How to Vibe-code a B2B contact finding app
This use case shows how a publisher vibe-coded a B2B contact finder that delivers emails and related structured data at scale.
Use this to find advertiser contacts by industry or build an entire B2B audience. Included is the tech used, specific prompts used, an outline of the final workflow, adjustments needed, and results.
Challenge
When NichePublisher decided to cut its audience of 12,000 in half – to remove anyone who had not opened recently and structure the data for every contact- it also needed a plan to replenish the audience by reaching out to new specialty publishers.
The size of the specialty publisher marketplace is difficult to estimate, but the best guess is around 20,000. So to arrive at a meaningful and dominant number of email subscribers – about 10,000 to 12,000 – meant finding 6,000 additional valid publishing contacts and structuring that data correctly.
After exhausting industry associations – including all state and national publishers associations – finding good contacts at scale would require use of AI and other tools.
There were other challenges:
- Automating a dupe check so new contacts did not repeat the current list.
- Data structure consistency.
- Mapping the search. With 30 categories to search, plus 120 possible sub-searches per category and five additional filters, the project required an organized search structure that a project manager could easily supervise.
- Ensuring that new contacts had the right job title and that emails were valid.
The offshore contractor had only limited familiarity with AI and would not be able to use it on their own at scale, keeping the data structured correctly with the right job titles.
Strategy
The publisher decided to try building an automated tool on a no-code app builder, Replit.
The original goal was to create a platform that both a contractor could easily use without errors and where the admin would have an approval process.
The requirements, or prompt, given to Replit were as follows:
a. You are building a contact finder for [name of industry] platform.
b. Any contractor should be able to log in with a password and perform searches by [list of] industry category and subcategory, and when finished, click submit, which goes to the admin for approval.
c. The admin will be able to view any list submitted, approve, and download by date range.
d. Data found will be structured consistently to include [ industry category & subcategory, [job titles], [job title types’, geo-location [national, local, global], format, and LinkedIn URL when exported. Do not add data outside of this list given.
Adjustments
Results
• Under the new workflow, the contractor found data for 10,000 high-level contacts over the course of two weeks.
• Cleaning in Neverbounce resulted in a 60% valid rate.
*Both Invalid emails from NeverBounce s and contacts without email at all were imported to Rocket Reach to find additional contacts, recovering about 10%.
*MailChimp under a different sender was used to test the list with a “warm letter.” This removed unsubscribes and hard bounces before adding contacts to the real ESP. The unsubscribe rate was about 1% on the first send, and under .3% afterwards.
•The process added about 6,000 valid new B2B C-level contacts, bringing the list number up to 11,000. Current sends have a 20 to 35% open rate and 2 to 4% CTR.
*Eventually, the admin was able to ask Replit to run the searches in sequence, removing the need for the contractor.
*Paid tokenization credits to run the app came to about $300.
