Market Analyst

Narrative

Avoma now provides structured AI notes alongside transcripts — summaries, action items, talking points. The question: does this data replace, duplicate, or complement our Claude extraction? A 5-call deep comparison answered definitively: complementary, not redundant.

Claude excels at insight quality — verbatim quotes with speaker attribution, interest scoring, positioning resonance. Avoma excels at deal mechanics — action items, decision authority, deployment risks. Avoma also misattributes team member quotes as positive moments and conflates features demoed with features requested. The integration plan feeds notes as supplementary context into the Claude prompt, letting Claude do what it does best with richer input.

Narrative

Sagetap calls come through a different channel than Avoma — they're analyst-mediated conversations with a different dynamic than direct sales calls. The pipeline needed to detect the source and route accordingly, since the external-call filter that works for Avoma would incorrectly exclude Sagetap calls.

The dashboard now shows source pills so the viewer can factor in the conversation context. An Avoma call with a prospect has different weight than a Sagetap call with an analyst. Making the source visible lets the reader calibrate.

Narrative

The analysis agent bridges the gap between structured dashboard views and open-ended questions. It loads the full dataset and can read raw transcripts for deeper context when a surface-level answer isn't enough.

Every answer cites specific calls — company name, date, and verbatim quotes. Sample size caveats are included automatically. The strategic goals in goals.md keep the agent focused on product-market fit decisions rather than generic summarization.

Narrative

The first three cycles built the pipeline and dashboard. This cycle asked: is the extracted data actually good? Walking through every page revealed display issues, but more importantly, extraction gaps. The prompt was generating insights from Qpoint speakers framing customer problems — not from customers themselves.

Urgency and positioning resonance were the biggest additions. Urgency distinguishes real pipeline from friendly interest — a 5-level scale with concrete/verbal/vague commitment tracking. Positioning resonance captures which pitch angles actually land with prospects, directly informing GTM messaging. Both require reading between the lines of conversation dynamics.

Narrative

The dashboard makes the extracted data explorable. Filtering by company, date range, interest level, and segment lets you slice the 79 calls in different ways. The README carries an important caveat: this is directional signal, not ground truth.

Several layers of distortion exist between reality and what the dashboard shows — conversation dynamics, AI extraction artifacts, small sample size, and selection bias. The dashboard acknowledges these limitations explicitly rather than presenting data as definitive.

Narrative

The pipeline ingests call transcripts from Avoma and processes them through Claude to extract structured insights — pain points, feature requests, objections, competitive mentions, interest levels, and company segmentation. The Batches API handles bulk extraction cost-efficiently.

Metadata sidecar support captures call date and duration alongside the transcript. A later patch (March 5) backfilled 15 of 18 calls that had null dates. The pipeline is deterministic: same transcript in, same structured JSON out.