The agent stack thesis at BCG, examined.

What Deckster was, and what it failed to become

Deckster launched inside BCG as a direct answer to a problem every large consulting firm shared: the firm's institutional knowledge — forty years of engagement reports, proprietary frameworks, sector-specific benchmarks — existed in formats that consultants could not access at the speed a modern engagement demanded. A senior associate starting a new mandate in consumer goods needed to locate comparable engagements, extract the relevant frameworks, and understand the current thinking across the firm's Consumer practice before the first client meeting. The process took days. Deckster compressed it to hours. That was the original value proposition, and it delivered on it.

The architectural limitation became apparent when BCG's leadership began asking whether Deckster could do more than surface and summarise. The answer was no — and the reason was structural. Deckster was built on a retrieval-augmented generation architecture optimised for question-answering against a fixed corpus. It could not be directed to undertake an analytical task, decompose it into sub-components, and produce a structured output that a consultant could commit to a client deliverable. It was a search interface with a conversational layer. When Voss's team ran internal benchmarks comparing the analytical output of Deckster-assisted consultants against the output of comparable teams at McKinsey using the rebuilt Lilli platform — benchmarks conducted under a structured competitive-intelligence exercise in Q3 2024 — the gap was significant enough to constitute a strategic vulnerability. The decision to rebuild followed within six weeks.

GAMMA's design departed from Deckster's architecture at the lowest level. Rather than a single model querying a single corpus, GAMMA runs a multi-agent orchestration layer: a coordinating meta-agent that receives a task specification from a consultant, decomposes the task into parallel sub-workflows, routes each to a practice-specialised sub-agent, and assembles the outputs into a structured deliverable. The meta-agent does not retrieve documents. It manages agents. The distinction is the entire thesis.

The GAMMA platform: architecture and deployment

GAMMA's infrastructure runs on a hybrid cloud configuration — BCG's own private compute environment for client-sensitive data processing, with API access to frontier models for general reasoning tasks. The model selection is not disclosed externally, but people familiar with the build describe a multi-model approach: a primary frontier model handling complex reasoning and synthesis, a smaller, faster model for classification and routing tasks, and fine-tuned domain models for four of the firm's highest-revenue practice areas. The fine-tuning corpus is BCG's proprietary engagement archive, augmented by structured outputs from the first cohort of GAMMA-assisted engagements — a feedback loop that compounds the system's advantage with every deployment.

Isabelle Fontaine, BCG's Global Head of AI and Digital Platforms, oversaw the infrastructure build. Her team completed the core orchestration layer in Q1 2025 and began phased practice-area deployment in February. The deployment sequence was deliberate: Corporate Finance and Strategy (CFS) first, followed by Operations, then Healthcare. CFS was selected because its analytical workflows — deal screening, synergy modelling, integration planning — are data-dense and well-bounded. The inputs are structured, the outputs are structured, and the quality of the agent's work is assessable against an objective standard. Healthcare and Operations followed for different reasons: Healthcare because BCG's proprietary clinical and regulatory dataset represents a source of fine-tuning advantage its competitors cannot easily replicate, and Operations because operations mandates generate the highest volume of structured process-mapping data, which feeds back into the training pipeline at the highest rate.

As of Q2 2025, GAMMA is in production across fourteen of BCG's twenty-three practice areas. The remaining nine — including BCG's Social Impact, Climate and Sustainability, and Centre for Growth practices — are scheduled for H2 2025 deployment. Voss has set a full-firm deployment target of Q1 2026. Whether that timeline holds depends on one variable his team cannot fully control: client data governance. Several of BCG's largest clients — two global banks, a US-based pharmaceutical company, and an European industrial conglomerate — have opened negotiations about the terms under which their engagement data can be used to fine-tune GAMMA's practice agents. Those negotiations are ongoing, and their outcome will determine how quickly the system's healthcare and financial services agents achieve the performance benchmarks the firm has internally set.

Three practice agents, three different bets

The CFS agent — designated GAMMA-CFS internally — handles the analytical scaffolding of corporate strategy and M&A mandates. On a buy-side advisory engagement, GAMMA-CFS runs target screening against BCG's proprietary M&A database and live third-party financial feeds, produces a ranked long-list with strategic fit scores mapped to BCG's proprietary Value Creation Matrix, and generates a preliminary synergy model populated with sector-specific benchmarks from comparable historical transactions. The workflow that a BCG associate team historically executed over two weeks GAMMA-CFS completes in under forty-eight hours. The engagement manager's time shifts from building the model to interrogating it — running stress tests, adjusting assumptions, and applying the qualitative judgment about strategic fit that the agent explicitly flags as outside its scope.

The Healthcare agent operates under tighter constraints and delivers a different kind of value. GAMMA-Health does not produce clinical recommendations. Its mandate is regulatory and commercial intelligence: mapping a pharmaceutical client's pipeline against the current payer landscape across thirty-two markets, identifying reimbursement precedents for comparable drug classes, and flagging upcoming regulatory inflection points that a commercial team needs to plan around. The agent monitors regulatory filings, payer databases, and health technology assessment decisions in near real time and pushes structured alerts to engagement teams when a development intersects a live client situation. Priya Sundaram, a Principal in BCG's Boston healthcare practice, has described GAMMA-Health's alert function as "the regulatory watch service we used to staff with three analysts working rotating shifts." Those analysts have been redeployed to client-facing interpretation work.

The Operations agent is the most structurally ambitious deployment and the one creating the most internal debate. GAMMA-Ops handles operational due diligence and process transformation diagnostics — the analytical core of BCG's operations practice. For a supply chain transformation engagement, GAMMA-Ops integrates with a client's ERP systems via a secure API layer, runs comparative benchmarks against BCG's proprietary operations dataset covering 4,200 companies across sixteen sectors, identifies performance gaps at the subprocess level, and produces a prioritised transformation roadmap with effort-impact sizing attached. The first full production deployment, on a European automotive supplier in Q2 2025, produced a preliminary diagnostic in eleven days against a historical baseline of six to eight weeks. The engagement team spent the remaining time on the interventions the diagnostic identified — not on constructing the diagnostic itself.

McKinsey, Bain, and the race no one is advertising

McKinsey's Lilli rebuild and BCG's GAMMA initiative are the two most advanced agent deployments among the major strategy consultancies, and neither firm discusses the other's progress in external communications. Internally, both firms track the other's deployment closely. The competitive dynamic is unusual: the primary battleground is not client acquisition but talent. The first firm to demonstrate, to MBA recruiting candidates, that its agents handle analytical drudgework at a level that frees junior consultants for higher-leverage activity will win the quality competition for the analyst class. BCG's internal recruiting narrative for the class of 2026 specifically references GAMMA as evidence that junior consultants on BCG engagements will spend their time on "judgment work, not Excel work" from the first week of engagement. McKinsey makes a structurally similar argument about Lilli. The candidates evaluating these claims will make them verifiable or falsifiable within eighteen months of joining.

Bain's position is more cautious and more strategically coherent than it appears from the outside. Bain has deployed internal AI tooling across its analytics function — the firm's proprietary Advanced Analytics Group runs a suite of internal models for client work — but has not announced an agent-orchestration rebuild equivalent to GAMMA or Lilli. People familiar with Bain's technology strategy describe a deliberate choice to wait for the orchestration market to mature before committing to a full re-architecture. The argument inside Bain, as it has been characterised to INTELAR, is that the first-mover advantage in agent infrastructure accrues to the firm with the best proprietary training data, not the first to build the orchestration layer. Bain's proprietary data corpus — strong in private equity and healthcare — is narrower than McKinsey's and comparable in depth but not breadth to BCG's. The Bain calculus, on this reading, is to let McKinsey and BCG incur the rebuild costs, observe which architecture compounds most effectively against the real constraint, and build the second version rather than the first.

That strategy has a clear risk. The agent systems McKinsey and BCG are building improve with use. Every engagement produces structured output that feeds back into the training pipeline. By the time Bain begins its own orchestration rebuild, the gap in system performance — driven by eighteen to twenty-four months of additional fine-tuning data — may be large enough that catching up requires acquiring a firm that already has the corpus rather than building one. Several mid-market analytics consultancies with strong proprietary datasets are, by multiple accounts, receiving early-stage acquisition interest from firms whose primary motivation is the training data rather than the client list.

What clients are actually experiencing

Three BCG client deployments in H1 2025 have become internal reference cases for GAMMA's production performance. The first, with a global consumer packaged goods company headquartered in the Netherlands, involved a sixty-three-market brand portfolio rationalisation — an engagement scope that would historically have required a forty-person team over twenty weeks. Under the GAMMA-CFS and GAMMA-Ops integrated deployment, a fourteen-person team completed a preliminary portfolio recommendation in nine weeks. The client's Chief Strategy Officer, reviewing the output with the BCG engagement partner, described the analytical granularity as "a level of market-by-market detail we would not have expected until week fourteen of a conventional engagement." The total fee was 38% below the comparable legacy scope. BCG's margin on the engagement was, by internal calculation, the highest the Consumer practice had recorded on a mandate of that complexity.

The second reference case, with a US-based specialty chemicals manufacturer, deployed GAMMA-Ops on an operational turnaround mandate. The client was facing margin pressure from a combination of input cost inflation and production yield inefficiency across seven manufacturing sites. GAMMA-Ops ran a cross-site benchmarking analysis — integrating production data, maintenance records, and procurement contracts via API — and identified a yield improvement opportunity at three sites that the client's internal operations team had not surfaced in a prior six-month internal review. The opportunity was worth an estimated $140M in annualised EBITDA improvement. The BCG team spent the first three weeks of the engagement getting the data integrations stable. The agent identified the yield opportunity on day twenty-two. The engagement partner's summary to the client's CEO attributed the speed directly to GAMMA's ability to run parallel analysis streams that a human team would have had to sequence.

The third deployment is the one generating the most internal conversation at BCG. A sovereign wealth fund — jurisdiction not disclosed in internal case materials reviewed by INTELAR — engaged BCG for a strategic portfolio review covering 47 direct investments across infrastructure, technology, and natural resources. The GAMMA-CFS agent was deployed to run performance attribution and strategic fit scoring across the full portfolio against a set of macroeconomic and sector scenarios the client's investment team specified. The agent processed the full portfolio in four days, produced a scenario-weighted risk map, and flagged seven positions as warranting immediate strategic review based on scenario sensitivity. The client's CIO, according to engagement notes summarised in BCG's internal GAMMA performance log, asked whether BCG could leave the agent running against the portfolio on an ongoing basis rather than as a discrete engagement deliverable. That conversation is the first documented instance of a BCG client requesting GAMMA as an ongoing managed service rather than as a project tool. Voss's team is evaluating what a managed service commercial model would look like. The answer will define BCG's next strategic move more consequentially than any single engagement.

What to watch

Five signals will determine whether BCG's GAMMA rebuild converts into a durable structural advantage or surfaces the same internal pricing and partnership tensions that have complicated McKinsey's Lilli rollout.

• The client data governance negotiations. Two global bank clients and the European industrial conglomerate mentioned above are in active discussions with BCG over the terms for using engagement data in GAMMA's fine-tuning pipeline. If those negotiations resolve on terms that allow BCG to use structured engagement outputs — even in anonymised, aggregated form — the system's financial services and industrial agents will improve at a rate no competitor can match without equivalent client data access. If they stall or restrict, the advantage narrows.
• Whether the sovereign wealth fund managed service converts. A recurring GAMMA deployment at a sovereign fund level would be the clearest signal that BCG is building a software revenue stream alongside its engagement revenue stream. That is a different business from consulting. It competes differently, it is valued differently, and it attracts a different class of investor interest if BCG ever revisits the partnership structure. Watch for a public reference to "ongoing AI advisory relationships" in BCG's communications in the next two quarters.
• Bain's next infrastructure announcement. If Bain moves to an agent-orchestration rebuild before the end of 2025, it signals the firm has concluded the wait-and-see strategy is losing more ground than it conserves. If Bain instead announces an acquisition — of an analytics consultancy, a domain-specific AI firm, or a proprietary dataset business — it signals the firm is trying to buy the training corpus it needs rather than build it. Either move would confirm that GAMMA and Lilli have shifted the competitive baseline for the entire industry.
• BCG's analyst intake for the class of 2026. If BCG reduces its global analyst intake by more than 15% from the 2025 baseline — while maintaining or growing revenue — it signals that GAMMA has crossed the threshold from productivity tool to headcount replacement. That threshold is the most consequential and least discussed implication of the agent rebuild. The class of 2026 recruiting cycle closes in October.
• Alejandro Voss's next public remarks. Voss has not given a major public address on GAMMA's architecture since a brief reference at the World Economic Forum's AI working group session in January 2025. A detailed public disclosure — at Davos, at a technology conference, or in a BCG publication — would signal the firm has decided that transparency about GAMMA's capabilities is now a competitive asset rather than a liability. The absence of disclosure to date has been deliberate. Its end, when it comes, will mark a strategic inflection.