Find where a B2B SaaS company loses new-logo pipeline, and what to test first.
An open-source Claude Code agent that audits a B2B SaaS company from the outside in and names the three to five experiments most likely to add new-logo pipeline.
1What it finds
This is an excerpt of a teardown the agent produced from public sources only. Links to the full report follow it.
Outreach has rebuilt its story around agents, and its own site tells that story. None of the six third-party surfaces checked does. Search answers still describe a sales engagement platform and leave Outreach out when a buyer asks for an AI revenue platform.
Story sync: 5 of 16 buyer-facing surfaces. Owned 5 of 10. Third-party 0 of 6.
- Gartner Peer Insights still describes "Outreach Sales Execution Platform", and Crunchbase lists "a sales engagement platform" at outreach.io.Gartner Peer Insights, Crunchbase
- Salesloft and Clari merged and moved to one brand in September 2026. Outreach's comparison page says so, and its only offer is "Request a demo", which opens a form with 12 required fields.outreach.ai/salesloft-vs-outreach
- Search answers for "best AI SDR tools" and "best AI revenue platform for mid-market sales teams" did not include Outreach.Search-engine answers, US, directional
What I would run in the first 90 days
| Experiment | Pipeline metric | Review |
|---|---|---|
| Consolidation assessment for Salesloft and Clari accounts | Opportunities created vs a matched holdout | Week 10 |
| Carry the new story to review sites, Crunchbase and AI-tool roundups | First meetings from ICP accounts | Week 12 |
| A short assessment between the calculators and the demo form | Opportunities per form submission | Week 8 |
Read the one-page teardown or the full audit with sources. Outreach has not reviewed it.
2Why it starts with the ICP
A/B tests and landing-page changes move conversion rates on traffic a company already has. New-logo pipeline moves when the company changes who it goes after, what it tells them and what it asks them to do.
So the agent works from the top down. It settles the ICP before it looks at audiences, and it settles the offer before it recommends a channel. When an upstream layer is broken, it says so and stops recommending fixes that depend on it.
3How it works
The agent profiles the company by stage and sales motion, then scores six layers from 0 to 3. Each score cites the findings behind it.
| Layer | What it checks | A common leak |
|---|---|---|
| 1. ICP | Who actually buys and wins, compared with who the company says it targets. | The stated ICP is wider than the customers the company can prove it serves. |
| 2. Audiences | Who feels the problem, who signs, who blocks, and what each needs to believe. | Every page speaks to one persona while the rest of the buying group gets nothing. |
| 3. Messaging | Whether each audience hears the current story, with proof, on every surface it meets. | The company changed its story, and AI answers, review sites and ads still tell the old one. |
| 4. Offers | What each audience is asked to do at each stage, and whether it is worth their time. | "Book a demo" is the only offer, so buyers who are not ready have nowhere to go. |
| 5. Motion | Which channels carry the offer to the ICP, chosen after layers 1 to 4. | Channels were picked before the ICP and offer were settled. |
| 6. Measurement | Whether anyone could tell that an experiment worked. | Source and stage definitions changed, so the trend line cannot be trusted. |
It then groups findings by root cause, keeps the three to five that matter most, and writes one experiment for each. Every experiment names the pipeline number it should move, a review date and the result that would stop it.
It produces either a full report or a one-page teardown written for a CRO, a founder or a hiring manager.
4Rules it follows
- Every finding cites a source and the date it was observed.
- It never invents a metric or a benchmark. Unknown numbers are marked unknown, with the data that would answer them.
- It delivers three to five priorities. Everything else goes in a parking lot.
- It sizes every lift as low, mid and high, and plans on the low case.
- Every experiment has a review date and a kill rule.
- It observes only. It does not fill in forms, start trials or contact the company.
5Use it
Install the plugin in Claude Code:
/plugin install newlogo-growth --marketplace wpsands/newlogo-growthThen ask for an audit or a teardown:
Audit acme.comOne-page teardown of acme.comA company URL is all it needs. It gets more precise with SEO, enrichment and AI-visibility tools connected, and with a CRM or analytics export when you have one. The connectors list covers the options. The method itself is five plain files you can read and change.