What is the real impact of AI on operational performance in 2026?

Introduction – From available technology to captured value

The lesson of electrification

Around 1900, manufacturers replaced their steam engines with electric motors. Productivity gains only appeared two decades later, when factories were redesigned around the motor: workshops reorganised around the production flow, lines rethought, jobs reconfigured. The motor was never the point; the factory was.

AI is following a comparable trajectory, on fast-forward. The engine is becoming a commodity: models are converging, the cost of inference is falling fast, and companies are already arbitrating between providers as they would between energy sources. Differentiation will therefore not be built on the engine, but on what the company does with it: the clarity of its strategy and value proposition, the redesign of its processes, the quality of its data, the ability of its teams to work differently. It is a business transformation, not a tooling choice.

The companies winning with AI have not won on technology, but on integrating it into their means of production. That is what explains the lead of AI-native scale-ups: not better AI engineers, but young processes, clean data and teams used to changing methods. And that is what makes AI, in a large group, an ExCom topic rather than an IT department project, under joint business and tech sponsorship.

Part I – Why now? P&L impact at scale in 2026

P&L impact is about to change scale in the most agile companies; large groups face the scaling wall

1. 2026 – P&L impact is materialising, but unevenly

Three time arcs in twenty-four months

• 2024 – individual use. GPT or Claude adopted spontaneously, each person on their own. Real personal gains, but little value captured by the organisation.
• 2025 – exploration. POCs multiply, the first agents appear, and budgets rise considerably. But use cases remain confined, little adopted, rarely industrialised.
• 2026 – scaling and profitability. The impact on the P&L is already visible among AI-natives. A double pressure is setting in: AI costs have become an identified line item, and boards expect quantified impact.

Spectacular gains, concentrated in AI-native organisations

Where the organisation is agile, versatile and built around data, the gains are considerable and time-to-market collapses.

Alan, a French Next40 scale-up, gives a public measure of this: 40% of its members' requests are resolved instantly by its AI assistants – up to 70% for simple requests – with satisfaction stable at 4.5 out of 5. Over the same period, membership grew by 34% with a constant member-relations team, and overall productivity rose by around 30% in one year.

"For us, AI is a company objective, not a side project: 75% of the Growth team uses it every day, and we went from 5% to 80% of visuals produced by AI in two quarters." (AI Growth Lead, Consultant, Scale-Up – Fintech sector)

The real challenge: scaling in large groups

In a large group, complexity is structurally higher – deeply entrenched processes and reduced capacity for change, broader and more fragmented data, more teams to train, and non-negotiable compliance and cybersecurity requirements.

2026 has provided several public illustrations of this, including among leading players.

• A major French industrial group (CAC 40) froze all of its AI funding 12 months after launching its use-case factory, in order to refocus its North Star, narrow the scope to 2 business units and fix the data first.
• Starbucks withdrew its computer-vision inventory system from its entire North American estate, nine months after deploying it, due to lack of reliability.
• Klarna: after replacing around 700 agents with AI in 2024, the company began rehiring humans in 2025 and switched to a hybrid model – AI for volume, humans for complexity. (Bloomberg, Reuters)

Sources: Reuters and CNBC (Starbucks); Constellations experience (CAC 40 industrial group); business press (Amazon)

"The ExCom wants results right away, not 'wait eight months while we clean the data'." (Performance Director, Mid-cap – Industrial sector)

"We launched a POC to analyse support emails, then stopped it: the data wasn't ready yet. We preferred to make it reliable first rather than run AI on incomplete databases." (Jacob, AI Lead, European Telecom Infrastructure Operator)

Real victories are showing the way among large groups that are moving forward by treating AI as an evolution of their production apparatus.

• Unilever: more than $1.7bn of value generated through its AI supplier-management platform (100,000+ suppliers, 190 countries), with purchasing costs down 4% and supply disruptions down 17%. The best P&L example outside banking.
• L'Oréal quadrupled its product development speed by focusing AI on predicting molecular efficacy.
• Goldman Sachs: 3 to 4x engineering productivity through coding agents (Devin + Claude) across around 12,000 developers, mainly on legacy modernisation.

Sources: consumer goods trade press; company communications.

2. The cost of AI: a line item to manage "by design"

The cost of AI has become a C-suite topic in less than a year. Details in Appendix A.

The AI bill has become variable and hard to manage: pricing has shifted from per-user subscriptions to pay-per-use consumption, and agents consume fast. It hits the P&L on several lines at once – investment, operating expenses and cost of sales – which makes it hard to attribute.

Unit prices fall, the bill rises: the cost per query is collapsing, but consumption is growing faster. Reasoning that is twice as long can cost four times as much, and a chain of agents left to run autonomously propagates its errors (up to 17.2x without a coordinator, Google DeepMind 2025).

The right indicator is not the price of a token, but the cost of a useful answer (cost-of-pass): a query that produces a relevant solution, compared with the same task done by an expert. For a financial close, the typical order of magnitude is around ten dollars versus a week of an accountant's time.

"Our goal was to set up a frugal, agnostic, multi-platform AI Operating Model, with data sovereignty." (AI Lead, Consultant, Large Cap – Telecom sector)

On this scope, a fully French alternative was evaluated with the same tooling and switching criteria defined in advance: independence from providers is tested, not assumed.

The levers for keeping costs under control come down to five verbs: stop uses that add no value, route each task to the model that suits it, cache what repeats, govern rights and caps, train teams not to hand a deterministic problem to a probabilistic model. Technically, this requires a multi-model architecture that arbitrates between performance, security, data sovereignty and cost.

• Uber: 2026 AI budget exhausted by April, leading to a cap of $1,500 per month per coding tool. (FT, AI costs)
• Constellations field insight – On a client master dataset, the preliminary check of site activity was handed to a read-only agent on the data warehouse. A hypothesis formulated by a business expert was tested across the entire database in twenty minutes: 67 sites were wrongly listed as active. Execution cost: less than $8 of inference and less than one cent of queries – the order of magnitude of a cost-of-pass for this type of check.

AI impact: an accelerated J-curve – gains materialise after a phase of redesign and adoption

This is the lesson of electrification: general-purpose technologies require underlying investments – process redesign, upskilling, reorganisation – that are hard to value. Productivity can dip at first, then takes off.

"This moment when gains are slow to come is actually the most decisive: it's where you separate the use cases that matter from those that will never make a difference. The right response is not to scale back ambition, but to sharpen the course, reprioritise, and consolidate the foundations – data, process, integration – on which everything else will rest. Organisations that pull back here mistake an investment phase for a failure. Those that hold firm, with discipline, are the ones that capture the value afterwards." (Louis Ravier, Partner, Constellations)

Detailed figures on employment, adoption and maturity are provided in Appendix B.

3. ROI: reallocating resources towards growth

At this stage, return on investment shows up as acceleration, wider scopes and a growth focus, rather than as headcount reduction.

Jobs are not being cut en masse: time is being freed up, and then reinvested – or allowed to leak away if no one manages it.

IKEA – Ingka Group: from call centre to revenue channel

Billie, the chatbot of Ingka Group, IKEA's main entity, handles around 47% of customer service requests – 3.2 million conversations and nearly €13 million in operational savings. Rather than cut headcount, the group retrained its 8,500 call-centre agents as remote interior design advisers: analysing the requests the bot could not resolve had revealed latent demand for advice, invisible as long as teams were absorbed by transactional work. The remote sales channel created in this way generates €1.3bn in revenue, around 3.3% of group sales, with a target of 10% by 2028.

Sources: Ingka Group, Reuters, Harvard Business Review. Public figures – see methodology note.

Measuring impact: the hardest exercise

Time saved is hard to track. During the adoption phase, you build the reflex first and calibrate fine-grained measurement afterwards. One company we surveyed dedicated a full-time person solely to estimating impacts; another gave up on bottom-up tracking.

In a Next 40 fintech scale-up, one person is dedicated to estimating the impact of AI, in order to measure, then decide what can be industrialised, rescoped or stopped. (AI Lead, Consultant, Scale-Up – Fintech sector)

"At this stage we've stopped tracking impacts bottom-up. We just do a top-down pre-analysis before launching the use case."(AI Lead, Consultant, Large Cap – Private Equity sector)

"AI doesn't change the rule of any transformation: measurement must not delay the launch, it must frame it. A quantified target set by the sponsor before starting, an ROI that serves first to prioritise and then to correct, real impact tracked over time. Without that, you don't know whether you're fighting the right battle – and you get neither team buy-in nor the budget for the next wave." Louis Ravier, Partner, Constellations

A race won first on growth

The first player in a sector to convert AI into a net gain on its value proposition and business model will gain an edge for several years. Hence the priority given to growth, broadening the offering and customer value, rather than headcount reduction – there will always be time to activate operational gains.

This is what Jevons' paradox describes: when efficiency increases, usage increases even more. In companies that genuinely automate, people are not laid off, they are hired – and each new hire produces what several FTEs used to produce.

Part II – How? Driving the AI transformation, from POC to Use-Case Factory

Moving from POC to scale is at least as much an organisational and model question as a tech one. Here is what three years of work with large French groups and mid-caps has taught us about how to structure an AI Use-Case Factory, and how to get it adopted.

1. The dual Business x Tech requirement

10/20/70: algorithms account for 10% of the work, the tech stack 20%, people and processes 70%.

ExCom sponsorship – Business x Tech

AI is not a Tech or IT programme: it involves strategy, the business model, governance, operations, then teams and their adoption. Cross-functional sponsorship is therefore essential: it guarantees the prioritisation of use cases, settles functional trade-offs and drives adoption once delivery is complete. Technology, for its part, holds the platform: the data and agent foundation, integration with the information system, quality, security. Neither can carry it alone: trade-offs are escalated to the executive committee.

"Deploying AI at scale is an ExCom matter: decisions are made at the highest level, and you need people who can interact with C-level executives. Aligning business and tech sponsors is not an organisational precaution, it is the programme's condition for existence. That's why 100% of our AI engagements are under shared sponsorship, and 100% of our teams combine strategy consultants and tech profiles from the best agencies." Louis Ravier, Partner, Constellations

2. The 4 Ps: the prerequisites before deploying the factory

Our experience converges on four business prerequisites, none of which is purely technological. They determine whether AI remains a commodity or produces a visible effect on the P&L: Purpose, Process, Pure data, People.

Purpose – One AI North Star guiding every business and tech deployment

Key issue: Formulate the vision and the North Star underpinning the changes targeted for customers and for the company: business model, offering, target operating model, P&L impact. The AI North Star becomes the common thread running through all deployments.

Deliverables:

• AI vision and North Star carried by the ExCom.
• Shared narrative through group-wide communication.
• Prioritisation of use cases by impact x feasibility, arbitrated at ExCom level.

Process – AI does not fix broken processes: fix & accelerate where impact is highest

Key issue: AI does not make up for a broken process; it accelerates and optimises the highest-impact processes. Prioritisation takes into account P&L impact and time saved, professionalisation and moving upmarket, and costs – internal and external FTEs, capex, opex.

Deliverables:

• Process prioritisation matrix.
• High-level scoping of changes to the selected processes.
• Associated technical and AI challenges, focused on the selected processes.

Pure data – Clean once with the business, then keep it clean by design

Key issue: Clean data, backed by a single source of truth and maintained through applied rules. Cleaning relies on AI as much as possible, together with business teams – sales, finance, operations – who alone can settle disputed cases, sometimes right down to the plants or the accounting files. Correction rules then become permanent controls, from data entry and throughout the flows.

Deliverables:

• Data corrected in every master dataset.
• Single source of truth and data lineage established.
• Joint business and tech data governance.

People – No operating model shift without full adoption

Key issue: Adoption is the decisive challenge for most groups. It determines the real evolution of the operating model and day-to-day ownership of the tools, processes and target strategy.

Deliverables:

• Transformation players identified: adoption leads, a guild of evangelists.
• Coaching and hypercare on the first use cases.
• Communication of the overall transformation project.
• Updated job descriptions and target organisation.

"A broken process, even accelerated by AI or reinforced with fifty people, will remain a broken process. You don't need to wait until 100% of your processes are best-in-class before getting started, but selecting your priority use cases without first scoping the processes at a high level is a waste of time and money." Louis Ravier, Partner, Constellations

"We put an enormous amount of money into AI, we hit a wall because the relevance isn't there, and then we end up coming back to invest in data. We'd have saved time doing it the other way round." (AI Lead, Consultant, CAC 40 – Beauty sector)

3. Use-Case Factory – a group-wide transformation programme

For five years, Constellations has supported large groups through their strategic, tech and AI transformations.

We draw on these programmes, steered at the highest level, to propose a governance framework for AI transformation. This framework is not a single method, but a set of reference points for moving from POC to an operational factory in less than 6 months: how to structure, how to organise, how to prioritise.

3.1 Structuring the factory: three components

Whatever the organisation chosen, the factory rests on joint business and tech sponsorship, co-accountable for the result, and on three components.

Business lead: a decision-making body

• Decisions driven by the business: priorities, specifications and functional requirements, ROI measurement, go/no-go.
• A capacity for fast decisions: the body exists to decide, on a weekly rather than monthly rhythm.
• Clear rituals: a living prioritised backlog, a regular Business x Tech arbitration committee.
• Ongoing steering: pace-setting, removal of blockers by a dedicated intervention team, planned reversibility.
• An owner (sponsor) named before each launch: accountable for the ownership and maintenance of the use case.

Constellations field insight – A complete data catalogue was configured in half a day in a test environment. The entry cost of a POC is now counted in hours: the effort has shifted from building to ownership and maintenance.

This shift changes the nature of sponsorship. Tech teams now know how to build fast; only a strong business sponsor can arbitrate priorities, appoint an owner and drive adoption over time. Without that relay, this POC remained a demonstration.

Tech: an engine room that capitalises

• Delivery: build and industrialisation of use cases and agents, quality, security, debt.
• Reusable assets and shared guardrails: components, prompts, connectors, evaluations, trust protocol. This is what makes it possible to do something a hundred times instead of once, and what distinguishes a factory from a series of POCs.
• Documented context: descriptions, glossary, business rules. Documentation is as much a performance lever as a cost lever and a governance obligation.
• A 360° view of AI tools and uses: to safeguard architecture, security, consistency of use and cost control.

Constellations field insight. At a European telecom infrastructure operator, we measured the effect of documentation on the reliability of a natural-language analysis agent. With identical questions and tooling, the share of correct answers rose from 40–60% with the semantic layer alone to over 80% once a context layer was added: descriptions, glossary, business rules. The agent improved faster with the maturity of the documentation than with new model versions.

"The procedure is the asset, the prompt is disposable. A model can be replaced in a few weeks; skills, controls and context are the anchor point." Aurélien Rayer, Partner Tech AI, Constellations

Change & adoption: a guild

• Communication around the company project: AI presented as a shared objective, not as just another tool.
• A network of evangelists: a group-wide guild, the people arm of the factory.
• Coaching rather than training: coaches who hold the pen on the first use cases, in hypercare mode, then support the move to scale.
• Documentation as a vehicle for knowledge transfer: AI makes documentation both more complete, because it costs little to produce, and far more accessible, because anyone can query it in natural language rather than reading through it. The team's knowledge no longer rests on a few people (or on external ones!).

"Coaching rather than training. A day of training doesn't get people up to speed – it's like Excel. You need coaches who hold the pen on the first use cases, then learn how to scale." Performance Director, Mid-cap – Industrial sector

Constellations field insight. Within a data team, two unanticipated effects were observed. The first is induced upskilling: with the agent producing the first draft, practitioners spend more time reviewing, checking and arbitrating than writing. They thus exercise, earlier on, a judgement usually reserved for senior profiles – provided the review remains systematic. The second is better knowledge transfer: documentation, now cheap to produce, is written systematically and kept alongside the code. Context no longer depends on the person who created it, and the team's memory becomes collective.

3.2 Organising the factory: two axes, four archetypes

A Use-Case Factory borrows most of its functions from the Transformation Office: company-level prioritisation, a body that steers and sets the pace, a shared production capacity, and a roll-out by business line. The form it takes, however, depends on the organisation it sits within.

• The composition of the Use-Case Factory: Business lead | Tech Engine Room | Adoption, integrated or separate
• Its position in the organisation: ExCom level or BU level

[Add image here: four-archetype matrix]

No archetype is universal. Each one responds to an organisational configuration, and can evolve with its needs.

1. AI Transformation Office

Integrated – ExCom level

A single body, reporting to the ExCom and cutting across BUs, brings together the business lead, tech and adoption.

[Add image here]

2. Federated AI Squads

Integrated – BU level

Each BU has full squads that carry their use cases end to end. A simplified steering layer handles common standards, budget allocation across BUs and transformation tracking.

[Add image here]

3. AI Portfolio Office

Separate – Business lead at ExCom level

Business lead and adoption are held centrally. Tech is delivered within the BUs, on a common platform and under a central design authority.

[Add image here]

4. Shared AI Engine

Separate – Business lead at BU level

Tech is pooled centrally. BUs hold the business lead and, most often, adoption; adoption can also be entrusted to a central guild attached to the platform.

[Add image here]

4. From POC to Use-Case Factory

Structuring a Use-Case Factory closely resembles setting up a Transformation Office: prioritisation at ExCom level, a unit that steers and sets the pace, a shared engine room, and a roll-out by business line.

4.1 Running the factory as a Transformation Office

Joint sponsorship, co-accountable for the result. Technology carries delivery: build, quality, security, debt. The business keeps the trade-offs and adoption: priorities, functional requirements, go/no-go, bringing teams on board.

A Use-Case Factory run as a Transformation Office, which prioritises on an ongoing basis, sets the pace, removes blockers with a dedicated intervention team, and plans reversibility.

• A capacity for fast decisions: the body exists to decide, on a weekly rather than monthly rhythm.
• Clear rituals: a living prioritised backlog, a regular Business x Tech arbitration committee.
• Documenting embedded AI: at software vendors, it gets switched on through updates, not purchases. Without an up-to-date register, it escapes steering.

An AI engine room, detailed below, which holds the reusable assets and shared guardrails: components, prompts, connectors, evaluations, trust protocol. This is what makes it possible to do something a hundred times instead of once, and what distinguishes a factory from a series of POCs.

Constellations field insight – What survives a change of model is not the model: reusable skills, automatic checks before going into production, documented context, working conventions, evaluation sets. A model can be replaced in a sprint; these assets represent a year of accumulated practice and remain the client's property.

"The procedure is the asset, the prompt is disposable." (Algorithmic Developer, Consultant, Mid-cap – Telecom Infrastructure sector)

A central guild for adoption, a single group-wide network of evangelists. It is the people arm of the factory, separate from the decision-making body.

4.2 The engine room: two models

The pipeline is the same everywhere: framing the expected outcome, scoring, design, build, evaluation and pilot, decision, industrialisation and handover. What changes is the wiring.

Model A – centralised. Suited to mid-caps, single-business groups and launch phases. A single factory centralises prioritisation, delivery, platform, guardrails and ROI tracking; the guild radiates out into the teams. Consistency, fast upskilling, shared costs. It saturates if demand explodes, and may over-serve some BUs.

Model B – light central hub with relays in each business line. Suited to large groups. A light central core brings together the Transformation Office, the engine room's design authority and the guild. Each BU has its own relays: a Transformation Office contact, a dedicated engine room, a guild relay. The Transformation Office and the guild remain single and cross-functional, and arbitration remains central to avoid fragmentation. Business proximity and scalability, at the price of central governance that must be strong.

[Add image here]

Part III – Productivity levers in the age of AI: concrete cases

The value of AI on a process breaks down into four levers. The first two capture time; the last two capture quality and simplicity.

• Remove – eliminate the task. Repetitive, high-volume or rule-based work is handled end to end; human time is redeployed to cases that require judgement.
• Accelerate – speed up what exists. Remove interfaces and re-keying, process more volume at constant quality, and above all unblock the bottleneck that slows down the whole chain.
• Secure – make execution reliable. Controls move into the flow rather than after the fact, and AI absorbs unstructured formats and exceptions that traditional automation couldn't handle.
• Simplify – lighten the deliverable. When intelligence moves into the interaction, heavy artefacts built to compensate for rigid tools lose their reason to exist.

Most pilots only go after the second lever, speeding up what exists. Lasting gains come from reading all four.

This framework is not new: top-tier strategy consulting firms have used it for decades to break down a process. We have adapted it to the considerable changes made possible by AI and, above all, documented it with concrete examples from our AI transformations with clients in France and across Europe.

[Add image here: the four-lever tree]

1. Remove tasks

AI handles repetitive, high-volume or rule-based work end to end, and redeploys human time to the cases that require judgement.

• Level 1 support resolved autonomously: 68% of cases handled, quality maintained at 82%.
• Procurement · Autonomous long-tail negotiation

Automating the negotiation of purchases below the governance threshold – a process nobody was managing. Mixed workflow: persistent engine (Restate/Temporal) + AI agents, human-in-the-loop before sending. Order of magnitude: ~€1bn of purchases per year, threshold ~€5,000, ~2,000 buyers; capturing 2% would represent a considerable gain (hypothesis). (Procurement Lead, Consultant, CAC 40 – Environmental Services sector)

• A management controller's monthly reporting turned into an agent built by the controller themselves, without an IT ticket.

2. Accelerate tasks

Remove the interface

AI can absorb the translations, re-keying and hand-offs between two steps or two systems. It reduces breaks in the flow and brings the input directly closer to the expected result.

AI shortens the distance between intention and action. The interface becomes conversational, contextual, sometimes invisible.

Improve efficiency

This is often the first lever activated, because it can be anchored in existing processes. AI makes it possible to process far more volume in the same time, reducing the time needed per operation and increasing the overall capacity of the process.

Remove the bottleneck

Value often comes less from widespread automation than from applying AI at the precise point that slows down the whole process. Speeding up the bottleneck speeds up the entire chain.

• Marketing due diligence compressed from three to four weeks to about one week – the blocking step in a deal.
• Monthly close and reporting generated on demand, instead of a recurring choke point in the financial calendar.

3. Secure tasks

Zero defect from scratch

AI makes it possible to build controls directly into execution rather than detecting errors after the fact. Quality becomes continuous, with anomalies identified and corrected much earlier.

• Revenue assurance detecting unbilled services in real time: around 1.5 points of EBIT recovered.
• Contract review flagging non-compliant clauses at scale: thirty times more terms detected.

Reduce complexity

AI can handle unstructured formats, exceptions and multiple rules without everything having to be standardised first. It therefore makes it possible to simplify processes historically designed around rigid systems.

• Document due diligence on non-standard formats, where traditional automation failed: ten to fifteen documents processed in minutes.
• Marketing mix modelling run with 0.5 data scientists instead of 2, with fully automated production.

4. Simplify the deliverable

When intelligence moves into the interaction, the deliverable itself can become lighter, less structured, sometimes unnecessary. We stop building heavy artefacts to compensate for the rigidity of tools.

• Fewer complex interfaces to develop: dialogue with AI replaces part of the front end.
• A dynamic brand page that adapts to the visitor replaces rigid, over-built journeys.

5. Use cases by function, mapped to each lever

Finance

Remove · Automated, self-service flash reporting
Multi-plant monthly consolidation (Excel and PowerPoint), which took about a day of a controller's time, is now handled by an agent built by the business team in four half-days. The output is produced in 1h15, in self-service.
(Industrial, Mid-cap, Constellations client)

Accelerate · Remove interface · Files without re-keying
An agent automatically reads and reconciles over 80% of a file's data between documents and systems, eliminating re-keying and back-and-forth between the business and the tools. Around 30 minutes saved per file.
(Banking, CAC 40, BNP Paribas)

Accelerate · Improve efficiency · Augmented invoicing
An agent gathers and reconsolidates sources (accounting, ERP, operational data), checks the consistency of each line (services actually delivered, prices, discounts), flags discrepancies and pre-validates invoicing before it is issued. On a multi-entity monthly cycle, the check goes from about 5 days to about 1 day, with fewer credit notes and disputes.
(B2B Distribution, SME, Constellations client)

Accelerate · Remove bottleneck · On-demand ad hoc reporting
A context layer combined with AI generates, on demand, the specific reports requested by the ExCom, without manual data reprocessing. The French subsidiary publishes its reports by the 2nd of the month, versus the 7th or 8th for other countries.
(B2B Services, Mid-cap, Constellations client)

Secure · Zero-defect · Revenue assurance
On reliable data, an agent predictively detects services in production that have not been billed, as well as recurring revenue-leakage patterns. First workstream: around €1.5m recovered on a scope of around €80m, i.e. nearly 1.5 points of EBIT.
(Telecom, Mid-cap, Constellations client)

Secure · Reduce complexity · Customer master data (One SIRET)
An algorithm for rationalising the customer master data, designed with AI, makes the data reliable before it is used. 15K records processed in 12 minutes against the full SIRENE database; automatic acceptance above a 95% score, with the rest sorted into review batches.
(Telecom, Mid-cap, Constellations client)

Simplify · Talk with your data
Teams query data in natural language and get the answer or the chart directly, instead of building and navigating dashboards. The fixed deliverable fades behind the question.
(Telecom, Mid-cap, Constellations client)

Sales & Marketing

Remove · AI-assisted post-production
Post-production is now almost entirely AI-generated, and briefs increasingly so. Savings reach 5 to 10% on very large budgets, with the opportunity to bring back in-house the value captured by agencies.
(Luxury, CAC 40, Constellations client)

Accelerate · Remove interface · Self-service content production
A self-service creative factory, backed by a validated brand library, lets local teams adapt assets by prompt. The share of AI-generated assets rose from 5% to over 80% in two quarters, against an initial target of 25%; the tool was built in a 15-day sprint.
(Fintech, Scale-Up, Constellations client)

Accelerate · Improve efficiency · Ad hoc performance analysis
Google Ads search-term analysis is done with AI in about 30 minutes for three versions, versus half a day previously – a factor of 4 to 6.
(HealthTech, Scale-Up, Constellations client)

Accelerate · Improve efficiency · Sales meeting follow-up
A tool analyses the meeting transcript, identifies objections and drafts the follow-up using the battle cards. Measured time saving is 30%, provided the sales knowledge base is well documented.
(Fintech, Scale-Up, Constellations client)

Accelerate · Remove bottleneck · Creative variations
Around 60 variations are produced per concept (static and video), versus about ten previously, with no observed drop in performance.
(HealthTech, Scale-Up, Constellations client)

Secure · Zero-defect · Reporting and incrementality testing
The body of analyses (reporting, incrementality tests) is automated; monitoring of all advertising accounts against reference rules is being rolled out.
(Luxury, CAC 40, Constellations client)

Secure · Reduce complexity · Marketing mix modelling
Data collection is automated by integrating less structured sources, and modelling is supervised by 0.5 data scientists (vs 2 historically), at around half the cost of an outsourced MMM (about €40k).
(Luxury, CAC 40, Constellations client)

Simplify · Generative Engine Optimization (GEO)
Targeted content and internal linking increase how often the brand is cited by conversational engines: +80% on the category worked on, versus +20% natural growth, in a 6-to-8-week pilot.
(Fashion retail, Mid-cap, Constellations client)

Simplify · Adaptive brand website
The brand website adapts to the visitor and includes a conversational search dedicated to gifting, with hand-off to an adviser when needed. Hypothesis being tested: around +5% in sales in a pilot market.
(Luxury, CAC 40, Constellations client)

Private Equity

Remove · Automated marketing due diligence
A complete outside-in audit (positioning, UX, acquisition, martech) is carried out end to end by agents.
(Mid Cap, Private Equity, Constellations client)

Accelerate · Remove interface · Due diligence without the interfaces
Agents plug directly into the data room and query all documents simultaneously: no more opening, reading and summarising file by file, or navigating the VDR interfaces.
(Large Cap, Private Equity, Constellations client)

Accelerate · Improve efficiency · Augmented origination
Agentic monitoring draws on external data and, above all, the fund's proprietary data (expert calls, qualified notes) to identify targets before intermediaries do. Around 1.5 FTEs saved and pipeline up 15%, currently being evaluated before scaling.
(Large Cap, Private Equity, Constellations client)

Accelerate · Remove bottleneck · Opportunity qualification
A scoring agent qualifies opportunities downstream of sourcing and harmonises the reading grid for deals: 3 to 4 FTEs freed up in a team of 50.
(Large Cap, Private Equity, Constellations client)

Secure · Zero-defect · Controls built into due diligence
An agent monitors the data room, checks each new document against the checklist and generates the list of red flags every night (customer concentration, change-of-control clauses, hidden liabilities) for the morning review. 100% of the data room is audited instead of a sample.
(Large Cap, Private Equity, Constellations client)

Secure · Reduce complexity · Vendor due diligence
A first VDD tool has been tested: it saves one round of analysis, without replacing advisory firms. Longer-term target: become the main tool for screening and validating targets on the smallest build-ups.
(Mid Cap, Private Equity, Constellations client)

Simplify · Synthetic consumer studies
Synthetic panels replace part of consumer surveys. The approach is still in the testing phase.
(Large Cap, Private Equity, Constellations client)

Tech

Remove · Autonomous network operations
An agent triggers more than 100 remediation actions on its own and brings the handling of a major network incident down from a few hours to about one minute (over 95%), with 237,000 events processed in 2026; 12 use cases in full autonomy, 400+ in production.
(Telecom, Large group, Constellations client)

Accelerate · Remove interface · Steering tool designed by the business
For a project-steering tool rolled out to 30,000 employees, business teams designed their own mock-ups through an Airtable and Claude integration, cutting the scoping phase by about three months.
(Shipping, Large group, Constellations client)

Accelerate · Improve efficiency · Software and data development
AI code: 5 to 7x productivity on agent-assisted development (practitioners' estimate), with plenty of room for improvement. The roughly 50% gain in the first month assumes several weeks of prior upskilling.
(Telecom, Mid-cap, Constellations client)

Accelerate · Remove bottleneck · Vibe-coded business tools
Business teams design their own tools without a tech team – for example, a customer-experience visualisation tool that reconstructs a user's actual journey in a single view (steps completed, friction points, drop-offs), where this information was previously scattered across several tools. Prototyped without a ticket to the tech team.
(Fintech, Scale-Up, Constellations client)

Secure · Zero-defect · Quality by default
Agents generate descriptions and tests across the entire raw data layer, reviewed unit by unit: ~95% of models documented and tested, 100% of sources under freshness control, drift detected automatically. Tests and documentation become systematic because they cost almost nothing.
(Telecom, Mid-cap, Constellations client)

Secure · Reduce complexity · Conversational maintenance
Teams ask questions in natural language; AI groups similar multilingual cases and proposes the solution, absorbing a heterogeneous history. -30% in maintenance costs on production lines.
(Industrial, Large group, Siemens Senseye)

Simplify · Buying within the conversation
The existing commerce system is connected to ChatGPT, Gemini or Copilot without replatforming: the transaction happens within the exchange, the e-commerce interface fades away and IT capacity is freed up.
(Commerce, Tech leader, commercetools AgenticLift)

Ops

Remove · Autonomous long-tail negotiation
Purchases below the governance threshold, previously left unmanaged, are negotiated by AI agents orchestrated by a persistent workflow engine, with human validation before sending. Annual gain of 2% on a scope of €1bn of purchases.
(Environmental services, Large group, Constellations client)

Accelerate · Remove interface · Prioritised pricing validation
An agent records the validations actually given each day and, using explicit rules, deduces what can be pre-approved. The validator no longer goes looking for requests in the tool: they receive an already-prioritised list of needs and only decide what requires it.
(B2B Distribution, Large group, Constellations client)

Accelerate · Improve efficiency · Level 1 ticket handling
Once decision trees, the FAQ and ticket analysis are equipped with AI, level 1 is automated and teams focus on levels 2 and 3.
(Telecom, Mid-cap, Constellations client)

Accelerate · Remove bottleneck · Customer service prioritisation
By prioritising the queue where the stakes are highest, the bottleneck is removed in the right place: +40% prioritisation accuracy, recurring incidents resolved 2.5 times faster, +80% first-contact resolution.
(Beauty, CAC 40, Constellations client)

Secure · Zero-defect · Sales team coaching
Agents support in-store sales advisers on product knowledge and sales practice. Impact is measured top-down at this stage.
(Luxury beauty, Large group, Constellations client)

Secure · Reduce complexity · Procurement under control
Every requisition is checked against policies, categories, quotes and price lists: reviews go from several days to a few minutes.
(Environmental services, Large group, Constellations client)

Simplify · Product assistant for store teams
A conversational agent draws on unstructured sources (database, product sheets, safety data sheets) so that store teams can answer customer questions on the spot.
(B2B Distribution, Large group, Constellations client)

‍