and the data check use one AI read per company, on Surge's own server. The AI reads up to 8 companies in one call and only sees the data we already hold (name, website, industry, size, country, description and the signals). It does not visit their site or LinkedIn. The read is saved and reused for 60 days, or until you change your ICP.
- Cost: about ₹0.10 per AI call (our estimate of our own compute), so roughly ₹0.01–0.02 per company. Free during your first 3 months. After that it is charged at our cost from your wallet, like every other AI call, and shows in each run's log.
- When the AI is off or fails, the rules alone score the lead and say so: "Who they sell to: not checked". Rules alone can reach 85 at most, so a lead with perfect rule matches still gets pushed.
The spam-complaint maths
Gmail and Yahoo ask bulk senders to keep the spam-complaint rate under 0.1% and never reach 0.3%. 0.3% is 3 complaints in 1,000 emails. Go above it and your mail starts landing in spam or getting rejected, and a sending domain can take weeks to recover.
An example (the complaint rates are assumptions to show the shape, not measurements): you have 1,000 leads and, as is common, only about 35% are a real fit.
| Emails | Complaints if poor fits complain 1 in 200 and good fits 1 in 1,000 | Complaint rate | |
|---|---|---|---|
| Send to everyone | 1,000 | 650 × 0.5% + 350 × 0.1% ≈ 3.6 | 0.36%: over the limit |
| Send only good fits | 350 | 350 × 0.1% ≈ 0.35 | 0.1%: safe |
Most of your replies come from the good fits anyway, so you lose very little by not sending the rest, and you keep your domain.
For sequencers and the API
Pushes carry custom.fit_check (0–100) and custom.fit_reasons (one line per area). The MCP tool approve_lead refuses leads under the line with the reason, unless the user explicitly asked to send that lead anyway (send_anyway: true).