Beyond Pilots: Why AI Strategizing in 2027 Demands a Forward-Deployed Mindset

Discover why AI pilots are no longer enough in 2027 – and how a Forward-Deployed mindset creates real value.
by
Manuel Freude

By Manuel Freude, AI Strategy Project Manager at Merantix Momentum

Walk into almost any enterprise boardroom today, and the tone has shifted sharply since the Generative AI boom. Two years ago, leadership teams applauded as business units spun up dozens of proofs of concept: knowledge-base chatbots, automated sales summarizers, and countless machine learning pilots. Today, executive boards are asking a harder question: "What measurable P&L impact did these initiatives create, and why are we still unable to scale them?"

The data from late 2026 shows how wide the gap between activity and financial return has become. In McKinsey's August 2026 State of AI survey, 80% of respondents report that AI boosts their individual productivity, yet only 37% of organizations attribute a positive EBIT contribution to AI, virtually unchanged from 2025. Only around 6% qualify as true AI high performers with at least 5% EBIT impact.

We see the reason in our client engagements, including with executive teams. When we confront people with the difference between transformation and disruption, the reaction is remarkably consistent. Most people are far more comfortable with transformation: a big program that includes everybody and keeps everyone aligned, confident, and happy. Yet the true potential of AI sits in workflow redesign, and that is usually a disruption exercise. People are asked to saw off the branch they are sitting on. That is an uncomfortable feeling, for the individual and for the team.

This is the "transformation trap": organizations choose the comfortable path, spread AI across many small initiatives, and never reach the redesign work where the P&L actually moves.

What "forward-deployed" means for AI strategy

The forward-deployed model was pioneered in Silicon Valley. Instead of working from a distance, forward-deployed teams closely work with clients to design, configure, and implement workflows and applications together with them. What defines the model is the way of working: iterating rapidly with the future users and building side by side with them, rather than in a black box.

Applied to AI strategy, this changes the order of work. Before the first sprint, our strategist clarifies in the actual workflow which measurable P&L target the initiative serves, how the process really runs today, which data foundation and governance guardrails exist, and which tasks stay with humans. Traditional consulting stops at the slide deck. Traditional IT outsourcing executes whatever specification it receives. Forward-deployed AI strategizing connects commercial direction and technical reality from day one.

To turn AI ambition into measurable value in 2027, your roadmap needs to answer four questions.

1. Where does AI move the P&L, and which workflows must be redesigned?

Too many roadmaps still begin with a tool ("Where can we deploy an assistant?") instead of a structural P&L lever. In many projects we have run structured interviews across business lines, and one thing is different from two years ago: many people now know what AI is and can name use cases that would help them. Those ideas fall into three clusters:

  • Small use cases that a large language model chat, such as Langdock or Copilot, can solve.
  • Lightweight agentic workflows that take over a defined sequence of steps.
  • Redesigned processes built on custom AI solutions, which are larger, more complex, and usually involve agentic software.

‍

The first two are the easiest to find, and they typically save an hour or two per person per week. That is real, but it is rarely what moves margins or revenue. The real work, and the uncomfortable part, lies in the third cluster: redesigning processes around custom solutions for the key business problems. Not many organizations cross that bridge on their own. That is why we typically engage with the people on the disruptor side of the organization to pursue exactly these cases. 

BCG's September 2026 Applied AI Index points in the same direction: companies that combine strategic clarity with applied AI generate five times more value than their peers, and those that track AI returns directly on the P&L, rather than on technical dashboards, capture three times the value (3.6% versus 1.2%).

A disciplined roadmap therefore maps cross-functional workflows and builds an Opportunity Portfolio in which every candidate is scored on operational friction, revenue impact, and strategic fit. Instead of fifty disconnected ideas, leadership narrows down to the three to five initiatives that can produce a step-change.

2. What will it take to actually work with forward-deployed teams?

A strategy slide promising "autonomous supply chain planning" means little until it has been tested against production reality. The first step is to map the landscape: which data pipelines are production-grade and which exist only in theory, which latency limits apply, what tooling is in place, and whether a proprietary model, a fine-tuned open-source model, or a vendor solution fits the use case.

One question deserves special attention: do you have a semantic context layer? By that we mean the layer that connects AI to your IT landscape (ERP, CRM, service support systems, document systems, contract systems) and encodes what the data points mean and how they relate to each other. Only then can AI work reliably on these data points rather than guess.

In practice, forward-deployed work often starts in a sandbox environment with synthetic data and our own tooling. This shows what is possible without getting dragged into lengthy access discussions about how to connect which tool to which system. Once the solution proves its value, we take the forward-deployed application and integrate it into the real-life systems at a later stage. Validating pipelines, latency limits, and semantic readiness upfront prevents six-figure investments from stalling during enterprise deployment.

3. How do tokenomics and build-versus-buy reshape the investment case?

Executive boards do not release multiyear budgets because of technical curiosity. They allocate capital against quantified business cases. In 2027 two shifts are rewriting the standard case.

First, build-versus-buy is inverting. McKinsey reports that 32% of surveyed organizations, and nearly half of AI high performers, decided against buying at least one software product or feature because internal teams could build it with agentic coding tools. Second, tokenomics has become a hard operating constraint. Even as unit token prices fell, multi-step agentic reasoning drove total consumption up, and 20% of organizations say AI operating costs, including tokens, already limit their usage.

A realistic financial model for every prioritized use case therefore accounts for:

  • Upfront development, testing, and system integration costs,
  • Ongoing inference expenses, API fees, and maintenance overhead at production scale,
  • Direct savings (hours recovered, manual throughput reduced) alongside top-line value (faster cycle times, lower churn),
  • Make-or-buy decision, including who maintains what you build.

‍

A look at what token efficiency can do. Prosus runs AI agents alongside 40,000 employees and 5 million partners to serve 1 billion customers. Token usage grew 52-fold to 8.2 trillion tokens per month, and the company reports having cut its AI costs by 90%, according to its own account from August 2026. There was no single breakthrough. Four disciplines compounded: routing each task to the right model (about 50% of the saving), reducing what each request carries (about 20%, with the average request context cut from 115,000 to 60,000 tokens), distilling specialists for repetitive tasks (about 20%), and better GPU utilization and load balancing (about 10%). Behind them sit three questions that are worth asking constantly: Are we using a more powerful model than this task requires? Are we giving it more information than it needs? Are we buying more computing power than the workload actually uses?

With clear payback horizons and risk-adjusted ROI thresholds, AI stops being an ambiguous cost center and becomes a transparent capital investment.

4. Are you organizationally ready to deliver?

In our experience, the biggest gap to ambition is not a missing tool. It is the fear of the blank sheet. Governance sounds like a complex thing to get right, so organizations talk about it for too long. The better path is to get hands-on and start with a real use case, and let governance grow with it. Readiness then comes down to three areas:

  • Governance & compliance: Are risk classification (including under the EU AI Act) and data protection settled early, so that projects don't get derailed midway through procurement? With agents entering the picture, add a simple question: who is allowed to do what, with which access, and where does a human sign off?
  • Operating model & talent: Does your team have the skills to run monitoring, drift detection, and user adoption after go-live, or do you need specialized support? Someone must own the solution after the project team leaves.
  • Change enablement: How will workflow automation fit into daily operations without cultural resistance? Adoption is rarely a technical problem.

‍

Mapping the exact "gap to ambition" gives leadership a clear list of prerequisites before development funding is committed.

What we see across our sectors

Effective AI strategizing cannot operate apart from sector realities. Two examples from our work show how we solved the AI transformation trap.

Manufacturing & Industry

In manufacturing, the challenge is to improve heavy, real business processes. That is not something you build once, hope it works, and then roll out.

In practice we roll out use cases one unit after the other. You cannot do everything at the same time: you need to prove that the solution works at one site. Once something is in place there, you improve it, build on scalable IT systems, and only then bring it to the next site.

It is hardly ever a copy-and-paste exercise, because factories differ from each other. Machines, sensors, and data standards vary, so a solution that runs well in plant A can behave differently in plant B. IT and OT are often separated, which leaves the data where the data science team cannot easily reach it. And the knowledge of experienced operators and inspectors is rarely captured in the data the model sees.

That is why we work iteratively and on site at the different locations, together with the people who run the processes.

Healthcare & Pharma

With our pharma clients we see three distinct situations, and each calls for a different starting point.

The value becomes visible with a roadmap. Many clients only recognized the value of AI once we had completed the roadmapping project. For them, operationalizing the AI transformation is the first big step: turning the roadmap into solutions that run in daily work.

The fear of being disrupted. A different group of clients is afraid of being disrupted by AI themselves. For them we developed methodologies to capture their potential AI disruption angles and work through them together. The resulting countermeasures feed into a broader AI roadmap, so that the concern becomes a concrete, prioritized plan.

The foundation is not in place yet. Especially mid-sized pharma, but also Mittelstand companies from other sectors, are rather new to AI, and their semantic context layer in the AI IT architecture is not in place. A proper assessment and redesign of that architecture is therefore part of our work with them, before use cases can run on it.

The three situations look different, but the starting point is always the same: understand where the client stands, and build from there.

Closing the gap between strategy and impact

In 2027, the dividing line won't lie between enterprises that talk about AI and those that ignore it. It will lie between organizations stuck in the transformation trap and those that finish the job: redesigning the workflows where AI actually moves the P&L.

Forward-deployed AI strategy means working on exactly those redesigned workflows, and sitting together with the real people on the ground: the business owners and the IT experts. That is how leadership turns experimentation into durable enterprise value.

In a time when every provider calls itself "AI-first", we like the notion of doing our work “experience-first”. You should not trust a partner because they use good AI - that is a commodity. You should trust us because, across more than 250 projects, we have seen what works and what doesn't. That is what our forward-deployed team brings into the room on day one: it knows which use cases stay at an hour or two saved per week, and where the real potential of workflow redesign sits.

Ready to get started? Get in touch via our website or reach out to us on LinkedIn

‍

Subscribe to the Merantix Momentum Newsletter now.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

More articles

The latest industry news, interviews, technologies and resources.

AI-Driven Compound Optimization in the Large Molecule Space

An expert interview on AI-Driven Compound Optimization in the Large Molecule Space.

Sovereignty Belongs to Those Who Can Act

Sovereign AI means knowing your critical dependencies, weighing them, and being able to shape them

First to the Finish: How AI in Market Access Will Determine the Future Market Leaders in the Pharmaceutical Industry

Why the future of market access lies not in more processes, but in intelligent automation.

AI-Driven Compound Optimization in the Large Molecule Space

An expert interview on AI-Driven Compound Optimization in the Large Molecule Space.

Germany wins gold for administrative AI

The award-winning solution you can touch: Momentum presents its groundbreaking AI solution in an interactive working space.

Beyond Pilots: Why AI Strategizing in 2027 Demands a Forward-Deployed Mindset

By Manuel Freude, AI Strategy Project Manager at Merantix Momentum

Walk into almost any enterprise boardroom today, and the tone has shifted sharply since the Generative AI boom. Two years ago, leadership teams applauded as business units spun up dozens of proofs of concept: knowledge-base chatbots, automated sales summarizers, and countless machine learning pilots. Today, executive boards are asking a harder question: "What measurable P&L impact did these initiatives create, and why are we still unable to scale them?"

The data from late 2026 shows how wide the gap between activity and financial return has become. In McKinsey's August 2026 State of AI survey, 80% of respondents report that AI boosts their individual productivity, yet only 37% of organizations attribute a positive EBIT contribution to AI, virtually unchanged from 2025. Only around 6% qualify as true AI high performers with at least 5% EBIT impact.

We see the reason in our client engagements, including with executive teams. When we confront people with the difference between transformation and disruption, the reaction is remarkably consistent. Most people are far more comfortable with transformation: a big program that includes everybody and keeps everyone aligned, confident, and happy. Yet the true potential of AI sits in workflow redesign, and that is usually a disruption exercise. People are asked to saw off the branch they are sitting on. That is an uncomfortable feeling, for the individual and for the team.

This is the "transformation trap": organizations choose the comfortable path, spread AI across many small initiatives, and never reach the redesign work where the P&L actually moves.

What "forward-deployed" means for AI strategy

The forward-deployed model was pioneered in Silicon Valley. Instead of working from a distance, forward-deployed teams closely work with clients to design, configure, and implement workflows and applications together with them. What defines the model is the way of working: iterating rapidly with the future users and building side by side with them, rather than in a black box.

Applied to AI strategy, this changes the order of work. Before the first sprint, our strategist clarifies in the actual workflow which measurable P&L target the initiative serves, how the process really runs today, which data foundation and governance guardrails exist, and which tasks stay with humans. Traditional consulting stops at the slide deck. Traditional IT outsourcing executes whatever specification it receives. Forward-deployed AI strategizing connects commercial direction and technical reality from day one.

To turn AI ambition into measurable value in 2027, your roadmap needs to answer four questions.

1. Where does AI move the P&L, and which workflows must be redesigned?

Too many roadmaps still begin with a tool ("Where can we deploy an assistant?") instead of a structural P&L lever. In many projects we have run structured interviews across business lines, and one thing is different from two years ago: many people now know what AI is and can name use cases that would help them. Those ideas fall into three clusters:

  • Small use cases that a large language model chat, such as Langdock or Copilot, can solve.
  • Lightweight agentic workflows that take over a defined sequence of steps.
  • Redesigned processes built on custom AI solutions, which are larger, more complex, and usually involve agentic software.

‍

The first two are the easiest to find, and they typically save an hour or two per person per week. That is real, but it is rarely what moves margins or revenue. The real work, and the uncomfortable part, lies in the third cluster: redesigning processes around custom solutions for the key business problems. Not many organizations cross that bridge on their own. That is why we typically engage with the people on the disruptor side of the organization to pursue exactly these cases. 

BCG's September 2026 Applied AI Index points in the same direction: companies that combine strategic clarity with applied AI generate five times more value than their peers, and those that track AI returns directly on the P&L, rather than on technical dashboards, capture three times the value (3.6% versus 1.2%).

A disciplined roadmap therefore maps cross-functional workflows and builds an Opportunity Portfolio in which every candidate is scored on operational friction, revenue impact, and strategic fit. Instead of fifty disconnected ideas, leadership narrows down to the three to five initiatives that can produce a step-change.

2. What will it take to actually work with forward-deployed teams?

A strategy slide promising "autonomous supply chain planning" means little until it has been tested against production reality. The first step is to map the landscape: which data pipelines are production-grade and which exist only in theory, which latency limits apply, what tooling is in place, and whether a proprietary model, a fine-tuned open-source model, or a vendor solution fits the use case.

One question deserves special attention: do you have a semantic context layer? By that we mean the layer that connects AI to your IT landscape (ERP, CRM, service support systems, document systems, contract systems) and encodes what the data points mean and how they relate to each other. Only then can AI work reliably on these data points rather than guess.

In practice, forward-deployed work often starts in a sandbox environment with synthetic data and our own tooling. This shows what is possible without getting dragged into lengthy access discussions about how to connect which tool to which system. Once the solution proves its value, we take the forward-deployed application and integrate it into the real-life systems at a later stage. Validating pipelines, latency limits, and semantic readiness upfront prevents six-figure investments from stalling during enterprise deployment.

3. How do tokenomics and build-versus-buy reshape the investment case?

Executive boards do not release multiyear budgets because of technical curiosity. They allocate capital against quantified business cases. In 2027 two shifts are rewriting the standard case.

First, build-versus-buy is inverting. McKinsey reports that 32% of surveyed organizations, and nearly half of AI high performers, decided against buying at least one software product or feature because internal teams could build it with agentic coding tools. Second, tokenomics has become a hard operating constraint. Even as unit token prices fell, multi-step agentic reasoning drove total consumption up, and 20% of organizations say AI operating costs, including tokens, already limit their usage.

A realistic financial model for every prioritized use case therefore accounts for:

  • Upfront development, testing, and system integration costs,
  • Ongoing inference expenses, API fees, and maintenance overhead at production scale,
  • Direct savings (hours recovered, manual throughput reduced) alongside top-line value (faster cycle times, lower churn),
  • Make-or-buy decision, including who maintains what you build.

‍

A look at what token efficiency can do. Prosus runs AI agents alongside 40,000 employees and 5 million partners to serve 1 billion customers. Token usage grew 52-fold to 8.2 trillion tokens per month, and the company reports having cut its AI costs by 90%, according to its own account from August 2026. There was no single breakthrough. Four disciplines compounded: routing each task to the right model (about 50% of the saving), reducing what each request carries (about 20%, with the average request context cut from 115,000 to 60,000 tokens), distilling specialists for repetitive tasks (about 20%), and better GPU utilization and load balancing (about 10%). Behind them sit three questions that are worth asking constantly: Are we using a more powerful model than this task requires? Are we giving it more information than it needs? Are we buying more computing power than the workload actually uses?

With clear payback horizons and risk-adjusted ROI thresholds, AI stops being an ambiguous cost center and becomes a transparent capital investment.

4. Are you organizationally ready to deliver?

In our experience, the biggest gap to ambition is not a missing tool. It is the fear of the blank sheet. Governance sounds like a complex thing to get right, so organizations talk about it for too long. The better path is to get hands-on and start with a real use case, and let governance grow with it. Readiness then comes down to three areas:

  • Governance & compliance: Are risk classification (including under the EU AI Act) and data protection settled early, so that projects don't get derailed midway through procurement? With agents entering the picture, add a simple question: who is allowed to do what, with which access, and where does a human sign off?
  • Operating model & talent: Does your team have the skills to run monitoring, drift detection, and user adoption after go-live, or do you need specialized support? Someone must own the solution after the project team leaves.
  • Change enablement: How will workflow automation fit into daily operations without cultural resistance? Adoption is rarely a technical problem.

‍

Mapping the exact "gap to ambition" gives leadership a clear list of prerequisites before development funding is committed.

What we see across our sectors

Effective AI strategizing cannot operate apart from sector realities. Two examples from our work show how we solved the AI transformation trap.

Manufacturing & Industry

In manufacturing, the challenge is to improve heavy, real business processes. That is not something you build once, hope it works, and then roll out.

In practice we roll out use cases one unit after the other. You cannot do everything at the same time: you need to prove that the solution works at one site. Once something is in place there, you improve it, build on scalable IT systems, and only then bring it to the next site.

It is hardly ever a copy-and-paste exercise, because factories differ from each other. Machines, sensors, and data standards vary, so a solution that runs well in plant A can behave differently in plant B. IT and OT are often separated, which leaves the data where the data science team cannot easily reach it. And the knowledge of experienced operators and inspectors is rarely captured in the data the model sees.

That is why we work iteratively and on site at the different locations, together with the people who run the processes.

Healthcare & Pharma

With our pharma clients we see three distinct situations, and each calls for a different starting point.

The value becomes visible with a roadmap. Many clients only recognized the value of AI once we had completed the roadmapping project. For them, operationalizing the AI transformation is the first big step: turning the roadmap into solutions that run in daily work.

The fear of being disrupted. A different group of clients is afraid of being disrupted by AI themselves. For them we developed methodologies to capture their potential AI disruption angles and work through them together. The resulting countermeasures feed into a broader AI roadmap, so that the concern becomes a concrete, prioritized plan.

The foundation is not in place yet. Especially mid-sized pharma, but also Mittelstand companies from other sectors, are rather new to AI, and their semantic context layer in the AI IT architecture is not in place. A proper assessment and redesign of that architecture is therefore part of our work with them, before use cases can run on it.

The three situations look different, but the starting point is always the same: understand where the client stands, and build from there.

Closing the gap between strategy and impact

In 2027, the dividing line won't lie between enterprises that talk about AI and those that ignore it. It will lie between organizations stuck in the transformation trap and those that finish the job: redesigning the workflows where AI actually moves the P&L.

Forward-deployed AI strategy means working on exactly those redesigned workflows, and sitting together with the real people on the ground: the business owners and the IT experts. That is how leadership turns experimentation into durable enterprise value.

In a time when every provider calls itself "AI-first", we like the notion of doing our work “experience-first”. You should not trust a partner because they use good AI - that is a commodity. You should trust us because, across more than 250 projects, we have seen what works and what doesn't. That is what our forward-deployed team brings into the room on day one: it knows which use cases stay at an hour or two saved per week, and where the real potential of workflow redesign sits.

Ready to get started? Get in touch via our website or reach out to us on LinkedIn

‍

Oops! Something has gone wrong.
Oops! Something has gone wrong.
Oops! Something has gone wrong.
Oops! Something has gone wrong.
Oops! Something has gone wrong.

Discover more whitepapers

Data-driven to the drug of tomorrow

Opportunities and barriers of AI in a GxP world.

Artificial Intelligence for Private Equity Portfolios

Increase in value for your entire portfolio

Towards Tabular Foundation Models

About the status quo, challenges and opportunities

The AI Canvas: Our tool for project evaluation

Discover the AI Canvas!

Data-driven to the drug of tomorrow

Opportunities and barriers of AI in a GxP world.

Artificial Intelligence for Private Equity Portfolios

Increase in value for your entire portfolio

Subscribe to the Merantix Momentum Newsletter now.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.