Every use case below follows the same recipe: identify a repetitive piece of work, define the success criteria in numbers, build the smallest agent that meets them, instrument the outcomes, keep a human on the consequential decisions.
◐ Customer Service5 use cases
Customer service that does not make people wait
Tier-one questions resolved in seconds, complex ones routed to the right person with the full history attached. The agent handles the obvious. The humans handle the interesting.
- Ticket triage and routing. Read the incoming ticket, classify it (refund, bug, billing, account access), tag it, set priority, and assign to the right team. Average handle time on inbox queues drops by half.
- First-touch resolution. Password resets, order lookups, return labels, shipping status, FAQ answers, account questions. The agent answers directly, posts to your help desk, and closes the loop without a human in the chain.
- Reply drafting for human agents. Agent reads the ticket, pulls the relevant knowledge base article, and drafts a reply. The human reviews, edits, and sends. Reply time drops, tone stays human.
- Multilingual support. A single agent that reads and replies in 20+ languages without losing tone or context. No more outsourcing language coverage.
- Voice agents for inbound calls. Natural-voice intake for high-volume call centers: identity check, intent capture, simple resolution, and human handoff with the case already built.
◑ Sales5 use cases
Sales reps doing the high-value half of the job
Lead qualification, CRM hygiene, and follow-up sequences are handed to the agent. Reps spend their day in conversations, not in Salesforce admin.
- Inbound lead qualification. Score the lead, enrich the company, check fit against ICP, and either book a meeting or send a polite decline. The rep wakes up to a calendar full of qualified intros.
- CRM updates from meetings. Listen to the call (or read the transcript), update the opportunity, set the next step, log the contact, and Slack the deal owner with anything that needs attention.
- Outreach personalization at scale. Read the prospect, the company, the recent news. Draft a first-touch email that does not sound like spray-and-pray. Send only the ones the rep approves.
- Pipeline hygiene. Every Monday morning, the agent walks the pipeline, flags stale opportunities, suggests next steps, and asks the rep questions about deals that have not moved.
- Proposal and quote generation. Generate the first draft of a proposal or quote from a templated structure and the deal context. The rep edits and sends.
◒ Marketing5 use cases
Personalization that respects the human reading it
Content variants, send-time decisioning, audience design, and post-campaign analysis. The agent does the iteration. The marketer keeps the taste.
- Content variant generation. Subject lines, push copy, ad headlines, on-site banner text. Generate 20 variants per campaign, score against the eval rubric, and ship the top three for live testing.
- Send-time and channel optimization. Pick the right channel and the right hour for each individual. Send-time models that actually move open rates instead of guessing about time zones.
- Audience design and lookalike expansion. Translate a marketer prompt ("customers similar to our best second-purchase cohort but acquired in the last 60 days") into the segment definition in your CDP. Validate the size, suggest channels, push to activation.
- Post-campaign analysis. Read the journey, the segments, the channel mix, the holdout. Write the post-mortem in plain English: what worked, what did not, what to try next.
- Liquid and AMPScript assistance. Engineer-level help in Braze and Marketing Cloud: write the personalization snippet, check it against the brand guide, suggest a fallback path when data is missing.
◓ Operations5 use cases
Back-office work that runs itself
Document processing, approval chains, vendor management, and quality checks. The agent moves the work forward; humans approve the consequential steps.
- Invoice and document extraction. Read PDFs, contracts, purchase orders, shipping documents. Extract structured data, match against the ERP, flag mismatches, route to the approver. Built-in audit trail.
- Approval workflow routing. The agent reads the request, checks policy, finds the right approver based on amount and category, follows up when the approval sits too long, and escalates by the SLA.
- Quality assurance for content and data. Spot duplicates, missing fields, broken links, brand-guideline violations, and translation errors before they ship. Surface them with context so the human can fix them in one pass.
- Vendor onboarding. Read the vendor docs, run the standard checks (insurance, sanctions, tax forms), score the risk, and queue the missing items for follow-up.
- Inventory and supply signals. Watch the data feeds, flag anomalies, draft the early-warning note to the planner. The first version of a forecast you can argue with.
◔ Finance5 use cases
Numbers checked before they get to the meeting
AP automation, fraud detection, reconciliation, and anomaly explanations. The agent does the matching. The controller signs off on the exceptions.
- AP automation. Three-way match between invoice, PO, and goods receipt. Flag duplicates and policy violations. Route exceptions to the right approver. Pay only what is owed, when it is owed.
- Expense report review. Pre-check expense reports against policy: receipts, categories, amounts, missing items. Approve the clean ones automatically. Surface the unusual ones with context.
- Fraud and anomaly detection. Watch transactions and approvals for the patterns that should not be there. Flag with reasoning, not just a score, so the analyst can act.
- Reconciliation and close. Match bank lines to ledger entries, propose the reconciling journals, write the reasoning. The team reviews and posts.
- Plain-English variance explanations. The agent reads the actuals, the budget, and the prior year. Writes the variance commentary in the voice of your FP&A team. Saves the analyst three days a quarter.
◕ HR & People5 use cases
People work with the busywork removed
Sourcing, screening, scheduling, onboarding, and policy lookups. The agent does the high-volume work. Recruiters and HRBPs spend their day on the human conversations.
- Resume screening and shortlisting. Read the resume, score it against the role spec (not against the candidate, against the criteria), flag the strong ones, write the rejection note for the rest with care.
- Interview scheduling. Look at the panel calendars, propose times, send the invite, send the reminders. The candidate experiences a clean booking flow, the recruiter never opens an email thread.
- Onboarding agent for new hires. Day-one questions: laptop, payroll, who do I ask about X, where is the wiki. The agent knows the org and the policies; the new hire is productive on Monday.
- Policy and benefits lookup. Employees ask in Slack, the agent answers from the policy doc with a citation. HR closes a third of their inbound.
- Performance review prep. Pull the manager and peer feedback, the goals, the work product. Draft a first cut of the review. The manager edits with judgment.
◖ Internal Knowledge5 use cases
Your wiki, but it actually answers
Retrieval-augmented agents grounded in your docs, your runbooks, your tickets, your codebase. Answers come with citations and a confidence note.
- Engineering runbook copilot. When an alert fires, the engineer asks: what is this, who owns it, how did we resolve it last time. The agent reads the runbook, the past incidents, the dashboards. Time to mitigate drops.
- Sales enablement search. A rep asks: what is our latest position on data residency, do we have a case study in healthcare, what is our pricing for a 500-seat customer. Citations included.
- Customer support knowledge. Same idea, customer-facing. The agent pulls from your help center, past tickets, and product docs to draft an answer the human can ship.
- Policy and compliance Q&A. A legal or compliance question gets a sourced answer from your policies. The agent flags the cases that need a human lawyer.
- Onboarding for technical teams. New engineer joins; the codebase is large. The agent answers structural questions ("where does payment flow live, who reviews changes there"), reducing the load on senior staff.
◗ Data & Analytics5 use cases
Insights without the SQL detour
Natural-language access to the warehouse, anomaly explanations, and lightweight forecasting. The analyst stays in the loop; the questions get answered faster.
- Natural language to SQL. Business users ask in English. The agent writes the SQL, runs it against a sanctioned schema with row-level security, shows the result and the query. The analyst stays in control of the source of truth.
- Anomaly explanations. When a KPI moves, the agent decomposes by dimension, finds the driver, and writes a one-paragraph explanation. The team starts the day with the why, not just the what.
- Data quality monitoring. The agent watches the freshness, the row counts, the schema drift, and the value distributions. Alerts come with context: what changed, since when, likely cause.
- Dashboard search. No one knows which of 600 dashboards has the answer. The agent does. It points the right person to the right view with the relevant filters set.
- Forecast assistance. Generate a first-cut forecast against history, surface the assumptions, let the planner adjust. Faster than a spreadsheet, more defensible than gut feel.
◘ Engineering5 use cases
Engineering throughput, with the boring parts on autopilot
Code review, PR triage, test generation, incident summaries. The agent does the first pass. Engineers spend their time on design and judgment.
- Code review and PR triage. Read the diff, check against the team conventions, suggest improvements, flag risky changes for senior review. PRs move faster without dropping standards.
- Test generation. Look at the function, the surrounding context, and the bug history. Generate the unit and property tests that would have caught the last three regressions in this area.
- Incident summarization. Read the alert thread, the dashboards, the recent deploys. Write the timeline, the impact, the likely cause. Saves the incident commander an hour during the worst moment of the week.
- Documentation maintenance. Detect drift between code and docs, propose updates, open the PR. The team reviews instead of writing from scratch.
- Stale-dependency triage. Watch the dependency tree, flag CVEs, suggest the right upgrade path, generate the PR. The engineer reviews and merges.