Listed call-center and BPO operators have spent 2025 and 2026 absorbing a thesis that is now well-documented in financial reporting: that generative AI will erode the headcount-arbitrage operating model that has powered the sector for two decades. As reported by the Financial Times, Bloomberg and Reuters across multiple pieces, hedge funds including **Marshall Wace**, **Point72**, **Citadel** and **Squarepoint** have built and disclosed short positions against several listed BPO operators on that thesis.
The clearest single data point: **Teleperformance** now carries the **highest disclosed short interest of any European tech-services stock**, with reported short interest growing from approximately **3.8% to 6.4%** over the relevant tracking window, and **May 2026 data** placing the figure closer to **12.24%**. **Concentrix** and **TTEC** shares have fallen materially in 2026. Indian-listed peers including **Firstsource** and **Hinduja Global Solutions** have come under similar pressure. The Andreessen Horowitz essay **"Unbundling the BPO"** (a16z, 2024) is widely cited as the canonical articulation of the venture-investor side of the same thesis.
Buyers reading those headlines reasonably ask: am I about to sign a multi-year contract with a category that is shrinking?
The same window of reporting shows a global BPO market estimated at **USD 353.6 billion in 2026**, with industry projections to **USD 741.6 billion by 2034** at a CAGR of approximately **9.7%**. Within customer-experience BPO, **Gartner has projected that 75% of customer interactions will involve AI in some form by 2026** — a figure widely cited by industry analysts as a directional ceiling on the share of volume that flows through AI-first surfaces.
These two halves are not contradictory. The market is growing **and** the operating model is restructuring. The structural pattern that practitioners now consistently describe is the **80/20 hybrid model**:
The buyers who sign well in 2026 are signing into the 20% with partners who run the 80% credibly. The buyers who sign poorly are signing into the 80% with partners whose operating margin still depends on staffing it with humans.
"AI-proof" is not a marketing claim a partner can self-certify. It is a small number of testable operating realities. The eight items below form a defensible checklist a buyer can put in an RFP or use in a due-diligence call.
Ask the partner to walk through, by name, the AI systems running in **production** on their delivery floor today. The set that matters: agent-assist (real-time prompt and answer surfacing), call summarisation into ticketing, post-call QA at 100% coverage rather than the legacy 1–3% sampled review, intent routing, and any first-tier deflection. If the conversation stays at the "we have an AI strategy" level, the partner is not in the AI-augmented tier.
A future-proof partner can show you, for an existing comparable account, the **actual split** of resolved volume between AI surfaces and human agents, the deflection rate by intent category, the **CSAT delta** between AI-handled and human-handled interactions, and the cost-per-contact breakdown for both. If the answer is qualitative, the partner has not yet operationalised the model.
The single fastest signal of operating-model maturity in 2026 is whether the partner will price **a meaningful share** of the contract on **outcomes** — resolved interactions, CSAT, first-contact resolution, conversion — rather than purely on agent-hours. Re-tooled operators accept this framing because their cost base is no longer dominated by headcount. Rebranded operators resist it because their P&L still depends on billable seats.
When AI deflection rises, the **SLA on the human handoff tightens**: once a customer has waited through an AI turn, the patience budget for the human is smaller. CET-aligned nearshore operations — Casablanca, Madrid, Lisbon, Bucharest — keep that SLA achievable 24/5 without expensive overnight shift premiums. Offshore-only delivery is harder to sustain in an AI-augmented architecture because the cost arbitrage that justified it shrinks as the deflected volume grows.
Ask specifically: how is **customer conversation data** used in model training, prompt logs, agent-assist memory, and post-call analytics? What is the **retention period**? Where does the data physically reside? Which models are hosted versus called via API? A 2026 partner should answer in writing, with named systems, vendor agreements and data-residency proof — and should be able to support **EU-only processing** on request. If the partner cannot articulate this, the regulatory and reputational risk passes through to the buyer.
Legacy SLAs (average handle time, service level on voice answer, abandonment rate) measure agent labour. AI-aware SLAs add: deflection rate on the targeted intent set, AI containment quality, escalation accuracy (did the AI hand off at the right point), and **AI-to-human consistency** (does the human resolution match what the AI committed to). A partner that has not rewritten its SLA template is selling the old contract under a new name.
Ask for an **AI-BOM**: the inventory of models, datasets, agents and prompts running in the production stack that touches your account. This is not yet legally required in most jurisdictions, but enterprise procurement is already requesting it in 2026, and the EU AI Act second wave will make it mandatory for high-risk systems from 2027. A partner that maintains an AI-BOM is operating at the level the next contract cycle will require.
Counterintuitively, the strongest signal of a future-proof partner is willingness to write **clean exit clauses**: data return formats, model and prompt repatriation where the buyer owns them, knowledge-base export, transition support, and termination notice symmetric with the buyer. Partners who fear AI displacement write sticky contracts. Partners who price on outcomes are happy to compete on renewal.
The eight points above can be folded into a normal RFP in one section. A defensible scoring grid:
A partner scoring under 60 on this grid is selling the old BPO. A partner scoring above 80 is selling into the same thesis the hedge funds are testing — but on the long side.
In production, blended cost per contact in mature AI-augmented contracts typically lands at **35–55% of the all-human baseline** once the transition is complete, with CSAT flat or improving. The exact figure depends on the deflection ceiling for your intent mix. The savings are real, but they are not the headline number. The headline is that the **risk profile** of the contract changes: less exposure to wage inflation, less exposure to attrition, more exposure to model quality and data-governance discipline.
For a deeper look at the underlying market dynamics, see our companion analysis on [AI-augmented BPO vs legacy call centers in 2026](https://callitdev.com/en/blog/ai-augmented-bpo-vs-legacy-call-centers-2026) and the [Morocco vs Eastern Europe nearshore comparison](https://callitdev.com/en/blog/morocco-nearshore-rd-hub-vs-eastern-europe-2026). For service detail, our [BPO services page](https://callitdev.com/en/services/bpo), our [AI/ML development practice](https://callitdev.com/en/services/software-development/ai-ml-development), the [Morocco delivery footprint](https://callitdev.com/en/why-morocco) and the [outsourcing cost calculator](https://callitdev.com/en/cost-calculator) cover the operational and commercial dimensions in more depth.
The short positions cited in the Financial Times and Bloomberg reporting are not a forecast that BPO disappears. They are a directional bet on **which operating model survives**. A buyer using that signal correctly reads it as a tool for shortlisting, not as a reason to insource everything. The buyers most exposed in 2026 are those locked into multi-year contracts with operators that have not yet repriced the cost stack around AI; the buyers least exposed are those signing new contracts with partners who already run the 80/20 hybrid in production and price accordingly.
Buyers locked into a legacy headcount-arbitrage contract have more leverage than is usually assumed, because most master service agreements include change-control mechanisms that can be used to introduce AI-aware terms without waiting for the renewal. A defensible 90-day plan looks like this. **Days 1–30**: instrument the existing contract — pull six months of contact data, classify volume by intent, and identify the top three intent categories that are candidates for deflection. **Days 31–60**: open a formal change request with the incumbent for an AI-augmented pilot on one intent category, with outcome pricing tied to deflection rate and CSAT delta, on a small but statistically meaningful share of volume (typically 5–10%). The incumbent's response is itself diagnostic: re-tooled partners welcome the pilot, rebranded ones stall it. **Days 61–90**: in parallel, run a short market sweep of two or three AI-augmented alternatives against the same intent category, using the eight-point checklist above as the scoring grid. By day 90, the buyer has either a credible repricing path with the incumbent or a credible alternative to put on the table at renewal — both materially better positions than waiting.
Call IT Dev runs an AI-augmented, multilingual, Morocco-nearshore BPO delivery model designed for European and North American buyers, with an explicit 80/20 hybrid stack and outcome pricing available on qualifying contracts. We are not the right partner for buyers seeking the very largest single-language headcount contracts at the lowest possible hourly rate — that is still the legacy operator territory, with the operating-model risks this article describes. We are the right partner for buyers who want the AI-augmented model on a defensible cost envelope, with CET-aligned human-tier delivery, multilingual coverage and clean exit terms. The fastest way to a concrete answer for a specific environment is a scoped discovery call against the eight-point checklist above.
It is not a political statement on AI and labour. It is not a prediction about specific listed operators. Naming Teleperformance, Concentrix, TTEC, Firstsource and Hinduja Global as having come under market pressure in 2026 is a factual description of disclosed short interest and reported share-price moves — not a recommendation on those securities. The point is the **buying behaviour** that follows from those signals, which is what a procurement, customer-operations or COO function needs to act on this year.
${CTA_BLOCK}
Yes, on the disclosed-data level. As reported by the Financial Times, Bloomberg and Reuters, funds including Marshall Wace, Point72, Citadel and Squarepoint have built short positions against several listed BPO operators. Teleperformance carries the highest disclosed short interest of any European tech-services stock, with the figure growing from approximately 3.8% to 6.4% over the relevant tracking window and May 2026 data closer to 12.24%. Concentrix and TTEC shares have fallen materially in 2026, and Indian peers including Firstsource and Hinduja Global have come under similar pressure.
No. The global BPO market is estimated at USD 353.6 billion in 2026 with industry projections to USD 741.6 billion by 2034 at a CAGR of approximately 9.7%. The market is growing and the operating model is restructuring simultaneously. The repricing reflects which operating models the market expects to survive, not the disappearance of outsourcing as a category.
AI handles the structured, repeated, deterministic 80% of contact volume (FAQ-grade billing, basic troubleshooting, status checks, intent classification, summarisation). Humans handle the complex, regulated, high-empathy, high-value 20% that resolves outcomes. Gartner has projected that 75% of customer interactions will involve AI in some form by 2026, broadly consistent with the upper bound of practical deflection on most intent mixes.
Willingness to price a meaningful share of the contract on outcomes — resolved interactions, CSAT, first-contact resolution, conversion — rather than purely on agent-hours. Re-tooled operators accept that framing because their cost base is no longer dominated by headcount. Rebranded operators resist it because their P&L still depends on billable seats.
Because AI deflection tightens the SLA on the human handoff. Once a customer has waited through an AI turn, the patience budget for the human is smaller. CET-aligned nearshore operations keep that SLA achievable 24/5 without expensive overnight premiums. Offshore-only delivery is harder to sustain because the cost arbitrage that justified it shrinks as deflected volume grows.
An AI Bill of Materials is the inventory of models, datasets, agents and prompts in the production stack that touches your account. Not yet legally required in most jurisdictions; the EU AI Act second wave will make it mandatory for high-risk systems from 2027, and enterprise procurement is already requesting it in 2026. A partner that maintains an AI-BOM is operating at the level the next contract cycle will require.
Blended cost per contact in mature AI-augmented contracts typically lands at 35–55% of the all-human baseline once the transition is complete, with CSAT flat or improving. The exact figure depends on the deflection ceiling for your intent mix. The risk profile of the contract also changes: less exposure to wage inflation and attrition, more exposure to model quality and data-governance discipline.
A defensible weighting: AI maturity 15%, hybrid model evidence 15%, outcome pricing 15%, nearshore resilience 10%, data governance 15%, AI-aware SLAs 10%, AI-BOM and audit posture 10%, exit terms 10%. A partner scoring under 60 is selling the old BPO. A partner above 80 is selling into the same thesis the hedge funds are testing — but on the long side.
CALL IT DEV — Software, AI and dedicated tech teams — Casablanca | Madrid | Dubai — contact@callitdev.com — +212-537-373777