Most HR leaders can say AI is in use across their organisation. Fewer can say what it is doing for the work. AHRI’s December Quarterly Australian Work Outlook report found that 93 per cent of organisations use AI in some capacity, and three in five report that work intensity is rising as a result. Adoption is high, the return is harder to see, and boards are starting to ask about both.
Use token data to open the conversation
Token usage shows who is using AI, how often and on what kind of work. Treated as a scorecard, it will mislead you: heavy usage can mean busywork, and light usage can mean someone automated a task well and moved on. Treated as a map, it shows where curiosity and friction sit. That is where HR is useful. Effective adoption depends on a candid conversation between leaders and their teams about which tasks people have handed to AI, which they tried and dropped, and where they are unsure what is allowed. HR can convene that conversation and bring the data. The question to keep returning to is hours reclaimed and quality of output, with prompt volume a distant second. At Deel, AI spend is tied directly to the hours it gives back.
The skills barrier is falling
Agentic AI can deliver significant operational returns, but in functions like HR, finance and payroll, those returns have mostly gone to organisations with the AI literacy to build agents, or the engineering budget to hire it. The work that most needed automating, such as filings on government portals, payouts through bank sites and reconciliations across systems that do not talk to each other, often had no API to connect to. That is changing quickly. Deel built Akai for its own operations teams for this reason. Someone records themselves completing a task manually. Akai observes the steps, builds the workflow with conditional guardrails and runs it, including on legacy portals with no API access. Critical actions, such as bank transfers, still require human approval.
What Deel saw internally
Every Deel operations team, across finance, tax, treasury, benefits and HR, now runs on Akai. Deel reports more than one 1 million hours saved to date and more than 250,000 cases handled automatically per month. A reconciliation that took over 20 days now takes minutes. In expense reimbursements, an earlier automation layer handled 78 per cent of 1,300 daily requests, and the rest sat in a manual queue. Akai now works as a second layer on that remainder. One payroll operator at Deel built a first agent in under an hour, without developers or IT tickets.
The trade-offs to plan for
Automating the edge cases changes the shape of a team’s work. People spend less time on repetitive processing and more on reviewing exceptions and making judgement calls, which asks for different skills. It also queries whether these governance questions are an HR responsibility: who approves a new agent, what data it can touch, and how it meets obligations such as GDPR. Deel’s own research with IDC also points to a quieter risk, namely that automating routine work removes the tasks that early-career employees once learned on.
Questions for your next leadership meeting
- What does our token data say about where AI is already doing real work, and where it is idle?
- Which manual, portal-based tasks in HR and payroll cost us the most hours each month?
- Who in our teams could build an agent if the tooling asked for no code?
- How will we know an agent’s output is right, and who signs off?
Akai by Deel is available at akai.run.
Guy Fenton, Global Director, Akai by Deel. Joining Deel in 2020, Guy has helped scale the business globally and currently leads go-to-market and AI transformation initiatives across APAC and global operations. He works closely with People and operations leaders to operationalise agentic AI—automating manual workflows across HR, finance, and payroll while keeping human judgment and governance at the centre.
