What can AI agents do for a business? Use cases and real examples

What AI agents do in customer service, sales, marketing and operations, two real Australian examples, and how to pick the first job for one. Sourced from Anthropic, the National AI Centre and iTnews. Updated 16 November 2026.

Key takeaways

  • An AI agent earns its place on jobs that need both conversation and action, have a clear measure of success, and keep a person in charge.
  • Customer service is the most proven use case: the agent can look things up and act, and every case either gets resolved or it doesn't.
  • Sales and marketing agents mostly take the admin and the repetitive drafting off people, with a person approving what goes out.
  • Most Australian small businesses using AI still use it for content, data analysis and security. The National AI Centre says agentic AI remains "largely untapped".
  • Kogan's agent started small: the earlier bot resolved about 6% of enquiries, the new agent went live after about six weeks, and it now covers 10 jobs.

What are the most common AI agent use cases for a business?

The best AI agent use cases are jobs that need both conversation and action, with a clear measure of success and a person in charge. Anthropic, which makes Claude, names customer support and coding as the two most proven. The same test points to customer service, sales follow-up, marketing campaigns and back-office admin as the usual places to start.

AI agent use cases by team

TeamWhat the agent doesWhat it connects toWhere a person signs off
Customer serviceAnswers questions, looks up orders and accounts, issues refunds, updates tickets, starts returnsHelp desk, order system, knowledge baseRefunds above a limit, complaints, anything it isn't sure of
SalesQualifies leads, drafts follow-ups and proposals, summarises calls, flags deals going quietCRM, email, calendar, call recordingsAnything sent to a prospect, and pricing
MarketingTurns customer language into campaign ideas, drafts ads for each platform and audience, reads results and proposes the next roundCRM, ad platforms, analyticsWhat goes live and what it spends
Operations and adminChases payments, files claims, follows up clients, prepares reports from several systemsAccounting, documents, spreadsheetsPayments and anything legal or financial
SoftwareFixes reported issues and writes code that automated tests can checkCode repository, test suiteCode review before release

Anthropic's test for where agents add the most value: tasks that "require both conversation and action, have clear success criteria, enable feedback loops, and integrate meaningful human oversight." It calls customer support and coding "two particularly promising applications". Coding works because "code solutions are verifiable through automated tests", though "human review remains crucial".

Most Australian businesses aren't there yet. In the National AI Centre's SME survey for December 2025 to February 2026, adopters mostly used AI for content generation (54%), data analytics (54%) and cybersecurity and threat detection (48%). It says "more complex applications – such as agentic AI, supply chain optimisation, and AI-assisted HR – remain largely untapped". For what that means for smaller firms, see what an AI agent can do for a small business.

Sources: Anthropic, Building effective agents (19 December 2024; the page notes its tooling has changed since); National AI Centre, AI adoption insights: December 2025 to February 2026 (first published 7 May 2026, updated 3 June 2026). Fetched 16 November 2026. The table is our summary; the operations row uses the examples from our post From Filing Cabinet to SaaS to AI.

What are real examples of AI agents used by Australian businesses?

Kogan is a well-documented Australian example. iTnews reported on 14 October 2026 that the online retailer's AI agent, built on Salesforce Agentforce, handles 10 customer service jobs, from order status pulled live from SAP to returns. Productbox Labs runs its own: an AI adviser on Off Plan Properties that answers buyers from project data.

Kogan: a customer service agent, built in stages

  • What it handles: "So far, 10 customer service use cases can be handled by the AI agent." The first, "where is my order?", is "designed to pick up the order status directly from SAP and pass it on to the customer". Order-status questions make up "roughly 60 percent of Kogan's inbound customer requests", a figure iTnews takes from a July Salesforce post.
  • What else: membership, mobile voucher redemption, unknown charge investigations, troubleshooting and identity verification, plus a product concierge for recommendations.
  • Photos of damage: an emerging use reads the photos customers send of damaged goods in a return. Kogan's CTO says the AI can detect when multiple people submit the same image.
  • How long it took: prototyped at an internal hack day, it "went live in production after about six weeks."
  • Where it started: the earlier, rules-based bot "was able to resolve about six percent of inbound customer inquiries." The CTO's advice: "You've got to be comfortable with a starting point". The article gives no resolution rate for the new agent.

Off Plan Properties: an adviser that hands over to people

Productbox Labs built and runs Off Plan Properties, a research site for buyers of new developments on the Gold Coast, the Sunshine Coast and in Northern NSW. In the site's own words: "Our AI adviser answers from verified project data. Our property team takes it from there when you are serious." It also "builds your property brief as you talk". That is the pattern we recommend for most first agents: the agent does the research and the questions, a person does the deciding. See the other sites we build and run.

Sources: Ry Crozier, iTnews, Kogan taps AI agent to handle customer service (14 October 2026); offplanproperties.com.au homepage. Both fetched 16 November 2026. No results figures are claimed for Off Plan Properties.

What can an AI customer service agent do?

An AI customer service agent answers customers and acts on what they ask: it looks up the order, account or policy, then issues the refund, updates the ticket or starts the return. Anthropic calls support a natural fit because success is easy to measure. Start with your most common question and hand anything unusual to a person.

Why support suits an agent, in Anthropic's words

  • "Support interactions naturally follow a conversation flow while requiring access to external information and actions"
  • "Tools can be integrated to pull customer data, order history, and knowledge base articles"
  • "Actions such as issuing refunds or updating tickets can be handled programmatically"
  • "Success can be clearly measured through user-defined resolutions"

That last point is why some providers charge per resolved case. Anthropic notes companies using "usage-based pricing models" that "charge only for successful resolutions". If you're quoted that way, ask exactly what counts as a resolution.

How to start

  1. Find your most common enquiry. At Kogan it was order status; at most businesses it's something similarly dull.
  2. Connect the agent to the one system that answers it, read-only at first.
  3. Have a person check its replies before they reach customers. The National AI Centre found about half of SMEs using AI already check outputs before they affect customers, but fewer tell customers when AI is involved.
  4. Add actions (refunds, returns, ticket updates) one at a time, with limits.

Your business is responsible for what the agent tells customers. Read who is liable when an AI agent gets it wrong.

Sources: Anthropic, Building effective agents (19 December 2024), Appendix 1; iTnews (14 October 2026, above); National AI Centre, AI adoption insights (updated 3 June 2026). Fetched 16 November 2026. The steps are ours.

What can an AI sales agent do?

An AI sales agent does the admin around selling so your reps spend more time with buyers: it qualifies new leads against your criteria, drafts follow-ups and proposals, summarises calls into the CRM and flags deals going quiet. A person should approve anything it sends to a prospect, at least until its drafts are reliably right.

Typical sales agent skills

  • Qualify a lead: check each new enquiry against your criteria and tell the right rep why it matters.
  • Draft a proposal: start from your past proposals and the notes on this deal.
  • Summarise a call: write the notes and next steps into the CRM.
  • Chase deals: point out deals that have gone quiet and draft the follow-up.
  • Tailor messages: write different versions for different buyers, for example a CFO focused on cost and a CTO focused on security.

The first three are skills shared by the whole team in the agents we build: build a skill once and every rep gets it. The agent is only as good as what it can see, so it needs access to the CRM, past proposals and call notes. Keep a person approving anything that goes to a prospect, and anything to do with price.

Sources: Productbox Labs homepage (skill registry: "Qualify a lead. Draft a proposal. Summarise a call.") and our post The LinkedIn Ads Paradox (11 August 2026). Read 16 November 2026.

What can an AI marketing agent do?

An AI marketing agent runs the whole campaign loop, not just the copy: it pulls the words customers use from sales calls and the CRM, turns them into campaign ideas, drafts ads in each platform's format for each audience, then reads the results and proposes the next round. A person approves what goes live and what it spends.

The four-step loop, and what the agent does in each

StepHubSpot's descriptionWhat the agent does
Express"Define what to say, how to say it, and why it matters right now."Reads recent sales call transcripts and pulls out the phrases customers used, then turns them into a few campaign concepts
Tailor"Use AI to turn your strategy into content that connects personally"Checks the concepts against the CRM to see which fits which customer group, and drafts versions for each
Amplify"Publish your content across diversified channels"Formats each version for the platform: video scripts, carousels, single-image ads
Evolve"Iterate quickly and effectively."Reads the weekly results (cost per click, click-through, conversions) and drafts the next round

Marketing is where Australian businesses already use AI most: content generation was the top use among SME adopters (54%) in the National AI Centre's latest survey. The difference with an agent is that it closes the loop, reading the results and changing the next campaign, rather than writing one post at a time. For the full blueprint, read The 4-Step Marketing Loop.

Sources: HubSpot, Loop Marketing (page undated; stage descriptions quoted verbatim); National AI Centre, AI adoption insights (updated 3 June 2026). Fetched 16 November 2026. The agent column summarises our post (11 August 2026).

How do I choose the first job to give an AI agent?

Choose a job your team does every week that has a clear right answer, draws on information already in your systems, and where a mistake can be caught before a customer sees it. Then ask whether a fixed-step automation would do: Anthropic and the National AI Centre both advise the simplest option that works.

A five-question check

  1. Does it happen every week? Rare jobs don't repay the set-up.
  2. Does it need conversation and action? If it only needs one, a chatbot or a fixed-step automation is probably enough.
  3. Can you tell when it's done right? Anthropic lists "clear success criteria" as a condition.
  4. Is the information already in your systems? If the answer lives in someone's head, write it down first.
  5. Can a person catch a mistake before it reaches a customer? Start where errors are cheap.

Anthropic's advice to builders is "finding the simplest solution possible, and only increasing complexity when needed". The National AI Centre tells buyers to "choose the simplest option that meets your needs, and to add complexity only where the value justifies it". If a fixed-step workflow does the job, use that. See whether you need an agent or a simpler automation, and what an AI agent costs to build and run.

Sources: Anthropic, Building effective agents (19 December 2024); National AI Centre, Choose a solution (published 22 April 2026). Fetched 16 November 2026. The checklist is ours.

What would a Productbox agent do for my team?

Productbox Labs builds one custom agent your whole team shares. It learns the business once, keeps client records, deals, emails, documents and meeting notes in one memory, and runs shared skills such as qualifying a lead, drafting a proposal or summarising a call. Each night it reviews its own conversations and improves. It goes live in about 30 days, and you own it.

Want a number before you talk to anyone? The AI ROI calculator estimates what an agent is worth for your team.

Got a job in mind for an agent?

Tell us what your team does every week. We'll tell you which job to start with, and whether it needs an agent, a simpler automation or nothing at all.