AGENTIC SOCIAL INTELLIGENCE
Working
with Saga.
How to use Pulsar's AI analyst for real research — what it's for, how it thinks, and how to hold it to the bar.
Internal · June 2026
Working
Part 1 of 2
— Part one
What Saga is.
And why it changes the work.
01 · The problem we're solving
The problem
How much of your week
do you spend describing the data —
vs. interpreting it?
Saga isn't a faster way to retrieve. It's a way to skip the describing and start at the interpretation — verdict-first, baseline-aware, position-taking. The work analysts wish they had more time for.
02 · Why we called it Saga
The name explains the bet
Saga.
Norse goddess of storytelling. Companion of Odin. Drinker from the well of memory.
We're switching from selling data and tools to describe data,
to selling story and meaning extracted from data.
03 · What Saga is
Definition
A senior analyst — not a search engine.
Saga is
A senior analyst — verdict-first, comparative, calibrated.
Interprets data. Takes a position when the data supports it. Surfaces what the data can't tell you.
Saga is not
A search engine. A neutral hedger. A jargon-laden academic.
Not a generic LLM bolted on to dashboards. Not a flatterer. Not a recap bot.
04 · How Saga is different
Four things that make it different
Not a faster way to ask questions. A way to stop asking.
Four ways Saga separates from the AI-copilot pack — each with a way to prove it.
01
Brief-driven, not prompt-driven
Brief Saga once and it runs on its own clock — watching, escalating, pushing finished intelligence to you. You don't re-open it. You don't re-ask. It produces work while you're doing something else.
Prove itBrief Saga and a copilot the same way. Walk away 30 days. One produces a stack of analyses tied to real data movements. The other produces nothing.
02
On the data lake, not the dashboard
Every copilot in this category sits on top of pre-aggregated analytics. Saga works directly on the raw corpus — novel clustering, custom embeddings, statistical work no dashboard pre-defines. It does what the product doesn't ship.
Prove itAsk the same novel-clustering question. Saga returns emergent affinity clusters not in any taxonomy. A copilot returns a summary of an existing dashboard.
03
Finished work, not summaries
A copilot narrates the dashboard it's been handed. Saga produces the deliverable — the brand brief, the competitive scan, the cultural read — multi-market, on one methodology, comparable by construction. It replaces the analyst's output.
Prove itThe Monday Brief lands in the inbox before standup. A 4-market scan on one prompt. 200 high-signal mentions, not 20,000 raw ones.
04
Methodology that compounds
A copilot is stateless — every session starts from zero. Saga captures your team's approach as named, versioned, owned prompt libraries that survive attrition and scale across hires. Your method becomes infrastructure.
Prove itA junior analyst producing senior-grade output from the captured house method — library named, owner credited.
05 · What Saga does
The four jobs
One agent, four kinds of work — routed across Haiku, Sonnet, Opus.
01
Brand health briefs
Scheduled executive summaries on your brand, competitors, and category. Replaces hand-built decks.
02
Interactive queries
Conversational research over the Pulsar data layer. Ask in plain language, get charts, citations, and follow-ups.
03
Autonomous scans
Always-on monitoring cycles that flag spikes, narratives, and anomalies before they hit your inbox.
04
Deep research tasks
Multi-step investigations — segmentation, audience deep-dives, narrative origin tracing — routed to Opus.
06 · How Saga thinks
Five operating principles
If Saga isn't doing these, call it out.
1
Verdict first
The answer is the first sentence. Not a recap of what was analysed. "Sentiment flipped negative on Wednesday — the X narrative is driving it."
2
No claim of change without a baseline
Every "up", "down", "shifted" carries a comparison — prior period, prior year, competitor, category average. "Mentions are up" without a baseline is invalid output.
3
Classify the signal
Every finding is one of strong signal, emerging signal, noise, or insufficient data. Findings of different strength never presented with equal weight.
4
Position-take when the data supports it
Hedged findings the data clearly supports are a failure mode. "This is a real shift" — not "it could be argued that this might represent..."
5
Surface the limits
What the data can't tell you is part of the answer. Sample size, platform coverage, language coverage, time-window edge effects. Mandatory where material.
→
The voice
Comparative. Change verbs (doubled, flipped, stalled — never "increased"). Anchored to a concrete example. No filler. No flattery.
Part 2 of 2
— Part two
How to actually
use it.
07 · Use it for these
10 questions Saga answers well
Replace [bracketed terms] with your brand, topic, or campaign.
Brand health
"How is [brand] doing vs the competitive set this quarter?"
Sentiment trend, SoV vs named competitors, audience drift.
Crisis monitoring
"Is this story escalating — and what's driving it?"
Velocity, breakout risk, tests amplifier / narrative / event.
Campaign measurement
"Did [campaign] lift sentiment?"
Pre-vs-post baseline lift, organic vs paid, theme alignment.
Competitive intel
"Who owns the [topic] conversation?"
SoV (mention & engagement), narrative ownership, audience overlap.
Audience profiling
"Who's actually in this audience — and how has it shifted?"
Cohort definition, period-on-period, indexing surprises.
Cultural / trend detection
"What's emerging in [category]?"
Velocity over volume; nano/micro influencer tiers signal first.
Virality prediction
"Will this peak today, or run another week?"
Pulsar Virality Model + calibrated forecast with bands.
Influencer discovery
"Who are the top voices on [topic]?"
Relevance, engagement rate, audience fit — not follower count.
Product launch
"How is [product] being received — and why?"
Volume curve, comparison vocab, complaint themes, cause-tests.
Exec / M&A monitoring
"What's being said about [name]?"
Aggregate discourse only — Saga does not profile individuals.
08 · Reading the answer
Anatomy of a Saga answer
Every answer is structured. Read it top-down.
1
VERDICT
The headline finding in one sentence.
2
EVIDENCE
The data behind the verdict — comparisons and a concrete example.
+
HYPOTHESES TESTED
For "why" questions: which causes held up, which were rejected.
3
WORTH A CLOSER LOOK
What you should have asked about but didn't.
4
SO WHAT
The strategic implication — what you might do next.
5
CAVEATS
What the data can't tell you; where the sample is thin.
09 · How to ask
Getting the best out of Saga
Five things to do every time.
→
Name the search
Use the brand, topic, or campaign name as saved in Pulsar. Saga queries your data layer — use its names.
→
Tell Saga the timeframe
"Last week", "since launch", "Q3" all work. No timeframe = baseline ambiguity.
→
Ask why, not just what
Saga tests possible causes — it doesn't just describe. "Why did sentiment flip?" beats "show me sentiment".
→
Ask follow-ups
Saga keeps context. Drill down without restarting the thread. The third question is usually the best one.
→
Want a chart or forecast? Ask.
Forecasts always come with confidence bands and trigger conditions. Don't accept a point estimate.
10 · The frameworks under the hood
What Saga applies — and you can too
Frameworks you can name in the question.
The five lenses of a mention
Volume · Velocity · Engagement · Sentiment · Reach. "Mentions are up" is one lens. "Mentions up but engagement flat" is two lenses and an insight.
Macro / Meso / Micro
Strong analysis moves through at least two altitudes — market, segment, post. Ask Saga to "zoom out" or "drop to post level".
Hypothesis-driven analysis
For any "why" question, Saga generates 3–5 candidate causes — internal, external, amplifier, algorithmic, artefact — tests each, reports which held.
Audience intelligence
Audiences as set objects — defined by attribute, behaviour, affinity; operated on by intersection, union, difference, lookalike, exclusion.
Pulsar Virality Model
Seven metrics per narrative + Pattern (SPIKE / PLATEAU / DECAY / STREAKY) and Trajectory (HEATING UP / COOLING DOWN / STEADY). Plus the ±15 V×I rule.
Forecast & scenario discipline
Never a point estimate. Always range + confidence. Base / Escalation / Resolution branches, each with a trigger condition.
11 · Quality control
Hold Saga to the bar
Spot when it's off. Then call it out.
When Saga is off
You'll know because it…
- Recaps before answering
- Reports change without a baseline
- Hedges a clear finding
- Tests one hypothesis only
- Forecasts as a point estimate
- Misses the so-what
- Uses generic language
What Saga won't do
Hard boundaries — by design.
- Profile a named individual
- Give a confident yes/no on a thin sample
- Hide its assumptions
- Manufacture insight from noise
Close · Q&A to follow
One thing to take away
Saga gets better
when the team holds it
to the same standard
we hold each other.
Ask hard. Read sceptically. Call it out when it's off.