EPAM Systems — Sequencing an Americas Commercial Transformation

Topic

Growth Strategy, Capital Allocation, Pricing Strategy

Industry

IT Consulting & Other Services

Internal ID

S&F-STR-216

Source

Strategy & Finance

Publication date

5 October 2026

Pages

5

License

Not specified

Download case PDF (5 pages)Request teacher’s pack

Setting: October 2026
Information cutoff: October 2, 2026, 4:00 p.m. ET
Case convention: The leadership decision is hypothetical. EPAM has not disclosed the meeting, budget, vote or alternatives described here.

The decision

Assume EPAM's hypothetical Americas leadership team must decide how to sequence its commercial transformation over the next two quarters. Management has described some North American clients shifting away from discrete engineering tasks and seeking broader AI-enabled modernization proposals; that explanation is not independently established by the filings. EPAM must decide whether to scale a domain-led, business-case selling motion across the region now; test that motion in selected accounts and verticals, expanding only after evidence of conversion, delivery quality and economics; or preserve selected time-and-materials (T&M) engineering work while adding commercial capability selectively.

The choice concerns sales and delivery capabilities, account selection, evidence gates and the pace of change. It is not a decision to replace engineers wholesale or to move the company to outcome pricing. Public information does not reveal a transformation budget, account-level pipeline or project economics. Recommend a sequence, specify what evidence should release the next stage, and state what would cause you to change course.

A growth challenge with several explanations

EPAM reported second-quarter 2026 revenue of $1.415 billion, up 4.5% year over year as reported and 3.4% in organic constant currency. GAAP operating margin was 10.8%, compared with 9.3% a year earlier; non-GAAP margin was 16.4%, compared with 15.0%. The company lowered its full-year revenue growth outlook to 3.2%–4.2% reported and 2.0%–3.0% organic constant currency. It repurchased $409 million of shares in the first half. The figures frame growth, margin and capital allocation; they do not show transformation costs, AI-specific margin or the relative return of reinvestment and repurchases. [P1]

The geographic picture differs depending on the measure. In segment reporting, Americas revenue was $792.262 million, up 0.6%; Europe revenue was $622.505 million versus $566.043 million, up 10.0% reported and 8.5% constant currency. Americas represented 56.0% of company revenue. Segment operating profit in the Americas rose 15.1% to $149.1 million, with segment margin of 18.8% versus 16.4%. Management attributed the margin improvement primarily to cost optimization. A separate client-location series reports $805.4 million in the Americas (56.9% of revenue) and $582.2 million in EMEA (41.2%). Do not combine client-location revenue with segment revenue. Neither geographic series establishes that a sales model caused growth or margin changes. [P2]

Vertical trends also resist a single explanation. Americas Financial Services grew 15.3% in Q2 and 13.0% in the first half; the filing cites increased spending from one large wealth-management client as well as insurance, asset-management and payments clients. Americas Software & Hi-Tech declined 10.8% in Q2 and 6.6% in the first half; Life Sciences & Healthcare declined 1.3% and 1.0%; Business Information & Media declined 2.4% and 0.1%. In Europe, Software & Hi-Tech grew 21.5% in Q2 and 31.8% in the first half, while Life Sciences & Healthcare grew 45.1% and 40.2%. Client mix, regional demand and commercial execution could all matter. A strong result in one region does not establish a transferable playbook. [P2]

Exhibit 1. Selected second-quarter segment results

Source and limits: EPAM Form 10-Q, “Americas Segment” and “Europe Segment,” including vertical tables. Segment reporting differs from revenue by client location. Management attributed Americas margin improvement primarily to cost optimization. Revenue growth is not evidence of a particular sales model's causal effect. [P2]

The offer and the delivery model

EPAM's 2025 Form 10-K reports $4.364 billion of T&M revenue and $1.063 billion of fixed-price revenue on total revenue of $5.457 billion. T&M was 80.0% of revenue, calculated from those reported amounts. T&M is billed by time-based rates. This mix indicates a substantial installed delivery model; it does not mean that T&M work is unprofitable, unsuitable for AI, or destined to disappear. The filing does not disclose a shift to outcome-based fees. [P3]

Delivery economics are not visible at the level needed to choose a commercial model. Q2 cost of revenue was $985.2 million, or 69.6% of revenue versus 71.2% in the prior-year quarter. The filing says cost of revenue includes salaries, benefits, stock compensation, travel and subcontractors; delivery compensation remains a cost whether employees are billable or not. Management attributed the lower ratio primarily to compensation declining as a share of revenue. No AI-specific utilization, redeployment, rework, client acceptance or project-margin information is disclosed. [P2]

An August 6 third-party transcript rendering attributes to EPAM's CEO an explanation that North American clients were reducing certain manual testing, UX and front-end work while seeking AI-led modernization, and that replacement work was ramping more slowly. The rendering describes a multi-quarter commercial transformation involving marketing, sales motions, client prioritization and business-case proposals. It also attributes AI-native revenue of about $160 million in Q2 and a $600 million 2026 goal to management. These are attributed statements and forecasts from a third-party transcript, not figures disaggregated in the filed release or 10-Q; the webcast/audio was not independently checked. The explanation could describe a lasting demand shift, a temporary timing mismatch, or both. [P4]

EPAM and OpenAI announced an Advanced Partner relationship on July 28, describing enterprise deployment work involving forward-deployed engineers. This is evidence of a partnership announcement, not exclusivity, signed backlog, client savings or deal margins. Accenture and Cognizant also describe significant AI programs and partnerships in their public reporting. Their measures, scale and reporting periods differ, so they indicate competitive activity without providing a like-for-like benchmark. [P5–P7]

What can be learned from AI and contract research?

AI productivity studies do not provide a dependable forecast for EPAM. Cui and coauthors' final Management Science article reports three randomized access experiments at Microsoft, Accenture and an anonymous Fortune 100 company, involving 4,867 developers. Its preferred instrumental-variables estimate is a 26.08% increase in completed tasks (standard error 10.3%) from Copilot use among compliers/users induced by access; it is not the intention-to-treat effect of access. The final article's full text and experiment-level details were unavailable in the research reviewed for this case. A distinct 2024 working-paper version covered 1,974 developers and used preliminary pull-request/build measures. Those earlier-version outcomes are not results of the final study. Neither version estimates EPAM's project productivity, margin, client value or workforce needs. [A1, A1a]

Other evidence underscores variation by task. In a preregistered study of 758 BCG workers on simulated consulting tasks, Dell'Acqua and coauthors found higher speed, output and quality on tasks inside the tested AI capability frontier, but lower correctness on one task outside it. METR's 2025 randomized study of 16 experienced open-source developers completing 246 tasks in familiar repositories found AI-allowed tasks took 19% longer (95% confidence interval: 2% to 39% longer). METR's later study had serious selection and time-measurement limitations and uncertain estimates. These settings and outcomes differ; together they support task-level testing and quality controls, not a universal productivity multiplier. [A2, A3]

Contract studies are similarly limited. A small freelancer field experiment, an observational study of 100 projects at one offshore vendor, and a single-bank contract portfolio study examine different settings and outcomes. They suggest that project uncertainty and contract design can interact, but they do not establish that EPAM should abandon T&M or prefer fixed-price work. [A4–A6]

Three sequences to consider

The following alternatives are hypothetical strategy packages.

1. Scale the domain-led motion now

Build a broader Americas capability to diagnose client business problems, assemble domain and engineering teams, and present modernization cases across a wider account base. This could improve reach while clients reconsider discrete work and competitors promote large AI programs. It risks hiring and coordination costs before conversion, repeatability, delivery quality and economics are known. Regional revenue growth does not identify which accounts or verticals will respond.

2. Pilot in selected accounts; gate expansion

Choose a limited set of accounts and verticals, test an integrated business-case proposal and delivery approach, then expand only when predefined evidence supports doing so. Candidate measures include qualified-opportunity conversion, time to contract, realized scope and contribution, client acceptance, quality/rework, utilization and redeployment. Pilots may produce more useful evidence with contained exposure, but narrow samples may not predict other clients, verticals or larger transformations. A slow test could also forfeit opportunities or under-resource a viable shift.

3. Preserve selected T&M engineering while adding capability

Retain T&M engineering where client demand, delivery performance and account relationships support it, while adding domain, consulting and business-case skills selectively. This may protect valued work and give account teams time to build a broader offer. It may also preserve a sales approach that is less suited to emerging client needs and create coordination problems between commercial and delivery teams. Public contract mix alone cannot determine which work to retain.

The alternatives can be combined, but a combination still needs priorities, decision rights and expansion gates. A credible recommendation distinguishes a test from a broad rollout and defines how sales claims will be checked against delivery results.

Assignment

Prepare a recommendation to the hypothetical Americas leadership team. Choose a sequence for the next two quarters and address:

1. Which accounts or verticals should be included first, and why? How will you avoid mistaking client mix or market demand for an effect of the selling motion?
2. What capabilities must sales, domain experts and delivery teams build together? How will client problem definition connect to feasible engineering scope?
3. Which measures and thresholds should govern continuation or expansion? Include conversion and timing, delivery quality and client acceptance, realized economics, utilization or redeployment, and risks of gaming or small samples.
4. What work should remain T&M, and how should scope uncertainty affect commercial terms? Do not infer a contract rule from the 80.0% historical T&M share or the academic studies alone.
5. What internal data are needed to refine the decision, and what evidence would make you change course?

State the strongest argument against your recommendation. Do not assume a private budget, vote, AI productivity uplift, client outcome or engineer-replacement plan.

Sources and cutoff

• [P1] EPAM Systems, Q2 2026 Form 8-K, Exhibit 99.1, August 6, 2026. Results, outlook and first-half repurchases. SEC release.
• [P2] EPAM Systems, Form 10-Q, quarter ended June 30, 2026, filed August 6, 2026. “Cost of Revenues”; MD&A “Americas Segment,” “Europe Segment,” and vertical tables. SEC filing.
• [P3] EPAM Systems, FY2025 Form 10-K, “Revenue” and Note 2 contract-type table, F-30. SEC filing.
• [P4] EPAM Q2 2026 earnings-call transcript, August 6, 2026, third-party StockAnalysis rendering, prepared remarks and Q&A. The official webcast/audio was not independently checked; management statements and forecasts are attributed, not filed or independently verified. Transcript rendering.
• [P5] EPAM and OpenAI, “EPAM and OpenAI Partner to Help Enterprises Unlock Higher Value Through Applied AI,” July 28, 2026. EPAM announcement.
• [P6] Accenture, Q3 FY2026 earnings release and presentation, June 18, 2026. Release.
• [P7] Cognizant, Q2 2026 results, July 29, 2026. Issuer release.
• [A1] Cui, Demirer, Jaffe, Musolff, Peng and Salz (2026), “The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers,” Management Science. Final publisher abstract and supplement listing. Publisher record.
• [A1a] Cui et al. (2024), “The Productivity Effects of Generative AI: Evidence from a Field Experiment with GitHub Copilot,” working paper, DOI 10.21428/e4baedd9.3ad85f1c. §§1–5, design, Tables 1–4 and caveats. Version read.
• [A2] Dell'Acqua et al. (2026), “Navigating the Jagged Technological Frontier,” Organization Science 37(2):403–423. Abstract and accessible working-paper method descriptions. DOI.
• [A3] METR (2025), “Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity,” and February 2026 update. Study; update.
• [A4] Jørgensen and Grov (2021), “A field experiment on trialsourcing and the effect of contract types on outsourced software development,” Information and Software Technology 134, 106559. Abstract and method/results text. Publisher article.
• [A5] Gopal and Koka (2010), “The Role of Contracts on Quality and Returns to Quality in Offshore Software Development Outsourcing,” Decision Sciences 41(3):491–516. Abstract/metadata. DOI.
• [A6] Fink and Lichtenstein (2014), “Why project size matters for contract choice in software development outsourcing,” ACM SIGMIS Database 45(3):54–71. Abstract/metadata. Accepted manuscript record.

All information is limited to sources public by October 2, 2026, 4:00 p.m. ET. The decision and alternatives are hypothetical; no post-cutoff outcomes are included.

Back to strategy case studies