TL;DR
Built the Competitor Insights Agent, a Notion AI agent that pulls competitor signal from Slack, public Notion pages, and the open web to answer on-demand questions about market players, profile competitors, compare features, and support sales enablement. Follows the same template as the VoC Agent, refocused on competitive analysis. Covers Boulevard, Mango Mint, Zenoti, Vagaro, and the long tail. Available in Notion and via the #competitor-insights-agent Slack channel.
The Problem
Competitive intelligence at Phorest was opportunistic, someone would notice a Zenoti AI launch, post in Slack, and the rest of us would hope it got captured somewhere durable. With Zenoti driving sharp customer loss in early 2026 and AI Voice/AI Scribe shifting the perception battle, that wasn't good enough.
- Different teams need different lenses. Product cares about features, GTM cares about messaging, Strategy cares about positioning, no shared base to query.
- Public knowledge was fragmented. Notes from sales calls, Slack threads, research pages, and competitor websites lived in five different places.
- No on-demand answer for sales. Reps in
#medi-sales-check-inwere getting hit live with "but Zenoti has X" objections with no fast way to get a structured response. - Web scanning was a side project, not a system. Individual ICs were doing great work in isolation, but there was no shared agent the company could lean on.
The Solution: VoC Template, Competitive Lens
The Competitor Insights Agent reuses the VoC Agent's architecture, same source-plus-synthesis pattern, pointed at competitive sources instead of customer ones, with web access added.
Connected sources
| Source | What it covers | How it's connected |
|---|---|---|
| Notion (public pages) | Competitor research, SWOTs, concept pitches, prospect insights, deal-loss notes | Native |
| Slack | #competitor-intelligence, sales check-ins, churn-risk channels | Notion ↔ Slack |
| Web | Competitor websites, blogs, product pages, help centres, industry publications | Notion AI web access |
Two modes of use
- Discovery mode, "How does Zenoti position AI Voice vs. our Front Desk AI?", "Which competitors offer embedded finance?"
- Monthly report mode, competitor moves: launches, pricing changes, partnerships, marketing pushes, perception shifts.
Key Challenges
Web vs. internal source weighting
The web is noisy. Lean too heavily on it and every answer regurgitates the competitor's marketing site. Instructions weight internal Notion and Slack first (where Phorest's own analysis lives), then use the web to fill gaps and verify recency.
Avoiding "so what" hallucinations
Models are great at compressing competitor pages into bullets. They're poor at the strategic "what does this mean for us" call. The agent's instructions explicitly stop at synthesis and cite the source, humans own the implication.
Source rot
Competitor sites get redesigned, URLs move, pages disappear. The agent had to be tolerant of broken links and prefer recently-cited sources over older ones.
Reusing the VoC template
The biggest unlock was not building this from scratch. Same instruction shape, same report structure, same Slack-channel distribution pattern, just swapped sources and tone. Building the second agent took a fraction of the time the first one did.
What the Agent Can Do
| Capability | Example question |
|---|---|
| Competitor profiling | "Give me a detailed overview of how Zenoti is approaching their AI feature push" |
| Feature comparison | "Which competitors offer embedded finance (invoicing, payroll, BaaS)?" |
| Positioning analysis | "How do competitors position their WhatsApp offering?" |
| Sales enablement | "Client said Zenoti has AI Scribe, what's their actual offering and our counter?" |
| Market sweeps | "Notable competitor launches, pricing changes, or partnerships in the last 30 days" |
| Deal-loss patterns | "Which competitors show up most in churn-risk and lost-deal notes this quarter?" |
What We Learned
- Templates compound. Reusing the VoC Agent's shape made this cheap to build and consistent to use. Two agents, one mental model.
- Internal context beats web context. Phorest's own deal notes and Slack threads are usually richer than the competitor's public site.
- A dedicated Slack channel is the right surface for sales. Reps don't open Notion mid-call; they ping a channel.
- Citations or it didn't happen. Every claim links back to the source page or URL.
- Discovery + scheduled reports is the same winning shape as VoC.
- Cover the long tail via config, not code. New competitors are added by updating the source list and instructions, no rebuild required.
Tech Stack
- Notion AI Agents, orchestration, instructions, web access, Slack distribution
- Notion ↔ Slack integration,
#competitor-insights-agentand related channels - Public Notion pages, internal research and deal-loss notes as governed input
- Web, competitor sites, help centres, industry publications