Teleperformance trades at 5.3x EV/EBITA — 69% below its 10-year median. The market has made up its mind. We ran the numbers and disagree.
Teleperformance SE is the world's largest Business Process Outsourcing (BPO) and Customer Experience (CX) provider — €10.2 billion in revenue, 490,000 employees, 100 countries, 400+ languages. From above €400/share in 2022 to €66.54 today: an 83% collapse. Not because the business collapsed — FY2025 EBITA margin is 14.6%, free cash flow is €901m, and the company generated €1,537m of EBITA in FY2024. The stock collapsed because investors became convinced AI will make human-led customer service obsolete.
That fear is not irrational. It deserves to be taken seriously. We did. Then we built the numbers.
"At 5.3x EV/EBITA — 69% below its 10-year median — Teleperformance is priced for terminal decline. Our analysis asks one question: is that pricing rational, or is the market systematically extrapolating short-term disruption into permanent impairment?"
Q1 2026 Core Services declined -1.7% LFL — the first confirmed organic contraction in the company's modern history. Short interest sits at 6.4% of float against a sector average of 2.4%. Trust & Safety revenue is in structural decline as tech platforms (Meta, Google, TikTok) deploy AI content moderation in-house. LanguageLine faces real competition from AI real-time translation. The Klarna effect — one fintech's announcement that AI replaced the work of 700 agents — was not fabricated. These are real signals from informed participants.
What the bear case gets wrong is the speed and completeness of disruption. 81% of Teleperformance's revenue comes from financial services and government — complex, regulated, high-sensitivity interactions. A patient in a healthcare crisis cannot be served by a chatbot. A disputed insurance claim requires human judgment and regulatory accountability. The Kodak analogy requires a zero-marginal-cost digital substitute. That substitute does not exist for complex BPO within a 5-year investment horizon.
The -0.7% reported revenue decline in FY2025 masks very different underlying dynamics. The bridge below shows that excluding FX and a one-time contract loss, TP grew organically.
| Component | €m (est.) | % of FY2024 | Confidence |
|---|---|---|---|
| FY2024 Reported Revenue | 10,281 | 100.0% | Baseline |
| + ZP Better Together & Agents Only (scope) | +221 | +2.1% | HIGH |
| − TLScontact visa contract non-renewal | −200 | −1.9% | HIGH |
| − LanguageLine / Spec Services organic | −110 | −1.1% | MEDIUM |
| + Core Services organic (+2.7% LFL) | +220 | +2.1% | HIGH |
| − FX headwind (EUR vs USD/INR/TRY/EGP) | −362 | −3.5% | HIGH |
| − Hyperinflation / Other | −41 | −0.4% | MEDIUM |
| = FY2025 Reported Revenue ✓ | 10,209 | −0.7% | Reconciles |
Excluding FX (−3.5%) and the one-off TLScontact contract (−1.9%), the underlying business grew +2.1% organically. FX drag alone explains most of the "decline." Source: TP FY2025 Results, 26 Feb 2026.
The analysis was structured around a single question: what probability of AI-driven terminal decline does the current price imply, and is that probability rational? Answering it required four independent models and triangulation across their outputs.
| Model | Purpose | Key Output |
|---|---|---|
| AI Scenario Analysis | Three named scenarios (A/B/C) with probability weights, distinct financial assumptions, and named analogies. The headline model. | Blended value ~€90–110/share |
| DCF — Scenario B Base Case | 5-year FCFF model under AI Adapter assumptions. WACC 6.7% (French OAT 3.3%, ERP 5.5%, beta 0.85). GG + Exit Multiple TV. | ~€87/share (+31%) |
| Comparable Companies | Concentrix, TELUS Int., Conduent, Capgemini BPS, Accenture. Sector context: entire BPO is AI-discounted. TELUS take-private at 7–8x EV/EBITDA (2025) is the key comp. | TEP at 30–35% discount to TELUS take-private |
| LBO Model | Entry at €66.54, 50/40/10 debt structure. Embedded stress test: −10% revenue AND −200bps margin simultaneously. IRR sensitivity across 5×5 entry/exit matrix. | Base IRR ~33% / MOIC ~4.2x |
Teleperformance's EBITDA metric changed presentation basis between FY2024 (€2,096m, 20.4% margin) and FY2025 (€1,485m, 14.6% margin). This apparent collapse is a methodological change, not an operational one — FY2021–2024 included the full IFRS 16 right-of-use depreciation addback (~€550–600m p.a.). EBITA margin, used throughout this analysis, is stable across all years: 15.1% → 15.5% → 15.5% → 15.0% → 14.6%. The disclosure change was not explicitly flagged in TP's press releases — a red DD flag for transparency.
| Metric (€m) | FY2021 | FY2022 | FY2023 | FY2024 | FY2025 |
|---|---|---|---|---|---|
| Revenue | 7,115 | 8,154 | 8,345 | 10,281 | 10,209 |
| LFL Growth (CC) | 25.7% | 5.7% | 5.1% | 2.4% | 1.3% |
| EBITA ★ (primary) | 1,071 | 1,262 | 1,290 | 1,537 | ~1,490 |
| EBITA Margin ★ | 15.1% | 15.5% | 15.5% | 15.0% | 14.6% |
| EBITDA ⚠ (presentation change) | ~1,478 | 1,750 | 1,775 | 2,096 | 1,485 |
| Net Free Cash Flow | 703 | 703 | 812 | 1,084 | 901 |
| Net Debt | 2,656 | 2,609 | 4,553 | 3,974 | 3,974 |
| EV/EBITA (historical) | ~23x | ~14x | ~11.5x | ~9.5x | 5.3x |
★ EBITA margin used as primary metric — comparable across all years. ⚠ EBITDA drop FY2024→FY2025 is methodological, not operational. FY2023 ROIC compressed by Majorel acquisition leverage spike (Nov 2023, ~€3.2bn).
We built three scenarios with distinct names, analogies, and financial projections — not generic bear/base/bull. Each carries a probability weight grounded in available evidence. The asymmetry between downside and upside is the core of the investment case.
AI automation accelerates faster than TP can adapt. Clients cut outsourced headcount structurally. Trust & Safety goes near-zero. LanguageLine faces existential AI translation competition. Core Services enters sustained organic decline.
TP transitions to hybrid AI-human model. Future Forward delivers €100m+ annual savings. TP.ai FAB generates new data services revenue. Core stabilises at 2–3% LFL. Margin holds at 13–14% as efficiency offsets volume pressure.
TP.ai FAB scales into a commercially significant platform. TP becomes the integration layer for enterprise AI in customer operations. Data services grow 30%+ p.a. Outcome-based contracts drive margin expansion to 16–17%.
Probability-weighted blended value: ~€90–110/share (35–65% above current €66.54). The market is pricing AI Victim at 40–50% probability. We price it at 15%. Even if we're wrong and raise it to 30%, the blended value still implies upside.
Our Scenario B DCF (WACC 6.7%, TGR 2.5%, terminal EBITA margin 14.0%) implies €87/share — a 31% premium to today's €66.54. The current price corresponds to approximately WACC 7.5% and TGR 2.0% — pricing in conditions worse than any recession scenario for a business generating €900m of annual free cash flow.
| WACC \ TGR | 2.0% | 2.5% | 3.0% |
|---|---|---|---|
| 5.5% | €126 | €148 | €179 |
| 6.0% | €101 | €118 | €141 |
| 6.7% ★ Base | €75 | €87 | €103 |
| 7.5% | €52 | €61 | €73 |
| 8.2% | €34 | €41 | €50 |
★ Base case in amber. Current price €66.54 implies ~WACC 7.5%, TGR 2.0% — pricing near-recessionary conditions permanently.
At 5.3x EV/EBITA entry, the LBO IRR sensitivity is extraordinary. In a 5×5 matrix, almost every entry/exit combination generates acceptable or exceptional returns. The critical risk: the requested 50/40/10 debt structure leaves only 10% equity — creating covenant breach risk under the stress scenario (−10% revenue and −200bps margin simultaneously). IC should require 20–25% equity or covenant resets at close.
| Exit EV/EBITA ↓ · Entry → | 4.0x | 5.3x ★ | 6.5x | 8.0x |
|---|---|---|---|---|
| 5.5x (Scenario A) | 27% | 11% | −2% | −11% |
| 7.0x | 38% | 21% | 8% | −1% |
| 9.0x (Scenario B) ★ | 53% | 33% | 19% | 9% |
| 11.0x | 65% | 44% | 29% | 18% |
| 13.0x (Scenario C) | 76% | 54% | 37% | 26% |
★ Base case (5.3x entry, 9.0x exit): 33% IRR. Green = ≥20%, Amber = 10–20%, Red = negative. Even Scenario A exit at 5.5x generates 11% IRR at today's entry price.
The key context: Teleperformance's discount is not unique — the entire BPO sector is AI-discounted. The TELUS International take-private in 2025 at 7–8x EV/EBITDA is the most important comparable. A PE fund paid 7–8x for a BPO business with lower margins than TP and arguably higher AI exposure in its client mix. TP at 5.3x represents a 30–35% discount to that transaction. Either the PE fund was wrong on TELUS, or the public market is wrong on TP.
Largest single Specialized Services asset (~€700m est.) faces AI real-time translation and US healthcare budget pressure. Management calls it cyclical. If structural, disposal at 8–10x EV/EBITDA crystallises value and eliminates the primary risk. This is the most consequential analytical question for the investment case.
FY2025 EBITDA of €1,485m (14.6%) vs FY2024 €2,096m (20.4%) is a methodological change, not an operational collapse. EBITA margin is actually stable at 15.0%→14.6%. TP did not explicitly flag the basis change in its press releases. Any analyst reading headlines would draw incorrect conclusions. Disclosure quality: low.
€94m synergies claimed FY2024, but EBITA margin compressed from 15.5% to 15.0%. Integration costs absorbed ~€41m from holding companies. The gross/net synergy split is not disclosed. DD must audit whether the €94m is genuinely incremental or partially offsetting new costs.
McKinsey AI strategist, no public company CEO experience, leading 490,000 people across 100 countries. 9-month pre-appointment embedding provides continuity. Q1 2026 tone was analytically sharp. Primary risk: vision vs. operational execution. Watch H2 2026.
50/40/10 structure at 5.3x EV/EBITA: ~€620m annual interest. DSCR 2.5x in base case — manageable. Drops to ~1.9x in the stress scenario (−10% rev, −200bps margin). Covenant breach risk is real. IC should require 20–25% equity or covenant reset provisions.
FY2025 FCF €901m is reported with no material one-off WC releases identified. The decline from FY2024 (€1,084m) is fully explained by €362m FX headwind and ~€80m restructuring. Underlying FCF generation is robust and predictable. Use €850m normalised for LBO debt service.
Multi-year contracts (3–5yr), embedded compliance frameworks, regulatory accreditations, institutional agent training. Client retention ~85–90%+ sector norm. 1,000+ clients, no single client above 10% of revenue. The Kodak analogy fails: there is no zero-cost digital substitute for complex regulated BPO in the relevant horizon.
550+ AI projects deployed. 50+ new deals won in Q1 2026 alone. TP.ai FAB processing 4m calls, targeting 40m by YE2026. CAIO Andreas Braun (BCG/Microsoft/Accenture). €100m+ annual savings on track per Q1 2026. Data Services for AI growing double digits. This is commercial adoption, not press release AI-washing.
"Teleperformance is not Kodak. It is IBM in 2012 — a legacy business in genuine transformation, available at a distressed multiple. The question is not whether AI disrupts TP. It will. The question is whether TP can adapt quickly enough. At 5.3x EV/EBITA, you don't need a perfect outcome."
When the market prices a single narrative (AI disruption) into a 69% discount to median, it creates asymmetry: the downside is already priced, the upside is not. At 5.3x EV/EBITA, even the bear case LBO generates 11% IRR. This is what a margin of safety looks like in practice — not no downside, but bounded downside.
TP reported −0.7% revenue growth in FY2025. Excluding the €362m FX headwind (EUR strengthening), organic growth was +2.8%. Public investors price the reported number. PE investors evaluate on constant-currency economics. This creates a systematic entry opportunity: when a quality business with global revenues faces GBP or EUR strength, the reported decline overstates the operational reality.
TP's EBITDA methodology changed between FY2024 and FY2025 — a change that makes margins appear to collapse from 20.4% to 14.6%. This was not explicitly disclosed in the press release. An analyst reading only the headline draws the wrong conclusion. Disclosure quality is a separate dimension from financial quality — and in this case, poor disclosure is masking a business that is actually more stable than it appears.
Kodak's product (film) had a zero-marginal-cost digital substitute. That is why Kodak was terminal. TP's product — complex human judgment in regulated, high-sensitivity interactions — does not. 81% of TP's revenue is financial services and government. These clients will not replace human agents with AI on a 3–5 year horizon regardless of technological capability, because regulation, liability, and client sentiment all prevent it. The analogy is wrong; the conclusion doesn't follow.
If LanguageLine is disposed at 8–10x EV/EBITDA (est. €600–750m), two things happen simultaneously: (i) leverage drops from ~2.7x to ~2.1x ND/EBITA; (ii) the primary AI-disruption risk asset is removed from the portfolio. The remaining core business then trades at approximately 4.0x EV/EBITA — implying zero AI disruption premium in the price. That is the investment case in one calculation.
The complete investment memo and financial model are available for download. The Excel model includes 10 sheets — AI Scenario Analysis, DCF, LBO (with embedded stress test), Revenue Bridge waterfall, Comparable Companies, Historical Financials (FY2021–FY2025), Assumptions, DD Flags and Investment Thesis — with 286 live formulas and zero errors.
All data from public filings as of May 2026. Not investment advice.