Why we name signals.
Signals are the inbound data points that drive every decision about what to work on next. Strategic direction, customer feedback, support tickets, market research, observability data, financial data, AI-agent telemetry — all of it is signal.
Traditional agile doesn’t name signals explicitly. Backlogs fill, retros run, customers are “in the loop,” and the signals are assumed. Cadence calls them out by name because in an AI-heavy org, the volume of inbound data is too high to treat implicitly.
The principle
If the org doesn’t name a signal, it can’t curate it — and it can’t turn it into work.
Naming signals up front makes them visible. Visible signals can be inspected, instrumented, and routed. Unnamed signals get absorbed into individual judgment, and that’s where blind spots live.
Five categories.
Every signal we care about falls into one of five categories. Orgs add their own, so the taxonomy isn’t exhaustive, but these five are the minimum vocabulary for a healthy signal pipeline.
Category · Nº 01
Strategic
- Align directives
- Market research
- Competitive intel
- Industry shifts
Category · Nº 02
Customer
- Direct feedback
- Help desk tickets
- Usage patterns
- Churn reasons
Category · Nº 03
Operational
- DevOps observability
- Bug rates
- Incident data
- Deployment metrics
Category · Nº 04
Financial
- AI cost
- Cost-to-serve
- Revenue impact
- LTV / CAC
Category · Nº 05
AI capability
- Agent effectiveness
- Evaluation scores
- Automation lift
- Organic agentic research
Category Nº 01
Strategic signals
Where the org is choosing to point its energy. These come down from the Align lane and across from anyone watching the wider market.
- Strategic direction from Align. Stated priorities, OKRs, deliberate bets. The Align lane is the canonical source.
- Market research. External research the org commissions or absorbs from analysts, consultants, and industry sources.
- Competitive intelligence. What other companies are shipping, pricing, and positioning against.
- Input from Enable. Capability shifts: a new model is dramatically better at something, a vendor is deprecating a tool, a pattern is becoming standard.
Category Nº 02
Customer signals
What the people paying you are saying and doing. Customer signals are a high-value input that most orgs underuse.
- Direct feedback. Interviews, surveys, NPS comments, social media, sales-call transcripts, anything the customer says in their own words.
- Help desk and support. Tickets, chat logs, escalations. One of the richest records of what’s actually broken, and usually the least mined.
- Usage patterns. Product analytics, feature adoption, drop-off points, power-user behavior, dead features.
- Churn reasons. Why customers leave. Often the highest-signal data the org has, and often the least systematically captured.
Category Nº 03
Operational signals
How the system is running. Most engineering orgs already track this category, but in Cadence we treat operational data as product input, not only engineering telemetry.
- DevOps observability. Logs, traces, metrics across the production stack. Pod-owned and continuously instrumented.
- Bug signals. Defect rates, defect age, regression frequency. Bug data measures quality, and it also tells you which areas of the product are friction points.
- Incident data. Incident count, MTTR, post-mortems. Each post-mortem feeds the signal library directly.
- Deployment metrics. DORA-style: deployment frequency, lead time for changes, change-fail rate. These tell you whether the system itself is healthy.
Category Nº 04
Financial signals
Cost and value. AI orgs that don’t track this category wake up with a model bill that destroys their unit economics.
- AI cost. Compute, vendor, license, evaluation, fine-tuning. Tracked per pod, per workflow, per agent — not as a single line item.
- Cost-to-serve. What it costs to deliver the product to one customer for one month. AI changes this number significantly in both directions.
- Value delivered. Revenue impact, retention impact, expansion impact tied to specific shipped work. The pod knows whether what it shipped moved the numbers.
- Unit economics. LTV, CAC, gross margin, contribution margin. The signals that tell you the business is sustainable, not just busy.
Category Nº 05
AI-capability signals
How well the AI is actually working. This category exists because AI agents fail in shapes traditional software doesn’t: silent regressions, drift, prompt injection, quality erosion. Those failures have to be measured deliberately.
- Agent effectiveness. How often the inbound and outbound agents produce useful output. How often pods accept what the agents propose.
- Evaluation scores. Continuous evals against representative tasks. Trends matter more than absolute numbers.
- Automation lift. How much of a pod’s output is AI-assisted vs. fully human. Not a leaderboard metric — an architecture signal.
- Organic agentic research. Research and insight surfaced by agents themselves — competitive scans, customer-feedback synthesis, opportunity detection. The agents work as additional researchers, not only as executors.
Curation: signal → product brief.
Naming signals is step one. The harder step is turning that flow of raw data into something the org can act on. That’s curation, and at the volume a Cadence org generates, it doesn’t happen without effective agentic help.
What curation does
- Cluster. Group related signals across categories — a customer complaint, a support ticket trend, a usage drop, and an error rate spike are often the same story.
- Summarize. Reduce volume to what a human can act on without losing the evidence trail back to the raw signal.
- Score for value. Estimate the business impact and the cost-to-act, so Align and POs can prioritize on signal, not on whoever talked loudest.
- Shape into briefs. The output of curation is product briefs: well-shaped, evidence-backed proposals that feed the Signal Inspection event (covered in the Events chapter).
Why agentic help is non-optional
A modest-sized Cadence org generates more signal in a week than any human can read, let alone curate. The choice isn’t between “humans curate” and “agents curate.” The choice is between “agents curate, humans direct” and “most signal goes uncurated.” Most orgs default to the second answer without knowing they’re defaulting at all.
Agents process volume; humans read the curated picture and make the call. The Product Owner role is the human side of this partnership. The signal library is where the agent side accumulates context.
The signal library.
As signals get curated, the patterns they reveal accumulate. The signal library is where that accumulation lives — the org-level equivalent of the pod-level libraries described in the Pods chapter.
What it stores
Curated context, not raw streams.
Themes, recurring patterns, customer-segment insights, operational hot spots, post-mortem learnings, value-delivery evidence. It’s the synthesis layer above the raw data warehouse.
Who uses it
Align, Product Owners, agents.
Align references it for strategy decisions. POs draw from it to shape briefs. Curation agents read from it and write back to it, so every new signal updates context and improves what the next pass surfaces.
Pod libraries vs. the signal library
The two libraries serve different scopes and shouldn’t be merged. Pod libraries are deeply local: what a pod has built, decided, and learned in its own codebase. The signal library is deliberately org-wide, the cross-cutting picture POs and Align need to make portfolio-level calls.
The signal library is the org’s memory of why it cares about what it cares about.
Without it, every new product brief starts from a blank page, re-litigating what the org already learned. With it, a brief arrives carrying its evidence, the history of related decisions, and a connection to current strategic direction.
The end product of all of this, named signals curated with agents and accumulated in the library, is a steady stream of product briefs aimed at the org’s health and growth. That’s where the Events chapter picks up, with Signal Inspection.