AI Workforce Management and Diallers: Planning Capacity That Is Not People

Research date: 23 September 2026. Last updated: 23 September 2026.

Workforce management exists to answer one question: how much capacity do you need, and when. For thirty years the unit of capacity has been a person, and the whole discipline — forecasting, scheduling, shrinkage, adherence — is built on the fact that people must be hired weeks ahead, work fixed shifts and take breaks.

An AI agent is not that. Its capacity is a concurrency setting, it can be changed this afternoon, and it does not take lunch. That does not make workforce management obsolete. It moves the bottleneck, and most teams discover where it moved only after go-live: you no longer need enough people to answer the calls, you need enough people to take the escalations — and that number is a function of your containment rate, which is not in any WFM forecast built before you deployed.

The dialler half of this category has a different problem. Its binding constraint is not technology or capacity. It is regulation, the rules differ by country, and the figure almost everyone quotes is not the universal standard it is presented as.

Brilo AI publishes this page and sells an AI voice agent. Brilo does not sell workforce management software and does not sell a dialler. What Brilo fails to publish about the single most important number on this page is stated below.

How we compiled this

Documented research, not a hands-on evaluation. Prices, regulations and corporate facts below were read from the vendor's own page, the regulation text, or the company's own announcement on 23 September 2026, and each carries its source. No WFM platform was trialled.

Nothing here is legal advice. The dialler rules are summarised from primary sources so you know what to ask about; how they apply to your campaigns is a question for your own counsel.

What WFM actually does

Four jobs, in order:

Forecast the contact volume you expect, by interval, usually in 15- or 30-minute blocks. Schedule people against it, respecting contracts, skills and availability. Account for shrinkage — the gap between hours paid and hours on the phone, covering breaks, training, sickness and everything else. Then manage intraday, because the forecast is wrong and someone has to react.

Every part of that assumes capacity is lumpy, must be committed in advance, and is expensive to change. That assumption is what an AI agent breaks.

What changes when some of your capacity is a setting

Capacity becomes elastic and immediate. Adding a person means recruiting, training and scheduling. Adding AI capacity means raising a concurrency limit. The lead time collapses from weeks to minutes, which removes the main reason forecasts had to be accurate far ahead.

Shrinkage does not apply — to that part. An AI agent has no breaks, no sickness and no adherence problem. It also has no bad days. The shrinkage calculation still applies in full to your human team, and that team is now smaller, which makes it more sensitive to a single absence, not less.

The bottleneck moves to escalations. This is the part that catches people out. If the AI handles 70% of calls, your humans are not handling 30% of the same easy work — they are handling the 30% that is hardest, longest and most complex, because that is precisely what escalates. Average handling time for your human team goes up after deployment. A forecast that assumed the old AHT will under-staff you.

Your forecast gains a new input, and it is unstable. Containment rate now sits between call volume and required headcount. Forecast 1,000 calls at 70% containment and you staff for 300; if containment drifts to 55% you need 450, a 50% increase in required people from a change nobody scheduled. Containment moves when configuration changes, when call mix changes seasonally, and when underlying models are updated. Treat containment as a forecast input you monitor weekly, not a constant you set once.

Occupancy stops meaning what it meant. Classic WFM targets roughly 85% occupancy because people burn out above it. An AI agent can run near 100% of its concurrency limit indefinitely. Applying a human occupancy target to AI capacity buys headroom you do not need; applying an AI utilisation target to people is how you lose staff.

Diallers: the constraint is legal, not technical

A predictive dialler places more calls than it has people to answer, betting that some will not connect. When the bet is wrong, somebody picks up and nobody is there. That is an abandoned call, and it is regulated — differently in different places.

United States

Federal rules cap it explicitly. Under 47 CFR § 64.1200(a)(7) a telemarketer may not "abandon more than three percent of all telemarketing calls that are answered live by a person, as measured over a 30-day period for a single calling campaign", with the rate calculated separately for each successive 30-day period if a campaign runs longer.

The definition is stricter than it first sounds: a call is “abandoned” if it is not connected to a live sales representative within two (2) seconds of the called person’s completed greeting — two seconds from the end of their "hello", not from connection. The rules also require you to "disconnect an unanswered telemarketing call prior to at least 15 seconds or four (4) rings", and to play a prerecorded identification and opt-out message, with an automated opt-out mechanism, when no representative is available. Source: 47 CFR § 64.1200, read 23 September 2026.

United Kingdom

Here the commonly quoted figure appears to be out of date. Ofcom's revised persistent-misuse policy, published December 2016 and enforceable from 1 March 2017, is consistently reported to have removed the 3% abandoned-call safe harbour, leaving no rate that is treated as acceptable by default and no threshold you must cross before enforcement can follow.

A deliberate caveat on that last paragraph. Ofcom's own publication could not be retrieved when this page was written — the site returned an access error to us on 23 September 2026 — so the statement above rests on consistent industry reporting of the policy rather than on the regulator's own wording, which is a weaker source than the US paragraph above. The direction of the change is well attested; the precise current position is not something you should take from this page. Confirm it with Ofcom, or with your own counsel, before you set a dialler policy on it. Sources: callcentrehelper.com reporting on Ofcom's persistent misuse statement, read 23 September 2026; Ofcom's own document not accessible at time of writing.

The practical point survives the caveat. "The 3% rule" is a US federal rule, and dialler marketing routinely cites it as though it were a universal and permanent standard. It is neither: it is one jurisdiction's rule, and the UK position has moved since. If you run UK campaigns and treat 3% as a given allowance, you are relying on a figure that is at best unverified for your jurisdiction — which is a reason to go and check, not a reason to relax.

What this has to do with AI

First, the distinction that matters most: all of the above governs outbound dialling. These are rules about calls you place. An AI agent answering your inbound line is not governed by them, and carrying the 3% figure across to an inbound deployment is the easiest mistake to make on this subject. If your AI agent only answers, this section does not apply to you.

For outbound, an abandoned call happens because no human was free at the moment the caller answered. If an AI agent can take that call, the connection does not have to be dropped — which changes the compliance position, not only the efficiency one.

One genuine open question, which you should put to counsel rather than to a vendor: the US rule is written around connection to a live sales representative within two seconds. Whether an AI agent satisfies that is not something this page can answer, and any vendor telling you confidently that it does should be asked to show you where. Treat it as unresolved.

The WFM market consolidated, and most buyers have not noticed

Thoma Bravo completed its acquisition of Verint and combined it with Calabrio, its existing portfolio company, to form a single organisation, announced on 26 November 2025. Verint states the combined business serves more than 80 of the Fortune 100. The combined group subsequently moved forward under the Verint name. Source: verint.com press room, read 23 September 2026.

Two of the names on most WFM shortlists are now one company under one private-equity owner. calabrio.com/pricing currently redirects to verint.com. If your evaluation shortlist was written before late 2025 and has both on it, it is really a shortlist of one, and you should refresh it before you treat those as competing quotes.

What it costs, where it is published

Sources: zendesk.com/pricing, cloudtalk.io/pricing, verint.com — all read 23 September 2026.

The pattern is the same one that shows up across this category: the capability you are shopping for is an add-on to an add-on. A CloudTalk Essential seat that needs a power dialler and conversation intelligence is €29 + €15 + €9 = €53 per user per month, nearly double the headline. Budget from the configuration you will actually run, never from the tier price.

Note also what the last two rows imply. CloudTalk prices seats and sells an AI voice agent by the minute, because the two things are not the same unit. Once part of your volume is handled per-minute rather than per-seat, a per-seat WFM licence is being bought for a shrinking population.

When you do not need WFM software

You have fewer than about 15 agents. Below that, a spreadsheet and a shared calendar genuinely work, and WFM licensing plus the administration it creates will cost more than the scheduling inefficiency it removes.

Your volume is flat. WFM earns its money on variability. A line with the same load every weekday does not need forecasting software to tell you to staff it the same way every weekday.

You have just deployed an AI agent and not yet measured containment. Buying WFM now means forecasting against a number you do not have. Run for a quarter, get a containment rate and a revised human AHT, then buy.

You are buying it to police adherence. It will do that, and it is a poor reason. The return in this category is in forecast accuracy, not in monitoring people more closely.

Brilo's position, and the number it does not publish

Brilo publishes a free tier at 10 minutes a month, Pro at $149 for 600 minutes with overage at $0.16, Growth at $499 for 2,500 minutes at $0.14, and a custom tier above 5,000 minutes. Source: brilo.ai/pricing, read 23 September 2026.

What it does not do: Brilo is a voice agent. It is not a workforce management platform — no forecasting, no scheduling, no adherence — and it is not a dialler. If you need to roster a human team, you need one of the products above, and Brilo does not replace it.

And the gap that matters most on this page: Brilo does not publish a concurrency limit. Concurrency is the entire capacity question for an AI agent — it is the equivalent of asking how many agents are on shift — and the pricing page quotes minutes, agents and workspaces but not how many calls can be handled at the same time. Minutes tell you the monthly budget; concurrency tells you whether the phone gets answered at 10am on your busiest Monday. Ask for the number in writing, and ask what happens to call 21 when your limit is 20 — queue, busy signal, or voicemail. That question applies to every vendor here, and Brilo is not currently one of the ones that answers it up front.

Related reading

For a predictable surge rather than the ongoing planning discipline, see how to scale call centre capacity for seasonal peaks, which covers concurrency and the four steps to take in the month before a peak. For what the escalation pool should be measured on once it exists, see AI call centre quality assurance. For modelling the spend, see AI call centre costs, pricing and ROI.