Tech Mahindra AWS Agentic AI CoE: The Pilot-to-Production Gap Play

Tech Mahindra (NSE: TECHM) announced the launch of its Amazon Web Services (AWS) Agentic Process Transformation (APT) Center of Excellence (CoE) on September 3, 2026, in Pune. The Tech Mahindra AWS Agentic AI CoE is designed to accelerate enterprise adoption of agentic AI by combining Tech Mahindra’s Business Process Services (BPS) expertise with AWS cloud and AI capabilities — specifically Amazon Bedrock Agents, Amazon Connect, and AWS Step Functions. The press release positions the CoE as a “scalable AI execution engine”. As it helps enterprises move from AI experimentation to measurable business impact.

The question is whether a CoE inside an IT services company can close a gap that research firms attribute to infrastructure, governance, and organizational readiness failures, not technology capability gaps.

Layer 2 — Competitive Positioning: The IT Services Agentic AI Race

Tech Mahindra is not the only Indian IT services firm betting on agentic AI. The competitive landscape among India’s top five IT services companies is intense, and each has taken a different strategic approach:

TCS (Tata Consultancy Services): India’s largest IT services firm with $29.08 billion in FY2025 revenue and 607,000+ employees. TCS has launched WisdomNext, a multi-model GenAI aggregation platform with an Agentic Orchestrator Workbench and 150+ industry-specific agentic solutions. In March 2026, TCS launched Rapid Outcome AI with NVIDIA targeting accelerated deployment. TCS has aligned with OpenAI and AMD for AI infrastructure, building AI-optimized data centers in India with 100 MW compute capacity scaling to 1 GW. TCS’s AI services generate approximately $1.8 billion annually (~6% of revenue). TCS also partnered with Microsoft in December 2025 to deploy 50,000+ Copilot licenses internally.

Infosys: $19.35 billion in FY2025 revenue with ~317,000 employees. Infosys has built Topaz, an AI-first platform with an Agentic Foundry of 200+ pre-built agents and Topaz Fabric with 50+ IT operations agents. It partnered with Anthropic in February 2026 to integrate Claude models for enterprise agentic workflows — a partnership that TechCrunch noted gives Infosys “access to one of the stronger model families specifically for multi-step, tool-using agents.” Infosys also partnered with AWS in January 2026, combining Topaz with Amazon Q Developer. AI-related services generated ₹25 billion (~$275 million, ~5.5% of revenue) in Q3 FY26. Infosys is a Microsoft Frontier Firm with 50,000+ Copilot licenses.

Google Cloud Partnership

Wipro: Expanded its Google Cloud partnership in August 2026 to scale Gemini Enterprise and agentic AI, launching the LIFT framework with 1,500+ Forward Deployed Engineers. Also a Microsoft Frontier Firm with 50,000+ Copilot licenses and 25,000+ employees upskilled. Wipro is focusing on healthcare and life sciences through its AI360 ecosystem.

Accenture: The premium-play leader. Launched a dedicated NVIDIA Business Group focused on Sovereign AI and Physical AI. Accenture’s AI Refinery platform sits on top of various clouds to manage multi-model environments. Accenture, Cognizant, Deloitte, PwC, and TCS are all partners in the AWS AI Agents and Tools marketplace launched in July 2025.

HCLTech: Partnered with OpenAI in 2025 to help enterprises deploy AI tools at scale.

Tech Mahindra’s positioning is distinctive in one specific way: it is the only Indian IT services firm that has launched a dedicated, named AWS agentic AI CoE tied to its BPS business. Competitors have AI platforms (Topaz, WisdomNext), partnerships (Anthropic, OpenAI, NVIDIA, Microsoft, Google Cloud), and internal Copilot deployments, but none have publicly announced a hyperscaler-specific agentic AI CoE focused on business process transformation.

This is a narrow but real differentiation. The BPS focus is key — Tech Mahindra is not just building AI agents for IT operations or software development. It is building agents that execute business processes (collections, payments, customer service, claims processing) on behalf of clients. This is the BPaaS (Business Process as a Service) model augmented with autonomous AI, and it is where the margin compression in traditional BPO is most acute.

Layer 3 — Public-Data Sweep: Tech Mahindra’s Financial Position and BPS Performance

Tech Mahindra’s FY2026 results provide important context for evaluating this announcement:

Financial profile. Tech Mahindra reported FY2026 revenue of ₹56,815 crores ($6.385 billion), up 7.2% YoY in INR terms but only 0.6% in constant currency — the lowest constant currency growth among India’s top five IT services firms. EBIT rose 39.2% to ₹7,152 crores ($797 million), with EBIT margin expanding 290 basis points to 12.6%. PAT was ₹4,811 crores ($537 million), up 13.2% YoY. The company declared its highest-ever dividend of ₹51 per share.

BPS segment performance. BPS revenue for FY2026 was ₹9,048 crores (~$1.01 billion), representing 15.9% of total revenue. It grew 6.3% YoY (₹8,512 crores in FY2025) — modest but positive. The BPS segment result was ₹1,499 crores, up from ₹1,192 crores in FY2025, a 25.8% improvement in segment profitability. BPS headcount was 64,330 in Q4 FY26, down from 65,450 in Q3 FY26 and 59,636 in Q4 FY25.

European Telecom Operator

Deal momentum. FY26 deal wins reached $3,794 million in TCV, up 41.6% YoY — the highest in five years. Q4 deal wins were $1,073 million. The company highlighted a large, multi-year AI-led transformation deal with a major European telecom operator where “agentic AI is embedded into the operating model via a proprietary orchestration platform to drive zero-touch operations.”

Turnaround context. CEO Mohit Joshi, who joined in 2023, has been executing a three-year transformation journey. FY2026 marked the end of the “Stabilization Phase.” The company has expanded EBIT margins for 10 consecutive quarters through Project Fortius (operational restructuring). Management has set FY27 targets of 15% EBIT margin and organic constant currency revenue growth above peer average. The company stated that 80% of its workforce is AI-enabled, 76% have completed advanced AI training, and 84% of customer-facing employees are AI-enabled.

Target Group acquisition. Tech Mahindra acquired Target Group in May 2016 for £112 million (approximately $145 million at the time). Target Group is a UK-based financial services outsourcing provider specializing in mortgage and loan originations, savings and investments, payments, collections, and in-life servicing. It manages assets of more than £17 billion and serves 50+ major financial institutions including Goldman Sachs, Morgan Stanley, Credit Suisse, Barclays, and Shawbrook Bank. Target operates as a standalone entity within Tech Mahindra.

Mobile World Congress

Prior agentic AI work. Collections Guru is not Tech Mahindra’s first agentic AI product. In March 2026 at Mobile World Congress, Tech Mahindra launched “Agentic Payment Assistance & Collections Optimization” for telecom operators, available on AWS Marketplace. This solution uses Amazon Bedrock for agentic reasoning, Amazon Connect for omnichannel customer engagement, and AWS Step Functions for workflow orchestration. In 2025, the company launched the TechM Orion Marketplace as an enterprise agentic AI marketplace. The CIO First coverage of the APT CoE launch notes that the CoE uses Amazon Bedrock Agents for multi-step task orchestration, API execution, and workflow automation.

Layer 4 — The Unasked Question: Is a CoE a Product or a Marketing Vehicle?

The most important question is whether the AWS APT CoE is a genuine product development capability or a go-to-market positioning exercise.

The evidence suggests it is both — and that tension is worth examining.

Evidence for genuine product capability: Collections Guru is a real, deployed solution with measurable outcomes. Mortgage Soup reported in June 2026 that Target Group’s deployment reduced average handling times by 30%, cut screen-toggling by 40%, and reduced dependency on subject knowledge by 40%. FinTech Wales confirmed these figures. The solution is deployed in a regulated financial services environment, which requires compliance, auditability, and governance — not just demo-quality AI. Tech Mahindra’s AWS Marketplace listing for the telecom collections solution describes specific AWS services used (Bedrock Agents, Bedrock Knowledge Bases, Amazon Connect) and specific governance features (human-in-the-loop controls, explainability, auditability).

Evidence for marketing positioning: The CoE announcement comes five months after the MWC launch of the telecom collections solution on AWS Marketplace — which was itself described as part of the same agentic AI program. The CoE reframes existing product work as a new strategic initiative. The press release does not mention the Orion Marketplace, the MWC launch, or any other prior agentic AI work — presenting the CoE as a new beginning rather than a continuation. The “measurable business impact” language is repeated throughout but the only metric provided is the 40% efficiency gain from a single deployment at a subsidiary.

Worth Scrutiny

The “CoE” framing itself is worth scrutiny. In IT services, Centers of Excellence are a well-established go-to-market construct. They signal investment and capability to clients without committing to specific product deliverables. A CoE is not a product, a platform, or a guarantee — it is an organizational structure that may or may not produce repeatable solutions. Forrester’s June 2026 report on agentic AI noted that “platform confusion freezes commitment while teams argue over whether to bet on a SaaS agent, an SI-built system, or a custom build.” A CoE is an SI-built system approach — and its value depends entirely on execution.

The press release says the CoE “will allow enterprises to deploy industry-specific AI solutions” across telecom, healthcare, BFSI, retail, and manufacturing. Five industries. One documented deployment. The gap between the ambition (five industries) and the evidence (one use case in BFSI collections at a subsidiary) is the gap the CoE needs to close.

Layer 5 — Honest Translation: What the Claims Mean

“Agentic AI.” The press release uses this term repeatedly but does not define it. In the current market, “agentic AI” means AI systems that can plan, reason, and execute multi-step tasks with some degree of autonomy. Gartner distinguishes genuine agentic systems from “agent-washed” products — chatbots, RPA tools, and workflow automation relabeled as agents. Of thousands of vendors claiming agentic capabilities, Gartner estimates only approximately 130 offer genuine agentic features. Collections Guru appears to qualify — it gathers and interprets case data, provides recommendations, and automates case reviews, which is genuine agentic behavior in a bounded domain.

“Scalable AI execution engine.” This is marketing language for a repeatable solution framework. The claim is that the CoE can produce Collections Guru-like solutions across industries without building each one from scratch. This is the “platform vs. services” tension that defines IT services AI strategy. Infosys has the Agentic Foundry (200+ pre-built agents). TCS has 150+ industry-specific agentic solutions. Tech Mahindra has Collections Guru and a telecom collections solution. The CoE’s value depends on whether it can produce the third, fourth, and fifth solutions at scale.

Not Across The Entire Operation

“40% efficiency gains.” This is the most important number in the press release, and it deserves scrutiny. The 40% figure is described as “anticipated” and “in the areas where it’s been rolled out” — not across the entire operation. Target Group’s COO said “we’re seeing encouraging early results.” Mortgage Soup’s June 2026 report provided more detail: 30% reduction in average handling time, 40% reduction in screen-toggling, 40% reduction in dependency on subject knowledge. These are operational metrics, not financial outcomes. The press release does not mention revenue impact, cost savings in dollar terms, headcount reduction, or customer satisfaction improvements.

“Governed, enterprise-scale AI transformation.” Governance is the right word to emphasize. KXN Technologies’ 2026 enterprise research found that 78% of enterprises now require human-in-the-loop validation for Tier 2 and above decisions, and 61% cite legacy system integration as the top barrier to scaling agentic AI. Only 21% of organizations have a mature agent governance model. The CoE’s emphasis on governance — real-time compliance checks, role-based access, action logging — addresses the exact gap that Gartner identifies as a primary cause of project cancellation.

“Pilot-to-production cycles.” This is the core value proposition. Halkwinds Research found that a single agent takes 3–4 months to reach production, while multi-agent systems require 6–9 months. High-performers move from pilot to production in ~90 days; laggards take 9+ months. If Tech Mahindra’s CoE can compress this timeline for clients, that is a genuine competitive advantage.

Layer 6 — Decision-Maker Framing: Who Should Care and Why

For enterprise buyers evaluating agentic AI: The APT CoE is more credible than most IT services AI announcements because it has a shipped product (Collections Guru), a named customer (Target Group), and quantified operational metrics. If you are a BFSI, telecom, or healthcare organization already working with Tech Mahindra or AWS, the CoE provides a structured engagement model for moving from pilot to production. Ask to see the governance framework, the solution templates beyond collections, and the timeline from CoE engagement to production deployment. The 40% efficiency gain is from a controlled deployment at a subsidiary — ask for evidence of similar outcomes at non-subsidiary clients.

For Tech Mahindra investors: This announcement is consistent with CEO Mohit Joshi’s AI-first strategy and the FY27 margin target of 15%. The BPS segment, where this CoE sits, represents 15.9% of revenue and grew 6.3% in FY26. If the CoE can accelerate BPS growth by making TechM’s business process services more competitive against TCS BPS, Infosys BPO, and Wipro BPS, it supports both revenue growth and margin expansion (agentic AI-enabled BPS should command higher margins than traditional BPO). The deal win from the European telecom operator, where “agentic AI is embed into the operating model,” suggests the strategy is already generating revenue. Monitor BPS segment growth and margin in FY27 quarterly results.

Horizontal AI Platform

For competitors (TCS, Infosys, Wipro, Accenture): Tech Mahindra has made a specific bet that a hyperscaler-specific, BPS-focused agentic AI CoE is a stronger go-to-market position than a horizontal AI platform (Topaz, WisdomNext) or a broad multi-cloud partnership portfolio. If the CoE model proves effective — producing repeatable, industry-specific agentic solutions that deploy faster than custom builds — expect competitors to launch similar hyperscaler-specific CoEs. Infosys already has the deepest AWS relationship among Indian IT firms (Topaz + Amazon Q Developer partnership, January 2026); it could formalize this into a similar CoE structure.

For AWS: The CoE strengthens AWS’s position in the IT services ecosystem. AWS launched its AI Agents and Tools marketplace in July 2025 with partners including Accenture, Cognizant, Deloitte, PwC, TCS, and Wipro. Tech Mahindra’s CoE is the most AWS-specific agentic AI investment from an Indian IT services firm — it uses Amazon Bedrock Agents, Amazon Connect, and AWS Step Functions exclusively. This deepens AWS’s lock-in with Tech Mahindra’s BPS clients and differentiates AWS from Microsoft (which has Frontier Firm partnerships with TCS, Infosys, Wipro, and Cognizant) and Google Cloud (which has the expanded Wipro partnership).

What Is Genuinely New vs. What Is Repackaged

Genuinely new: The formalization of a dedicated AWS-specific agentic AI CoE within Tech Mahindra’s BPS business. No other Indian IT services firm has publicly announced a hyperscaler-specific agentic AI CoE. The integration of Amazon Bedrock Agents with BPS process expertise — combining autonomous AI workflows with human-in-the-loop governance in regulated environments — is a real capability, not just a press release. Collections Guru’s deployment at Target Group, with documented operational metrics, provides a credible reference case.

Improving: Tech Mahindra’s agentic AI product portfolio is expanding. The MWC launch (telecom collections, March 2026), the Orion Marketplace (2025), and now the APT CoE (September 2026) show a building pattern. CEO Mohit Joshi has consistently emphasized AI as the central growth vector. The FY26 deal wins ($3.79 billion, up 41.6% YoY) include AI-led transformation engagements. The European telecom deal with embedded agentic AI is particularly significant — it validates the model beyond the Target Group subsidiary.

Scale at Speed

The Question the Press Release Doesn’t Answer

The most important question is whether a CoE inside an IT services company can solve a problem that research firms consistently attribute to organizational readiness, not technology.

McKinsey found that fewer than 10% of organizations have scaled agentic AI to deliver tangible value, citing data quality as the primary roadblock in 80% of cases. Cisco found that only 13% of organizations qualify as “Pacesetters” — fully AI-ready across strategy, infrastructure, data, talent, governance, and culture. Forrester found that 88% of enterprise agent pilots never make it to production. Gartner predicts 40% of agentic AI projects will cancel by end of 2027.

The causes of failure are consistent across every research firm: legacy system integration, data quality, governance gaps, unclear business value, and insufficient risk controls. These are organizational problems, not model capability problems. A CoE that provides technology solutions (Amazon Bedrock Agents, workflow orchestration, governance tooling) addresses part of the problem but not all of it.

Augment with Agentic AI

Tech Mahindra‘s advantage is that its BPS business already operates the processes it is trying to augment with agentic AI. Collections Guru works at Target Group because Tech Mahindra owns Target Group, controls the systems, and has deep domain expertise in arrears management. The CoE’s challenge is replicating that level of process intimacy at client organizations where Tech Mahindra does not own the infrastructure, data architecture, or operational context.

The 40% efficiency gain is real. The question is whether it is transferable. The answer to that question will determine whether the AWS APT CoE is a genuine competitive moat or another pilot-stage announcement in a market where 88% of pilots never reach production.

Tech Mahindra has the BPS expertise, the AWS partnership, the financial resources, and the AI-first strategic mandate. What it has not yet demonstrated is that the CoE model can scale beyond a subsidiary deployment to multiple industries, multiple clients, and multiple regulatory environments. Until it does, the CoE is a promising execution engine with one proven cylinder — not yet the scalable AI execution engine the press release describes.