Latest
Deal Heidi cuts headcount by 18% and co‑founders leave after $475 million raiseDeal What buyers look for when they acquire an AI companyGlobal AI agents in accounting and finance teamsGlobal AI agents in retail and e-commerceGlobal AI agents in logistics and supply chainsUnited States Texas elects first Democratic governor in decades, putting AI data centre boom at riskGlobal OpenAI launches GPT‑6 Cyber model for defensive cybersecurity
United States edition
Friday, 6 November 2026
Infercline.
Advertise
The AI Opportunity Desk

Which business processes can AI agents handle today

Agents already take on real work in customer service, finance and freight. The jobs they handle best share a pattern, and so do the ones they still get wrong.

The Infercline Desk · Friday, 6 November 2026 Sign in to save

Most of the talk about AI agents is about what they might do. Operators need to know what they do now, in live businesses, at a cost that makes sense. The honest answer is a shorter list than the marketing suggests, but a longer one than sceptics admit. The processes that work share a pattern: high volume, digital information, clear rules for a good result and a person able to check the outcome.

Customer service is the most mature area. Vendors such as Intercom and Zendesk now price their AI agents by the outcome or the resolution, meaning a customer request the AI settles without passing it to a person, according to their pricing pages at the time of writing (November 2026). That pricing only makes sense because these agents finish a meaningful share of routine questions on their own: order status, password resets, delivery questions and policy answers. Gartner predicted in March 2025 that agentic AI would resolve 80 per cent of common customer service issues without a human by 2029. That is a forecast, not a result, but it shows where the industry expects this work to go.

Finance and bookkeeping are close behind. Xero says its AI agent, JAX, is designed to automate routine tasks such as bank reconciliations, data entry and getting paid. Intuit announced in July 2025 a set of AI agents for QuickBooks covering payments, accounting, finance and customer work. These tasks suit agents because the information is already digital, the rules are written down and an accountant can check the result quickly. Matching invoices, chasing overdue payments and categorising transactions are the common starting points.

Freight and logistics offer some of the clearest published numbers. In April 2025 C.H. Robinson said its AI agents had performed over 3 million shipping tasks, including well over 1 million price quotes, 1 million orders and more than 1 million appointments. It said loads were accepted in under 90 seconds for more than 5,200 customers, against up to four hours before. Uber Freight said in May 2025 that it runs more than 30 AI agents across the shipment lifecycle. Much of this work is reading emails, extracting details and entering them into systems, at a volume no team could match by hand.

Sales support is a growing area, with care. Agents can research prospects, draft first emails, log calls and keep records up to date. The risk is not technical. It is reputational and legal, because marketing rules differ by country and customers notice when they are being handled by software. Most businesses keep a person sending or approving outbound messages.

Internal operations are another fit. Answering staff questions about policies, routing IT requests, preparing reports from several systems and drafting documents from templates are all repetitive and digital. ServiceNow's 2025 agreement to buy Moveworks, a company focused on AI agents for employee support, shows how large software firms see this market.

The pattern also shows where agents still fall short. Work that depends on relationships, negotiation or reading a room stays with people. Work where a mistake is costly and hard to spot, such as legal advice, medical decisions or large payments, needs a person deciding, even if an agent prepares the file. Work where the information lives on paper or in someone's memory needs to be digitised first. And work that happens rarely seldom justifies the set-up effort.

Adoption figures support a cautious view. McKinsey's 2026 State of AI survey found that about two in ten respondents had reached the stage of scaling AI agents across their organisation. Many more are experimenting. The gap between the two is usually not the technology but the work around it: clear processes, clean data, controls and someone who owns the result.

Why it matters to your business

Knowing which processes agents handle today lets you start where the odds are best. A first project in customer service, bookkeeping or document-heavy operations has published examples to learn from and vendors with pricing built around it. A first project in a judgement-heavy area is more likely to disappoint and to sour your team on the whole idea.

It also helps you spot opportunity. If competitors in your sector still handle quotes, bookings or routine queries by hand, an agent that does it faster can be an advantage your customers notice.

What to do on Monday

  1. List your top ten processes by monthly volume. Mark each one digital or not digital, and rules-based or judgement-based.
  2. Shortlist the processes that are high volume, digital and rules-based. These are your candidates.
  3. For each candidate, find out whether your existing software vendor already offers an agent for it before looking elsewhere.
  4. Choose one candidate where a mistake is cheap and easy to spot, and make it your first trial.
  5. Leave relationship, negotiation and high-risk decisions with people for now, and revisit the list every six months.

More on this topic: AI agents by industry

Sources