AI Web research is one of the most exciting features we've ever released on Gumloop. It's the magic many users have been waiting for. Simply prompt the internet and get the answers you want returned to you as a step in your workflow.
"Is x company GDPR compliant?" "what's x person's job title?" "what's x company's linkedin url and follower count?". AI will browse the web, digging into websites and fetching you the answers you want, directly via Gumloop.
For those who already use Gumloop, imagine if our Perplexity node and AI Data Extraction node had a baby with even better performance.
How does it work
The node takes a natural language prompt and browses the web similar to how a human would, visiting websites and analyzing content until it feels like it's found the answers your looking for or hit a dead end.
All you need to do is define the search prompt. For example, if you want to find whether a company is SOC2, HIPAA and GDPR complaint you can specify that prompt and pass in 'company name' as the dynamic value. Any company name you then pass in will kick off a search of the internet.
You can then loop this AI Web Research node over 500 company names and enrich all that data!
(It's not using perplexity btw. It's something new ✨)
Why is it important
It's one of Gumloop's first agentic features and is the first time you don't need to know exactly how to accomplish the task in order to automate it.
Previously for this sort of research work, the Gumloop user would have to know exactly what website to scrape, how to traverse it and even then it might not be feasible once multiple scrapes are involved. Now, open ended web research is a single prompt away.
It unlocks millions of new use cases on the platform.
Use cases
PERSON RESEARCH: pass in a person's name, job title, company, first validate that the info is correct, then research that person's board memberships, public posts/announcements/speaking engagements, and whether they recently changed jobs. This flow really demonstrates how the integration can be used to not only enrich rows, but also validate rows + fix errors! (flow here)
DEEP STARTUP RESEARCH: provide a startup description + target venture fund name --> research the investment thesis and determine if there's a fit --> if there is a fit, extract information about the best partner at the firm to cold email (flow here)
COMPLIANCE CERTIFICATION RESEARCH: take in a csv of companies (company name + company website) and enrich each row with information on what compliance certifications that company mentions, whether they have a compliance leader, the linkedin of the compliance leader, and if there's any news or audits published about recent compliance issues the company has had. (flow here)
HIRING RESEARCH: take in the name + website of a company, return whether they are hiring, and if they're hiring highlight what kinds of roles they are hiring for + job board postings (flow here)
10-k PDF EXTRACTION: extract the link to the company's most recent 10-k. then extract key financial metrics – their revenue, fiscal year challenge(s), total number of employees, M&A transactions they've been involved in. This flow demonstrates how the node can be used for some pretty specific financial numbers + reading complex pdfs like 10k's. (flow here)
NICHE BUSINESS RESEARCH: takes in a solo practitioners website, highlights the treatments they provide, any reviews they have, and whether they accept telemedicine visits. This how the node can be used for smaller/niche businesses (gyms, dentists, hoteliers, etc.) (flow here)
REAL ESTATE INVESTMENT OPPORTUNITY RESEARCHER: Provide a geographical area and get in depth research about neighbourhood housing markets. I really like this one because it shows how the node can be used to generate a few options, then in a second node, enrich/research each option in detail. (flow here)
INVESTOR FIT RESEARCHER: go from startup name + target investor --> figuring out a fit --> if a fit exists, choose the best email to cold email and actually draft a complete cold email to that investor with all the right details. I love this one because it's two nodes with various steps of reasoning, and a useful output in the end. (flow)
More Info
Check out our docs here for more info about exactly how to leverage this new node.