Finding Trades by Browser Tab and Phone
Every new build or renovation needs subcontractors, architects, engineers, electricians and plumbers in one area and one window. A project manager assembles the list by hand: a license registry in one tab, an association directory in another, review sites in a third. The same contractor turns up several times under slightly different names. A license is checked by pasting a number into a search form, and insurance is taken on trust.
Then come the calls and texts from personal phones. Who promised a portfolio by Thursday and who asked for a call back lives in a notebook. When the manager is away, the follow-ups stop.
Gaps a Directory or a Generic Bot Leaves
- A directory is a list, not a shortlist. It shows who exists, without saying who holds an active license and insurance and is free next month.
- One person, many records. The same contractor appears under a legal name, a trading name and the owner's name. Without entity resolution the list multiplies and the same person gets called again and again.
- Collected data is trusted blindly. Sources contradict each other and go stale. A field with no source and no confidence cannot be believed, and nobody notices when it changes.
- A chatbot is not outreach. A generic bot writes a message but does not know what the profile is missing or what the person promised.
- Automation acting alone breaks trust. A profile overwritten from an ambiguous reply costs more than the manual work it was meant to replace.
From Project to Shortlist
Each step writes into the same candidate profile, so the next one starts from what is already known.
- Sourcing. Agents search license registries, association directories and review platforms by trade and area. They respect robots.txt and terms of use, and back off on captchas and rate limits.
- Profile. LLM extraction fills a fixed schema, with OCR for scanned documents. Duplicates merge on license number, phone, website and email. Every field stores its source, a timestamp and a confidence score.
- Match. When a project is created, the manager gets a deduplicated list ranked by fit, with the factors explained. Hard filters are data: an active license, required insurance, the service area. A candidate who fails them is left off the list.
- Outreach. An AI agent contacts the candidate within working hours in their time zone, names the company and says where the contact came from. AI voice calls and automated texts go only to people who have given prior consent, as the US Telephone Consumer Protection Act (TCPA) requires, and an opt-out takes effect at once. The agent asks only for what the profile is missing and agrees a due date for documents. Pre-screening calls produce a transcript and a summary.
- Decision. Every message, call and promise lands in one timeline per contact. The manager reviews it and decides who goes on.
Technology Stack
Provenance and Outreach Rules
The pipeline gathers, checks and contacts; the project manager chooses who is hired.
- Every fact has an origin. When sources disagree, the alternatives are kept with their confidence instead of the last one winning.
- Low confidence means review. A low-confidence answer is shown to a person before anything is written in. Duplicates without a reliable key are flagged for review, and every automatic change is labeled as automatic in the history.
- Honest outreach. Messages go from company numbers rather than personal ones, and consent and opt-outs are recorded.
- Automation respects context. Nothing runs on archived or completed projects, and messages go only through connected channels.
Operating Limits
These limits are part of the pipeline's specification.
- Eight sources merged into one profile, each field with its evidence.
- Four outreach channels, with every contact in one timeline.
- No profile edit on low AI confidence: the answer goes to a person for review.
- No contact attempts outside working hours in the candidate's time zone.
- Reminders on a promised document the day before, on the due date and the day after.
Other Hiring That Works the Same Way
Companies that hire independent professionals with public but scattered qualifications run into the same friction:
- candidates are found by hand across registries, directories and review sites;
- licenses, certifications or insurance must be checked before work is awarded;
- outreach history leaves with the person who made the calls;
- promises and follow-up dates live in memory.
Facility maintenance firms, licensed healthcare staffing agencies, field service companies and logistics subcontracting all face it. The pipeline can be pointed at other trades, regions and sources, and connected to the company's own CRM and messaging channels. AI agents and human-in-the-loop in our glossary explain the building blocks, and our AI agent development page describes how we deliver them.
sources merged into one candidate profile
outreach channels across SMS, messenger apps and AI voice calls
profile edits made on a guess