The demo shows what's possible when your sales signals are synthesised in one place — objections clustered, conversions ranked, intent flagged. Here's how to get something similar yourself, and what it looks like to build the real thing.
Sam Tasman-Jones·~15 min read·Claude Projects or Claude Code
The demo is a GTM intelligence dashboard — a single view that turns your raw sales conversations into structured signals. Three columns: what's converting and why, what objections are surfacing and how often, and who in your pipeline is showing active buying intent.
In the demo it's all hardcoded. In a real version, this gets rebuilt every time new conversations, notes, or CRM data come in. The output is the same shape — just live.
Below are two ways to get something like this for yourself: a quick snapshot you can pull together in 30 minutes with no code, and a starting point for building a live version of your own.
⚡ Option 1 — No code
Get a snapshot in 30 minutes
This uses Claude Projects to analyse your existing data and produce a structured point-in-time read of your GTM signals — the same three sections as the dashboard, delivered as a report rather than a live view. You can take the same approach in ChatGPT, Cursor, or any AI assistant that accepts file uploads. If your platform doesn't support persistent project context, paste your data files and instructions directly into the chat at the start of each session — you'll need to do that each time, but the output is the same.
1
Pull together your raw data
Collect what you have — call transcripts (Zoom, Teams, or any transcription tool), discovery notes, CRM deal comments, Slack threads about deals. Plain text or PDF works fine. Even rough notes are useful. Aim for at least 5–10 sales conversations.
2
Create a Claude Project and set the context
Create a new Project in Claude.ai. Upload your files. In the Project instructions, paste this — it sets the frame for everything that follows:
Project instruction
You are a GTM intelligence analyst for [your company / product name]. Your job is to turn sales conversations into a structured intelligence report that mirrors a live dashboard — three sections that give a complete read of what's converting, what's blocking deals, and who to prioritise next.
When asked to analyse data, always produce output in this exact structure:
── SECTION 1: CONVERSION SIGNALS ──
• Converting language: A tag-style list of specific phrases, moments, and themes that appear in sessions that progressed or closed (e.g. "saves 2 days per review", "live demo moment", "story-first workflow").
• What isn't converting: A separate tag list of language from sessions that stalled or churned.
• Lead sources: Break down where leads are coming from and which sources convert at the highest rate.
• What's working — ranked: 2–3 specific tactics or assets that appear most often in winning sessions. For each: a name, what makes it work, and suggested messaging.
── SECTION 2: OBJECTION THEMES ──
For each major objection cluster (4–6 themes), output:
Theme name | Count | % of total
Verbatim quote — the clearest example from the conversations
What's really driving this: 2-sentence analysis
Recommended counter: the most effective response based on what worked
Who raised it: list of initials or first names
── SECTION 3: PIPELINE PRIORITISATION ──
• Buyer profiles: 2–3 patterns of who is engaging and how they convert. Name the pattern, describe the signals.
• Active buy intent: Specific prospects showing strong signals — what they said, what they asked, what the next move is.
• Priority follow-up list: Ranked, with a one-sentence reason for each.
Be specific throughout. Pull direct language from the conversations. Do not generalise or pad with generic advice. If you lack data for a section, say so clearly.
3
Run the analysis
With your files uploaded, start with this prompt — it gives you the full picture in one pass:
Analysis prompt
Analyse all the conversations I've uploaded and produce the full GTM intelligence report using the structure in your instructions.
For Section 1 (Conversion Signals):
— Tag list: what specific words, phrases, and moments appear most in sessions that converted or progressed? Keep each tag to 2–5 words. Give me a separate list for what's NOT converting.
— Source breakdown: can you identify where these leads came from (LinkedIn, referral, event, inbound, direct)? Which sources are converting at the highest rate?
— Ranked plays: identify the 2–3 specific moments or tactics that appear just before a positive outcome. Name each one, explain what makes it work, and write 3 bullet points of messaging I can use.
For Section 2 (Objection Themes):
— Cluster all objections into named themes. Aim for 4–6 clusters.
— For each: theme name, total count, percentage of all objections, the single best verbatim quote, a 2-sentence read on what's really driving it, and the most effective counter based on what's worked in these conversations.
— Order by frequency, highest first.
For Section 3 (Pipeline Prioritisation):
— Describe 2–3 buyer profile patterns — the type of person or company that engages and converts versus those that stall. Name the pattern and describe the signals.
— Flag any specific prospects showing strong buying intent: what did they say, what did they ask, what should I do next with each of them?
— Give me a ranked follow-up list for this week — who to contact first, and one sentence on why each is a priority right now.
Be specific throughout. Quote the conversations directly where possible.
Then drill in with follow-ups like "what's the best response to the pricing objection?" or "which deals are at risk and why?"
4
Use the output
You now have a structured read of your pipeline — paste it into Notion, share it in Slack, or use it as the brief for your next sales team meeting. Run it again next week with new conversations added to the Project. The snapshot gets sharper each time.
⚙ Option 2 — Full build
Build your own live version
If you want this as a recurring, live dashboard — not a one-off analysis — you need to build it. It's achievable with Claude Code or a similar AI coding tool if you're comfortable getting into it. The rough architecture looks like this:
Single HTML file (like this demo)Notion database viewInternal web app
↓
Step 5 — Delivery
Push the output where your team lives
Slack weekly digestEmail summaryShared Notion pageLive link
The demo also has a chat layer — the Signals Assistant that lets you ask plain-language questions against your data ("who should I follow up today?", "what counters the pricing objection?"). This is built as a lightweight Claude API wrapper that queries your structured data store. It's an optional layer on top of the dashboard — worth including in your build plan if you want it, as the architecture is slightly different from the static dashboard render.
The nuance in any build is steps 1–2: how your specific data is structured and where it lives will shape the whole pipeline. The prompt below describes the full target output in detail and tells Claude to gather that information from you before writing a single line of code.
Starting prompt — paste this into Claude Code or Claude.ai to kick off your build
I want to build a live GTM intelligence dashboard — a single-page web application that synthesises my sales signals on a recurring basis. Here is the exact output I am trying to build:
THE TARGET OUTPUT — three-column dashboard:
Column 1 — Conversions:
• A tag cloud of specific phrases from sessions that converted, and a separate tag cloud for sessions that didn't
• A conversion rate trend chart (line chart, % of sessions converting over time)
• A source breakdown bar chart (where leads come from, which sources convert best)
• 2–4 "conversion tool" cards ranked by conversion rate — each showing a tactic name, rate, messaging bullets, and a task checklist
Column 2 — Objections:
• A tag cloud of language from sessions that stalled or churned
• 4–6 objection theme cards — each showing: theme name, frequency count, a verbatim quote, 2-sentence body, which conversion tools counter it, and who raised it (initials strip with hover tooltips)
Column 3 — Pipeline:
• 3–4 persona/buy signal cards (buyer profile patterns + named prospects with active intent signals)
• An input feed showing recent data sources with timestamps (Slack, Drive, transcripts, etc.)
Plus an optional chat layer — a sidebar panel powered by the Claude API where users can ask plain-language questions ("who should I follow up today?") and get answers grounded in the structured data.
THE ARCHITECTURE I want to build:
1. Data collection — a scheduled script (or manual export trigger) that pulls from my data sources and saves as clean, structured text files
2. Claude synthesis — a Claude API call that processes the raw data through a structured prompt and outputs a JSON file matching the dashboard schema (one object for each of the three columns)
3. Dashboard render — a self-contained HTML/CSS/JS file that reads the JSON and renders the three-column view with charts (Chart.js) and interactive card components
4. Chat layer (optional) — a lightweight Claude API wrapper that accepts plain-language queries, passes them to Claude along with the structured JSON as context, and returns answers in the dashboard sidebar
5. Delivery — static hosting (Vercel or similar) or a local file opened in browser
Before writing any code, ask me a series of questions to understand my specific setup:
- What data sources do I have and in what format (call transcripts, CRM notes, Slack exports, email threads, discovery notes)?
- Do I have API access to any of these sources, or will I be working with manual exports?
- How often do I want the data refreshed — daily, weekly, or triggered manually?
- What tech stack am I comfortable with, and where will this be hosted?
- Do I have a Claude API key? Any other API keys relevant to my data sources?
- How many people will use this, and how will they access it?
- Do I want the chat layer, or just the static dashboard?
Once you have my answers, write a tailored plain-English build plan covering: data pipeline design, JSON schema structure, dashboard component breakdown, and chat layer approach. No code yet — I want to understand the architecture before we build anything.
Fair warning: Claude Code will get you most of the way there, but the last 20% — connecting to your specific CRM, handling your transcript format, deploying somewhere your team can access — tends to have nuance that depends on your setup. That's where having someone who's done it before helps.
Going further
→
Add a Slack digest — pipe the weekly synthesis output into a #gtm-signals channel so your whole team sees it without opening another tool.
→
Score and route automatically — once the intent signals are structured, it's a short step to auto-tagging high-intent prospects in your CRM or triggering a follow-up task.
→
Layer in win/loss data — tag each conversation with its outcome before uploading. The signal quality improves significantly when Claude knows which sessions converted and which didn't.
Want this built for your actual system?
The prompt above will get you started. If you hit the wall where your specific data setup, stack, or delivery requirements make it complicated — I've built this kind of thing before and I'm happy to help scope it properly.