Industries / Finance and business services
Call centers and BPO
Call centers and business process outsourcing (BPO) firms answer customers' calls, chats and emails and do routine office work such as processing forms and updating accounts, either in-house for one company or as contractors for many, often from lower-cost countries.
1965
The Birmingham Press and Mail, a newspaper in England, installed a GEC PABX 4 telephone exchange with automatic call distribution, which passed incoming calls to rows of agents at their own positions; the trade publication Call Centre Helper calls it "the earliest example of a call centre we can find in the UK".
42
Milestones achieved across 6 domains.
17
To watch for: announced, not yet confirmed outside.
13
Open deliverables. Nobody has shipped them.
In one minute
Where it started
The earliest call center on record opened in 1965, when the Birmingham Press and Mail in England installed an exchange that handed incoming calls to rows of agents; AT&T's toll-free 800 service (1967) and Rockwell's call distributor at Continental Airlines (1973) then let American firms build large telephone rooms. Contractors grew up to run this work for other companies, among them France's Teleperformance, founded in 1978. From the 1990s the work moved to lower-cost countries: a British Airways unit in India founded in 1996 became WNS, General Electric's back office there became Genpact, and by 2010 the Philippines employed more call center workers than India, by its industry association's count.
Where it stands
The Philippine trade body IBPAP says its IT and business process industry earned more than $40 billion from exports in 2025 and employed about 1.9 million people, and India's NASSCOM puts India's business process revenue at $59 billion for the year to March 2026. The two largest firms, France's TP (€10.2 billion in revenue in 2025) and Concentrix of the US ($9.8 billion), each employ roughly 450,000 to 490,000 people, most of them in India, the Philippines and Latin America. AI agents now take many simple contacts: TP's shares fell as much as 29% in one day in February 2024 after Klarna said its assistant did the work of 700 agents, and TP cut its 2026 outlook in November 2026, citing a faster-than-expected fall in content moderation work; yet only 20% of service leaders in a Gartner survey had cut agents because of AI, and Klarna began hiring people again in 2025.
What is open
Most claims about AI in this industry come from vendors testing their own products on private data, so most open jobs are public tests anyone can rerun: AI quality scoring against several human graders, voice agents with real consenting callers, a reach-a-human check under Spain's new law, card numbers leaking through speech-to-text, and speech recognition for Philippine, Indian and African English on phone-quality audio. Others are open tools for work still done by hand or inside closed software: a bake-off of routing rules on public call data, a checker for dialer logs against US and UK rules, a forecasting benchmark for call arrivals, roster test cases that encode real labor law, staffing math for queues where a bot answers first, and tests of AI agents keying forms into mainframe screens or turning one screen recording into a working automation. All of them can be built with today's speech, language and computer-use models on synthetic data or with consenting volunteers.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Routing
Routing is the work of getting each call, chat or message to the right queue and the right agent, and for outbound work, of running the dialers that place calls within the rules on how they may do it.
7 · 3 · 2
Last achievedSigned caller ID2021
Watch forPairing callers by predicted outcome
2
openAgents and coaching
This is the work of the people who answer customers' calls and chats: how they are hired, trained and coached, where they work, why they leave, and the software that now prompts them as they talk.
7 · 3 · 2
Last achievedSmaller gains in other firms2026
Watch forAccent translation for agents
2
openSelf-service
Self-service is letting customers get answers or finish tasks without a person, through phone menus (IVR, interactive voice response), speech recognition, help pages, chatbots and AI voice agents, under rules on telling people they are dealing with a machine and letting them reach a human.
7 · 3 · 2
Last achievedSpain guarantees a human2025
Watch forAI agents replace support staff
2
openQuality
Quality work checks how well each contact was handled: recording calls and chats, scoring some or all of them against a checklist, searching recordings for patterns, keeping card numbers and personal data out of what is stored, and asking customers what they thought.
7 · 3 · 2
Last achievedSatisfaction hits nine-year low2024
Watch forAI scorers against human labels
2
openBack-office processing
Back-office processing is the work that outsourcing firms and in-house shared service centers do away from the phones: typing data in, checking forms and claims, matching invoices to orders, checking new customers' identity documents (KYC, "know your customer"), reviewing posts against a platform's rules, and labeling data for AI.
7 · 3 · 2
Last achievedA test for software-driving agents2024
Watch forAgents settle small insurance claims
2
openStaff scheduling
Staff scheduling, or workforce management (WFM), is forecasting how many calls and messages will arrive in each half-hour, working out how many agents are needed to answer them within a target time, building shifts and breaks that cover that need within labor rules, and adjusting them during the day.
7 · 2 · 3
Last achievedReal call data, in public2005
Watch forSoftware agents adjust staffing
3
open
Advances to build with
3 research notes

Mistral Large 4
A trillion open parameters
Mistral published its Large 4 AI model, a trillion numbers by its count, free to download and sell, though firms that rent out models need a separate license.

Muse Ember
The jobs agents repeat
Meta's Muse Ember matched Claude Haiku 5.5 on an outside firm's tests at a third of the cost per task, though both list the same prices for text in and out.

Microsoft ThinkingBox
All twenty times
By Microsoft's own count, 50 steps of training took an AI model from passing 38 of 507 test tasks on all 20 tries to passing 107.
Open in call centers and bpo
3 of 13 open deliverables
Open · Routing
A bake-off for routing rules
A public benchmark that replays real call center arrivals through a simulator and scores several routing rules on the same calls: first come, first served to the longest-idle agent; fixed skill priorities; a speed-weighted priority rule; UCB-QR; and an open version of outcome-based pairing (predict the result of each caller-agent match, then pick the best). Add synthetic call sets with a known "true" caller-agent fit built in, so each rule's share of the possible gain can be measured against a known answer.
Built with what exists
Technion SEELab's USBank (about 1,000 agents) and AnonymousBank call logs, free for academic use after notifying the lab; the UCBQR simulator on GitHub (it lists no license, so ask the authors before reusing it); Python with SimPy or Ciw for event-by-event queue simulation.
Open · Agents and coaching
An open lab replication
A preregistered online experiment in which paid participants play support agents for a fictional shop, answering AI-simulated customers drawn from a fixed set of cases with known correct outcomes. Half get an AI reply assistant and half do not; the study measures correct resolutions, time per case and policy errors, split by each participant's skill at the start.
Built with what exists
τ-bench retail tasks (open source); an open-weights model for both the customer and the assistant so others can rerun it, such as Mistral Large 4 (see the Note "A trillion open parameters"); Prolific for paid participants; OSF for preregistration; oTree or a small web app.
Open · Self-service
Real callers for voice benchmarks
Run the open τ-voice customer service tasks (retail, airline, telecom) with consenting human volunteers calling the agent over a real phone line, in place of synthetic callers, and publish how the scores differ from the simulator's on the same tasks.
Built with what exists
Sierra's open-source τ-voice framework and tasks; a voice agent on a phone line through SIP, such as OpenAI's Realtime API; standard 8 kHz G.711 phone audio; ElevenLabs Scribe v3 to transcribe calls for review (see the Note "Fewer word errors"); a one-page written consent form for every caller.