What is multi-hop pathfinding?
Multi-hop pathfinding is the core algorithm behind warm-introduction software. It traverses a company's relationship graph from the user's node, through intermediary contacts, to a target person — surfacing every possible route and ranking them by likelihood of conversion.
A one-hop path is direct: you know the target. A two-hop path goes through one intermediary: a colleague who knows the target. A three-hop path is friend-of-friend-of-friend. Pathfinding software typically searches up to 3 hops; beyond that, the trust transfer degrades sharply and the routes stop converting.
The technical problem is straightforward — graph traversal is a textbook computer science problem — but the production implementation involves dozens of decisions about scoring, ranking, filtering, and routing that determine whether the surfaced paths actually convert.
Why pathfinding matters for B2B sales
Two reasons:
The graph is too big for humans. A 50-person company sits on roughly 100,000 LinkedIn connections, 200+ customer champions, and dozens of board and partner ties. Searching this manually for routes to a specific target is impossible. Software has to do it.
The strongest path isn't obvious. Multiple paths exist for most targets — but only one or two convert. A path through your CEO with a strong tie converts at 40%; a path through an intern with a weak LinkedIn connection converts at 1%. Without ranking, reps default to whatever they find first — usually wrong.
Pathfinding solves both problems: it scales to the full graph, and it ranks routes by conversion likelihood.
How pathfinding algorithms actually work
Three core steps:
Step 1: Graph construction
The system ingests relationship data — email metadata, calendar events, LinkedIn connections, CRM contacts, partner directories — and builds a graph where each contact is a node and each known relationship is an edge.
Each edge gets metadata: when was the last interaction, how frequent are touches, what's the depth (was it a single email or six months of meetings), what's the source (LinkedIn, email, calendar). This metadata becomes the input to the scoring function.
Step 2: Path enumeration
When the user queries a target, the algorithm searches the graph for every route from the user's node to the target's node, up to 3 hops. For a well-connected user querying a well-connected target, this might return 50–200 candidate paths.
The enumeration is fast — modern graph databases handle this in milliseconds across millions of edges. The challenge isn't enumeration; it's ranking.
Step 3: Scoring and ranking
Each path gets a composite score based on:
- Tie strength of each edge — How recent, frequent, and deep is each connection in the path? Weak edges produce weak paths.
- Path length — Shorter paths transfer trust better. A 2-hop path beats a 3-hop path almost always.
- Recency of the path overall — A 3-year-old chain is colder than a 3-month-old chain.
- Shared context — Did the intermediaries share an employer, board, or event with the target? Shared context boosts conversion rate by 2–3x.
- Policy fit — Is this ask allowed under the company's governance rules? (More on this below.)
The output is a ranked list — usually presented as top 3 paths — that the rep can evaluate and act on.
Why path length matters more than people expect
Conversion rates by hop count:
- 1 hop (direct): 40–60% reply rate. The rep already knows the target.
- 2 hops (one intermediary): 30–50% if intermediary is strong, 5–10% if weak.
- 3 hops (two intermediaries): 10–20% if both intermediaries are strong, near-zero if either is weak.
- 4+ hops: Marginal. Trust transfer degrades to the point where outcomes converge with cold outreach.
The math forces a design choice: most warm-intro software caps pathfinding at 3 hops. Going further produces noise without yield.
Where naive pathfinding fails
Three common failure modes:
The "shortest path" trap. The algorithm finds the fewest-hops path and surfaces it — but the shortest path runs through an intermediary with a weak, stale tie. The rep makes the ask, the intermediary ignores it, and the relationship gets damaged. Strength scoring has to outweigh hop count in ranking.
The "popular intermediary" overload. Your CEO is on 50 paths. The algorithm surfaces them all and reps queue up asks. The CEO becomes a bottleneck, gets ask-fatigued, and stops responding. Pathfinding has to model intermediary cadence limits.
The "ignored governance" mistake. The algorithm finds a path from an SDR to a customer's champion. Surfacing this path tempts the SDR to reach out directly — which burns the CS team's trust. Governance rules have to filter paths before they're shown.
What separates production pathfinding from demo pathfinding
Five things to look for in a production system:
- Strength scoring trained on actual outcomes. The system learns which path features predicted conversion in past intros and reweights accordingly.
- Intermediary cadence modeling. The system knows that your CEO can be asked for at most 6 intros per quarter and routes accordingly.
- Governance rules baked in. Customer intros route through CSMs, partner intros through partner managers, board intros through the founder.
- Multi-pillar source data. Pathfinding through team + customer + board + partner produces 5–10x the warm-path supply of pathfinding through team only.
- Closed-loop learning. Outcomes write back to the graph, improving rankings over time.
Tools that hit all five become orchestration layers. Tools that hit one or two are graph viewers.
Boomerang's approach to pathfinding
Boomerang searches across the four-pillar relationship graph (team, customer, board/investor, partner), ranks paths by tie strength and policy fit, applies intermediary cadence limits and governance routing, and writes outcomes back so the system learns which paths converted. The result: surfaced paths that reflect not just graph topology but the reality of which routes actually produce pipeline.
The pathfinding algorithm matters. The four-pillar source data matters more. And the governance and learning loop matter most — those are what turn a graph search into a working warm-intro motion.
For teams evaluating warm-intro software, the question to ask isn't "does it find paths?" — every tool in the category does that. The question is whether it finds paths that convert.
