Research Agents
Research Agents fill in what your connected tools don’t tell you. Point an agent at a question — “what’s this company’s latest funding round?” or “are they hiring for roles that signal expansion?” — and it searches your internal data and the public web, then writes the answer back to the customer record. Agents run on a schedule or fire when new records appear, keeping data current without manual research.
How they work
Section titled “How they work”Each agent has a prompt, data sources, and an output format. When it runs, the agent queries internal data — support tickets, meeting transcripts, product usage, feature requests, satisfaction trends — and searches the public web for company news, hiring activity, or funding announcements. It synthesizes what it finds into a structured answer: a score, a date, a yes/no, a short summary, or a pick from a list you define.
Results attach to customer, contact, or issue records. Run an agent once to backfill, or schedule it to refresh on a cadence.
What agents can access
Section titled “What agents can access”Agents draw from two pools. Internal data includes product usage, support history, meeting summaries and sentiment, satisfaction trends, feature requests, and contact engagement — the same data behind chat and canvas. Public web covers company websites, news, job postings, funding databases, and press releases.
You choose which sources each agent uses. A “tech stack” agent might only need web search. A “renewal risk summary” might pull from usage data, recent tickets, and meeting sentiment without touching the web.
Built-in agents
Section titled “Built-in agents”Customer health weights signals — product usage, support volume, meeting sentiment, engagement frequency — into a status: Thriving, Healthy, Emerging Concerns, or At Risk. You configure which signals matter and how much.
eCSAT (estimated satisfaction) scores every support interaction 0–10 with detected emotions and themes. Scores roll up into per-customer trends, so you spot declining accounts before anyone flags them.
Escalation risk estimates how likely a support issue is to escalate based on tone, urgency, repeat contacts, and business impact.
Custom agents
Section titled “Custom agents”Define a prompt, pick data sources, set the output format, and choose when it runs. Teams have built competitive intelligence from web mentions, expansion signals from job postings, renewal context combining usage with support history, and ICP scoring from public company data.
Outputs can be numbers, dates, text, booleans, URLs, or a selection from a list you define.
Where enrichments flow
Section titled “Where enrichments flow”Agent results feed the rest of Sonora.
CX Cues trigger on enrichment values. When health drops to At Risk or escalation probability crosses a threshold, a cue appears with context and a suggested action.
Customer Intelligence includes enrichments in answers. Ask “show at-risk customers” and health scores surface. Ask about sentiment and eCSAT history appears automatically.
Canvas blocks display enrichment data — health distributions, satisfaction trends, or custom agent results filtered by segment.
Customer profiles show health status, satisfaction trends, and custom agent outputs inline. Scores sync back to Salesforce or HubSpot so your team sees them where they already work.