# Reasoning across your network Source: https://docs.havilo.ai/concepts/asking Consider the following query: > Plan an event for our startups with Turkish founders, ideally founders who have had at least one previous exit. Identify the problems those startups have in common, find a few mentors whose expertise would actually help, use my Event Invite template to prepare invitations for everyone, and prepare an updated Quarterly Founder Outreach metrics document for our LPs. A person could do this. But what would it actually take? They would have to work through every startup and founder in the network. They would inspect names, biographies, email addresses, phone numbers, Notes, LinkedIn profiles, Crunchbase records, and any other evidence that might establish who actually qualifies. Then they would read the pitch decks, investor reports, company updates, meeting notes, and correspondence for every qualifying startup—not just to summarize them, but to understand what each company is struggling with and find the problems they share. From there, they would do the same work across the mentor network: study each person's experience, judge whose expertise maps to those shared problems, prepare the right invitations from the existing template, and carry the final results into the LP report. That is precisely how Havilo works. It can follow the same evidence, preserve the same distinctions, and apply judgment at every step—without the limits on time, attention, or scale that make an investigation like this impractical for a person. ## Understanding what you mean Natural language criteria are rarely as simple as a field and a value. “Turkish founder” describes a person, not automatically the company they founded. Relevant context might appear in explicit biographical information, places someone was born or has lived, language, contact details, their name, education, or professional history. Those signals do not all mean the same thing, and no single one has to be treated as proof. A company's headquarters, customers, employees, or other founders may support the conclusion that the company is Turkish, but they do not establish that every founder is. The reverse is also true: a Turkish founder does not necessarily make the startup Turkish. Havilo preserves the direction of the relationship instead of substituting whichever nearby attribute is easiest to find. “Ideally” matters too. A previous exit is a preference, not a requirement that should eliminate every otherwise relevant company. Havilo interprets both what a criterion means and how strictly it should be applied. When wording admits materially different interpretations and the choice would change the work, Havilo can ask rather than quietly assume. This is not a special rule for one nationality or one query. The same care applies whenever a request depends on experience, intent, fit, reputation, timing, or another idea whose evidence is distributed across the network. ## Connected context The answer to a question can draw from entity fields, relationships and their roles, Notes, dossiers, meetings, correspondence, public research, uploaded files, and earlier work. Havilo uses those sources together without pretending they are interchangeable. A declared field can establish a known fact. A relationship can show who founded, advised, employed, or invested in whom, including when and in what role. A Note or meeting can describe a problem that was never added to the data model. Public research can fill a gap or show that something has changed. This is why the event request can move from founders to their startups, from the startups to their shared problems, and from those problems to mentors with relevant experience. Each transition retains the entities and relationships that give it meaning. ## Search, analysis, and matching The AI Engine can retrieve known information, investigate a criterion that requires judgment, compare candidates, discover patterns across a group, and explain why a result fits. Matching is grounded in the target rather than a universal score. The right mentor depends on what a company needs now, what the mentor has actually done, whether that experience transfers, and sometimes whether the fit makes sense from both sides. The strongest result may also depend on an existing relationship or a credible path to an introduction. The same request can combine these forms of work. Havilo can first find the qualifying startups, then identify the problems they share, then look for mentors whose experience addresses those common problems. It does not have to collapse the entire request into one keyword search. ## Meaning at network scale The meaning of a criterion should not disappear as the candidate pool grows. Havilo can apply the same interpretation across a list, a cohort, an uploaded dataset, or a network containing millions of entities without reducing every person or company to the same generic summary. Some requests need a broad scan followed by close analysis of a smaller set. Others compare every member of one group with every member of another. Uploaded CSV rows can participate without first becoming permanent entities, while reliable matches can still connect a row to knowledge already in the workspace. ## Work you can use The result can be a direct answer, a table, a document, or a set of prepared emails. A saved Note can serve as a template, allowing Havilo to preserve its language and fill the intended details for each recipient. A later request can refine the candidates, work from an earlier result, or turn selected rows into another piece of work. A prepared email is a real draft, not a preview. Havilo saves it into the mailbox you send from, so it waits in Gmail or Outlook beside everything else you are writing. You can change it in either place and the other keeps up. It goes nowhere until you send it. When more than one of your mailboxes can send, Havilo waits for you to choose which. With no mailbox connected that can hold a draft, Havilo still writes a new email and keeps it in Havilo until you connect one. A reply needs the mailbox the conversation is in. Havilo does not stop at describing how someone could complete the event request. It can produce the candidate set, the reasoning, the invitations, and the updated reporting document in the same connected system. See [Artifacts](https://docs.havilo.ai/concepts/artifacts.md) for the formats Havilo produces and [Provenance and history](https://docs.havilo.ai/concepts/provenance-history.md) for how facts and analysis remain distinguishable. ## Ask about a SharePoint document [Connect SharePoint](https://docs.havilo.ai/guides/sharepoint.md) explains administrator approval and library selection.