AI Workforce Management for Support Teams (2026): Forecasting, Rosters and AI Agents

AI workforce management means two different things for a support team in 2026: software that does the workforce management work itself (forecasting, building rosters, checking rules, flagging gaps), and AI agents that resolve part of your volume, which changes what you are forecasting in the first place. Gartner predicts agentic AI will resolve 80% of common customer service issues without human intervention by 2029, yet it also predicts that by 2027, 50% of organisations that expected to significantly cut their service workforce will abandon those plans. Planning the humans you still need is the hard part, and it is where most guides stop short.

This article covers both senses. The first half is about what AI can and cannot do inside the planning cycle. The second half is about the less discussed problem: when an AI agent resolves 40% of your tickets, your human workload does not fall by 40%, and your forecast has to know why.

Key takeaways

• There are two kinds of AI workforce management. AI that plans (forecasts, schedule generation, rule checks, gap flagging) and AI that works (agents resolving tickets). Most WFM content covers only the first; support teams now need both.

• Volume falls faster than workload. In our illustrative example, AI agents resolve 40% of contacts but human workload falls only 25%, because the contacts left over take 25% longer on average.

• Forecast the human residual, not total demand. Plan human seats on the contacts AI does not finish, including handovers, and re-baseline monthly while your automation rate is still moving.

• Copilots change handle time too. In a study of 5,179 support agents, an AI assistant raised issues resolved per hour by 14% on average and 34% for novice and low-skilled workers, so AHT assumptions should vary by tenure.

• An optimiser does what you tell it. Assembled warns that a scheduler told to maximise coverage efficiency will recommend 95% occupancy; a person still sets the targets and limits.

Which tools do AI workforce management for support teams?

Lorikeet Workforce Manager builds a roster from your ticket and call volume by channel, day and hour, checks each draft week against contracted hours, rest between shifts, channel skills and approved leave, then flags under-covered shifts and suggests qualified, available agents. It is built for support teams of 20+ agents running live chat or phones. Alternatives worth comparing:

• Assembled says its forecasts account for AI agent capacity and human coverage together, alongside BPO vendors, which suits teams running a blended workforce at scale.

• Zendesk WFM offers AI-powered forecasting that shows how many agents you will need, a natural fit if your team already runs on Zendesk.

• NICE applies machine learning, optimisation and automation across the contact center planning cycle, including intraday changes; large or regulated contact centres often need this depth.

• Microsoft Dynamics 365 has an AI Agent Estimator that forecasts AI agent capacity and usage alongside human staffing, for teams already on Dynamics.

For the full side-by-side, see our guide to workforce management software for support teams.

What is AI workforce management?

AI workforce management is the use of AI to forecast demand, calculate staffing, build schedules and respond to changes in a support or contact center operation, with less manual analysis than a spreadsheet or a traditional WFM tool. NICE describes it as applying machine learning, optimisation and automation to the contact center planning cycle. Genesys adds that it factors in agent skills and time-off requests when planning shifts. Xima splits the forecasting side into predictive AI, which anticipates workforce needs from history and seasonal patterns, and prescriptive AI, which suggests what to do about them.

Those definitions describe AI as the planner. For a support team in 2026 there is a second meaning that matters just as much: AI agents that resolve customer contacts end to end. They do not plan the roster, but they change the demand the roster is built for. Our WFM glossary entry covers the underlying vocabulary, and the contact center workforce management guide walks through the full forecast, Erlang C and shrinkage cycle. This article sits on top of both.

What WFM work can AI do today?

AI can now do most of the repetitive analysis in workforce management: building the volume forecast, generating a draft schedule, checking it against working rules, and flagging where coverage falls short. What it does less well is know about events nobody recorded, and decide what trade-offs are acceptable.

Scheduling is where the step change is largest. Assembled notes that for a team of a thousand agents across a week, the number of valid schedule combinations is roughly 10^30,000, which is why legacy tools fall back on rigid shift templates. A 30-agent support team is nowhere near that scale, but the same problem shows up as a team lead spending Friday afternoon fixing a spreadsheet by hand.

NICE frames the forecasting goal well: not a perfect forecast, but one that responds quickly when the underlying pattern changes. It also lists what intraday tools do once the day starts: alert when volume or handle time departs from forecast, identify intervals at risk of under- or overstaffing, and reforecast the rest of the day. For a support team, that is the difference between knowing at 10 a.m. that the afternoon is short and finding out from the queue at 2 p.m.

In Workforce Manager, each week starts as a draft checked against contracted hours, rest between shifts, channel skills, approved leave and demand. If an agent is on leave Monday and phone coverage falls below forecast, it flags the shift and suggests qualified, available agents. Capacity planning shows shortfalls up to 12 months ahead, factoring in growth. The launch post says intraday management is next.

How do AI agents change workforce forecasting?

AI agents change workforce forecasting by removing a slice of contacts that is not a random sample: they usually resolve the simpler, more repetitive requests first, so human volume falls faster than human workload, and average handle time on what is left goes up. The table below is an illustrative example, not a benchmark; use your own resolution and handle time data.

The assumption doing the work: the 400 contacts AI agents resolve used to take people 5 minutes each, well under the 8 minute average. Removing 2,000 minutes of easy work leaves 6,000 minutes spread over 600 harder contacts. Then the staffing maths makes the gap wider again, because smaller queues need proportionally more spare capacity to hit the same service level. The contact center WFM guide works this through Erlang C: AI agents take 29% of contacts, but the roster falls 16%. Our guide to call center staffing with Erlang C goes deeper on the formula.

There are two schools of thought on how a WFM system should handle this. Writing in Call Centre Helper, NiCE argues that WFM systems forecast only what agents actually handled, so as self-service improves the history adjusts on its own, the way it did for IVRs. Microsoft and Assembled take the other view: forecast AI agent capacity alongside human staffing, so planners can model how demand splits between the two before it hits the budget.

Both are right at different moments. When your automation rate is stable, history-based forecasting on the human residual works. When it is moving month to month, as it does in the first year of an AI agent rollout, history describes a mix of work that no longer exists, and you need to model the split explicitly. Practical rules:

• Split the forecast in two. Forecast total demand, the share AI agents resolve, and the residual for people, per channel and interval. Automation rates can differ by hour and by channel, so a daily average hides where the gaps are.

• Count handovers as human work. A contact an AI agent starts and passes to a person lands in your human queue. It belongs in the residual forecast, with its own handle time.

• Forecast AHT on the residual, not the old mix. Expect it to rise as the easy contacts leave, and track it monthly.

• Re-baseline after every big automation change. A new workflow that resolves refunds end to end changes next month's residual more than any seasonal pattern.

Our article on workforce forecasting for support teams covers the forecasting method itself in more detail.

How should you staff a blended human and AI support team?

Staff a blended team on the human residual, with skills matched to harder contacts and a buffer while the automation rate settles, rather than cutting headcount in proportion to what AI resolves. The market is moving the same way: in a Gartner poll of 163 customer service and support leaders in March 2025, 95% planned to retain human agents to strategically define AI's role, and Gartner's Kathy Ross called a hybrid approach, with AI and human agents working in tandem, the most effective strategy.

Four things change in a blended roster:

1. Skills, not just seats. If AI agents resolve the routine work, the human queue skews towards complaints, exceptions, vulnerable customers and multi-step problems. Channel skills and seniority matter more per shift than they did.

2. Tenure changes handle time. Copilot tools affect people unevenly. The NBER study by Brynjolfsson, Li and Raymond found an AI assistant raised issues resolved per hour by 14% on average, 34% for novice and low-skilled workers, with minimal impact on experienced staff. A single AHT for the whole team gets less accurate as tools spread.

3. Coverage hours may shift. AI agents work every hour the channel is open, so the hours where human coverage matters most are the ones with the most handovers and complex contacts. Check the residual by interval before keeping last year's shift pattern.

4. Headcount moves slower than volume. Hiring and attrition take months. Use capacity planning to decide whether to hire, hold or let attrition run, and revisit it as the automation rate becomes clear. Our guide to scaling support without hiring covers the decision from the budget side.

Lorikeet's AI agents resolve customer problems end to end across phone, SMS, chat, email and WhatsApp, and loop in your team when they cannot solve an issue. Workforce Manager plans the human side from your volume by channel, day and hour; Lorikeet customers skip the roster upload because it reads ticket history directly.

What should you check before buying an AI workforce management tool?

Check that the tool can show its forecast working, handles the channels and working rules your team actually has, and plans on the volume people will really handle. NICE's own evaluation list includes forecast accuracy and transparency, multi-skill and omnichannel planning, rule handling, intraday reforecasting and explainable recommendations. For a support team of 20 to 100 agents, add these questions:

• Does it plan chat with concurrency? Assembled advises verifying that the scheduling model handles concurrency and asynchronous work, because voice-only forecasting will not staff an omnichannel team accurately.

• Can it separate AI-resolved and human-handled volume? If your AI agents and your WFM tool cannot see each other's numbers, someone reconciles them in a spreadsheet every week.

• How long until a first roster? Teams without a dedicated workforce planner should ask how much setup sits between importing data and a usable draft.

• Does it check your rules before publishing? Contracted hours, rest between shifts, skills and leave should be checked on every draft, not found by an agent on Monday.

• What does it measure? NICE recommends tracking service level, wait time, occupancy, overtime, schedule stability, adherence and employee experience alongside forecast error. Our article on WFM metrics covers how to read them.

What still needs a human?

AI can generate the forecast and the roster, but people still own the targets, the exceptions and the trade-offs. NICE says as much in its own FAQ: AI can automate parts of forecasting, scheduling and intraday analysis, but planners and managers still handle exceptions, business changes, employee considerations and policy decisions.

• Setting the targets. Assembled puts it plainly: AI scheduling optimises for what you configure, and a system told to maximise coverage efficiency will recommend 95% occupancy, a burnout rate. The judgement behind the rules matters more with AI, not less.

• Events nobody logged. A pricing change on Thursday, a partner outage, a campaign marketing forgot to mention. The forecast only knows what it is told.

• Judging the automation rate. How fast AI agents will take on more work is a business decision tied to which workflows you build next, not a pattern in history.

• Fairness and wellbeing. Who always gets the late shift, who is close to burnout, who needs development time. Rules encode rest and hours; people cover the rest.

• The hardest contacts. As AI agents take the routine work, the human queue is where escalations, complaints and vulnerable customers land. Staffing it well is a people decision as much as a numbers one.

Putting it into practice

1. Export 8 to 12 weeks of volume and AHT by channel and interval, and tag which contacts AI agents resolved.

2. Forecast total demand, the AI-resolved share and the human residual separately, with handovers counted as human work.

3. Recalculate AHT on the residual, and split it by tenure if your team uses copilot tools.

4. Run the staffing maths per interval and add shrinkage, as in the contact center WFM guide.

5. Let a tool draft and rule-check the roster, then review targets, fairness and known events yourself.

6. Re-baseline monthly while the automation rate is still changing.

If you would rather not rebuild the spreadsheet every week, Workforce Manager imports your current roster model from a spreadsheet or helpdesk export, turns your volume into a staffing forecast you can fine-tune for known events, and shows coverage by channel and hour. You can start a 30-day free trial and build a first roster from your own data. If you are still sizing the cost side, our cost per ticket guide is a useful companion.