The most useful thing in this workflow is a moment where the correct action was to say no.
Partway through building a prospect list, the assistant proposed a shortcut: skip ahead and use the connected data source to find the executives, then do everything together. It was a reasonable-sounding offer. It was declined.
That decision is the whole lesson. A clean list has to exist before enrichment starts, because enriching a half-built list means you can never tell which rows were missed and which were genuinely empty.
This guide walks the three stages in order, shows where the assistant will try to collapse them, and ends with an honest number: 31 direct phone numbers from 119 senior finance executives.
Why does the order of stages matter?
Because each stage produces something the next one depends on, and mixing them makes errors invisible.
The three stages. One, companies. Build the target account list. Two, people. Find the named individuals inside those accounts. Three, contact data. Enrich the finished list of people. Stages one and two cost nothing, because no contact data is requested. Only stage three spends credits.
Collapsing stages two and three sounds efficient and is not. If enrichment runs while the people list is still being assembled, a row with no phone number could mean the person has no findable number, or it could mean that row arrived after the lookup ran. Those are completely different facts and the spreadsheet shows them identically.
Keeping the stages separate also means a mistake in stage two is free to fix. A mistake discovered after stage three has already been paid for.

How do you connect the data source?
Create an API key, add a custom connector, confirm it before using it.
- In the Scalelist dashboard, open API in the left panel and click New key.
- In Claude, click your name at the bottom left, open Settings, then Connectors.
- Scroll to the bottom and click Add custom connector.
- Name it Scalelist and paste the MCP URL. Your API key goes immediately after the equals sign at the end of that URL.
- Click add.
The MCP is what makes this possible. Model Context Protocol is the standard that lets an assistant connect to an external tool and act on your behalf rather than only answering questions. Without it the assistant can discuss your prospect list. With it, it can go and find the numbers.
Confirm the connection by asking whether it is connected to your Scalelist MCP. It replies with the list of tools it can now use.
Stage one and two: building the list
Ask for companies first, then people, as two separate prompts.
Stage one here was a list of the top 100 finance companies in Chicago. That runs entirely on the assistant’s own knowledge and web search, with no connector involved and no credits spent. The output came back grouped oddly rather than as a clean list, which is cosmetic and fixable.
At the end of that response, the assistant offered to find emails or phone numbers. The answer is yes, but not yet.
Stage two asked for the C-suite executives at each company, with the output specified as a spreadsheet rather than paragraphs, and with named columns: first name, last name, job title, company name, company website and company type.
This is where the shortcut appeared. The assistant proposed using the connected data source to find the executives, and separately asked whether it could look for emails.
Both were denied. The reason for the first: executives should be found through research at this stage, so the list is complete and checkable before any paid lookup runs. The reason for the second: this run was about phone numbers, and allowing an unrequested email lookup would have spent credits on a channel that was not being tested.
After redirecting, stage two returned 119 C-suite executives across 80 companies, cleanly laid out in the requested columns.

Stage three: finding the phone numbers
One prompt, with the output columns named: find direct phone numbers for each contact, and create columns for phone number and phone status, found or not found.
A permission prompt appears the first time, asking to allow bulk phone enrichment. Choosing “always allow” stops it asking again on every subsequent batch.
Result: 31 direct phone numbers found, 88 not found, across 119 contacts. Roughly a 26 percent hit rate. Each mobile found costs 10 credits through the Mobile Finder.
The assistant explained the low number rather than leaving it unexplained: this is typical for senior finance executives, whose numbers are heavily guarded.
That explanation is worth as much as the data. A 26 percent return looks like a broken tool if you expect the 83 to 94 percent that owner-level and director-level lists produce. It is not broken. It is the correct answer for this audience.
What should you expect by seniority?
Phone coverage falls sharply as you move up the hierarchy at larger companies, and it is at its highest for owner-operators at small ones.
| Audience | Phone hit rate |
|---|---|
| SaaS executives, single state | 94 percent |
| Owners and CXOs, financial services, 51 to 200 staff | 86 percent |
| Medical practice owners, single metro | 83 percent |
| Senior finance executives, major city | 26 percent |
The pattern is about who controls the number. An owner-operator’s business line is their working number and appears in business records. A senior executive at a large institution has a number that is deliberately kept out of those records.
Credits are charged on success only, so the 88 misses cost nothing. On a 26 percent audience that cost model is the difference between a viable test and an expensive one.
Frequently asked questions
How do you find phone numbers for prospects? In three stages: build the company list, find the named people, then enrich the finished list through a connected data source. Keeping the stages separate means you can tell a genuine miss from a row that was never processed.
Should I let the assistant combine steps? No. When it offers to find people and contact data in one pass, decline. A clean list of people has to exist before enrichment, or an empty cell is ambiguous between “no number exists” and “this row arrived too late”.
Why are senior executive phone numbers so hard to find? They are deliberately guarded and kept out of public business records. This run returned 31 numbers from 119 senior finance executives, around 26 percent, against 83 to 94 percent for owner-level audiences.
Do I pay for phone numbers that are not found? No. Credits are charged on success only, so the 88 misses in this run cost nothing. That matters most on low-coverage audiences.
What is an MCP connector? Model Context Protocol is the standard that lets an AI assistant connect to an external tool and act on your behalf, rather than only answering questions. It is what allows the assistant to look up a real phone number.
What to do next
Three things. Companies, then people, then contact data, as three separate prompts. When the assistant offers to merge stages, decline. And judge a hit rate against the audience, not against a universal benchmark.
Run stages one and two only, and look at your list of people before spending anything. Once the names are right, run the phone enrichment and compare the hit rate against the table above before concluding anything about the data.