An inbound assistant that clears FAQs before a human picks up

Support desks spend a large share of their time on the same authentication steps and frequently asked questions before any real problem gets solved. For an operation that measures itself in headcount, the question is whether a voice assistant can absorb that front-end load, verify the caller, answer the routine query, and route the genuinely complex case to a person, so that human time is spent where judgement is actually needed rather than on repetitive gatekeeping.

Vendors often express the payoff as full-time-equivalent (FTE) capacity, which sounds concrete but hides the assumptions underneath. An honest evaluation converts that framing into per-call economics the buyer can measure directly: what a call costs today, which call types the assistant genuinely closes, and how much human time each automated call still consumes for edge cases. The operational task is turning an FTE claim into something a small desk can verify from its own logs.

PolyAI reports that Atos deployed an inbound voice assistant handling FAQs, authentication and security checks, tone adaptation, and routing of complex queries to human agents.

According to PolyAI, the assistant reduced agent call volumes by approximately 30% and handles the workload of 50 to 95 FTEs at roughly 50% of the cost of a single FTE. No publication date is stated, so these figures carry no date and should be re-verified before being used in any internal business case.

Rather than accept the FTE headline, an Indian SMB should decide what to measure from its own data. Capture cost per handled call today, the share of calls that are pure FAQ or authentication, average human handle time by call type, and the tail of calls that need escalation. Then ask which of those buckets the assistant actually closes end-to-end versus merely triages, because a call it routes onward still consumes agent time and should not be counted as removed.

Inspect the assumptions behind '50-95 FTEs' and '50% of an FTE': ask what call volume, shift pattern, and cost base define an FTE in that example, since none of those may match a small Indian desk. Verify how authentication and consent are handled — confirming applicable rules, consent, retention and escalation design with qualified counsel or compliance owners — how the assistant behaves with multilingual callers and code-switching, and whether tone adaptation degrades on regional accents. Confirm what happens to escalated calls during peak hours when human capacity is thin.

Atos is a large enterprise services organisation whose call volumes make an FTE-equivalence framing meaningful; a range as wide as 50 to 95 FTEs is only intelligible at that scale. A small Indian desk running a handful of agents cannot map that band onto its own staffing, and the '50% of an FTE cost' comparison rests on a labour-cost and volume structure that differs entirely from a smaller business, so the ratio is not portable as stated.

The reported 30% reduction assumes a call mix heavy in standardised FAQ and authentication traffic and a language environment the assistant was tuned for. Indian callers switching languages mid-call, smaller and spikier volumes, and a higher proportion of relationship-driven queries can all shrink the automatable share. Because the FTE framing bundles many hidden assumptions, treating it as a target rather than as a prompt to measure your own per-call economics would be a mistake.