Title: How RelayPoint Fixed Its AI Visibility and Doubled Pipeline with Abhord
Company overview
RelayPoint is a mid-market B2B SaaS platform that provides demand forecasting for CPG and food-and-beverage manufacturers. The company sells to operations leaders and revenue teams at firms with $50M–$500M in annual sales, integrating with common ERPs and data warehouses.
Engagement window
- Project start: October 7, 2025
- Measurement period: October 7, 2025 – January 15, 2026
- This is a refreshed edition with newly observed model behaviors and updated recommendations relevant as of January 27, 2026.
1) Initial problem
By late 2025, RelayPoint noticed that general-purpose LLMs rarely included the brand in responses to queries like “best demand forecasting software for mid-market CPG” or “predictive inventory tools for food manufacturers.” When the brand was mentioned, it was often misattributed as:
- A logistics brokerage defunct since 2018 (entity collision).
- A JavaScript library named “Relay” (partial-name confusion).
- A freight routing tool (incorrect category).
Symptoms quantified (baseline, week of October 7, 2025):
- Inclusion rate across four major LLMs for 15 intent queries: 6%.
- Correct entity attribution when mentioned: 58%.
- AI-assisted sessions to site (attributed by Abhord’s answer-layer tracking): 412/month.
- Pipeline influenced by AI-originating touches (last 90 days): $1.1M.
2) What Abhord’s analysis uncovered
Abhord ran an entity and evidence audit across RelayPoint’s public footprint and third-party references. Key findings:
- Entity collisions: Three similarly named entities dominated model memory. RelayPoint lacked a machine-readable “do-not-confuse-with” profile.
- Evidence gaps: High-signal facts (industries served, integrations, and customer logos) were scattered across PDFs and webinars, not easily digestible by models.
- Freshness decay: Core facts (pricing tiers, supported ERPs) hadn’t been updated in >300 days. Several models now down-rank stale claims in comparative answers.
- Source concordance deficit: Third-party descriptions varied; analyst write-ups used outdated positioning (“SMB inventory tool”), reducing confidence in model summaries.
- Weak answer-ready surfaces: Long-form thought leadership performed well in organic search but underperformed for LLM ingestion; there were few concise, verifiable snippets.
3) Optimization strategy implemented
Abhord and RelayPoint executed a four-part GEO/AEO plan over 10 weeks.
A. Canonical entity disambiguation
- Published a machine-readable Entity Card (JSON-LD + OpenAPI fragment) with: official name, aliases, “not-us” entities, industry tags, integration list, founding