AI as the engine under commercial strategy.

Most companies talk about AI as if it were a flashing light. Something to bolt onto existing processes. A chatbot in the webshop. A text generator in the marketing department. An AI tool at the helpdesk. A panel appears in the CRM with suggestions. A dashboard turns up with AI statistics on it.

It looks modern. It changes little.

AI as an engine works differently. There, AI doesn’t become an addition to the process. It becomes the driving force underneath it. The process itself changes. The role structure changes. What people do changes. The team gets smaller. The decisions get heavier.

The difference between the two is significant. And it becomes visible in the results after about a year.

This piece is about that difference. What a flashing-light implementation looks like in practice, what an engine implementation looks like, and why most companies unconsciously choose the first while the second is the actual lever. We’re writing this because we have up to ten conversations a month where directors ask us how to use AI in their commercial organisation. The answer is almost never a tool. It’s a different way of working.

The flashing light

A flashing-light implementation looks like this.

A department picks an AI tool. Often a general tool such as ChatGPT Enterprise or Copilot. Sometimes a specialised tool for one discipline: an AI content generator for marketing, an AI call analyser for sales. The tool gets integrated into existing workflow. People get trained. A dashboard appears to measure usage. And there’s a story for the board: we’re doing AI.

What changes is the speed at which existing tasks get done. A marketer writes a blog three times faster. A salesperson gets an automatic summary after a meeting. A customer service agent gets suggested replies to customer questions. All useful. All measurable in productivity.

What doesn’t change: the nature of the tasks. Who does them. Whether they’re still needed. Whether the work that now goes faster is work that should be done at all.

The result after six to twelve months is usually a modest productivity gain, a sense of progress, and a few new budget lines for AI licences. The department managers are happy. The director is happy, because he can tell his investor the company has gone “AI-driven”. The marketing manager has a new slide in her annual report.

For some applications, that’s enough. A transactional process that gets faster, a routine handling that runs more efficiently. There’s nothing wrong with that. The problem starts when the flashing light gets presented as strategic AI transformation while it’s actually just a productivity gain inside an unchanged structure. That isn’t strategy. It’s a cost line with a new name.

The engine

An engine implementation looks different. Here, it starts from a fundamentally different question.

Not: “How can we add AI to what we’re doing now?”

But: “What would our work actually be if AI had existed when we designed it?”

That second question builds again. The first dresses up the existing. And that sounds like a nuance, but in reality it’s the difference between two utterly different end states.

In an engine setup, AI does the heavy production work. Reading data, detecting patterns, generating variants, optimising distribution, maintaining segments dynamically, analysing conversations, summarising transcripts, testing hypotheses, producing content in ten variants per channel. All work AI is now better at than humans, or at least dramatically faster and cheaper.

Humans do the heavy thinking. Making choices, holding lines, setting priorities, valuing quality, judging what matters, building trust, having conversations that don’t fit a script. Work AI isn’t good at yet and won’t be good at in the foreseeable future.

That isn’t a made-up division of labour. It’s a logical split that every team we work with discovers naturally after six months. The question is only whether you adapt the team’s design to it, or whether you try to make a team built for the old balance work in the new balance. The latter never lasts.

We wrote a separate piece on why most AI projects in commercial organisations fail, and why the failure patterns are organisational, not technical.

What a flashing light looks like in practice (three examples)

Let’s work through the difference with three examples from different disciplines. First the flashing-light version, then in the next section the engine version.

Example 1: marketing content. The marketing department wants to produce more content. They buy ChatGPT Enterprise for the whole team. The junior content marketer uses it to draft faster. The content manager edits. A production guideline appears requiring AI for version one of every text. The team produces 40 percent more articles per week.

Example 2: sales enablement. The sales team wants better meeting prep. They buy an AI tool that summarises LinkedIn profiles and company info before a first meeting. The salesperson gets a morning briefing for the afternoon’s appointments. He’s 15 percent faster in prep and walks in better informed.

Example 3: customer service. The customer service department wants consistent responses. They implement an AI tool that gives support agents suggested replies based on the ticket. The agent edits and sends. Average handling time drops by 20 percent.

Three improvements. Three productivity gains. Three things easy to sell to the board. And in all three cases the structure of the work stays the same. The marketing team stays the same size, does the same kind of work, delivers the same kind of output. Just a bit more of it, a bit faster. Same for sales and customer service.

What doesn’t change? The position in the market. The customer segmentation. The proposition. The role split between marketing and sales. The way decisions get made. The story outwards. None of those gets touched by these three implementations.

That’s what we mean by flashing light. It blinks. It works. And it changes nothing that matters strategically.

We wrote a short piece on what the flashing light vs engine distinction means in practice.

What an engine looks like in practice (the same three examples)

Now the same three examples, but as engines.

Example 1: marketing content. The marketing department asks a different question. Not how to make content faster, but what the role of content actually is in their commercial process. They discover that 70 percent of the content they produce is barely viewed, that the remaining 30 percent often doesn’t connect to specific buying moments, and that sales almost never uses content in customer conversations. With that analysis, they redesign the approach. AI gets deployed to continuously read buying signals from data (which questions live in which segments), to produce variants per signal across five channels at once, and to monitor and adjust the performance of each variant. Humans do different things. Three choices a month: which buying signals do we pick up, which channels get priority, and which stories deserve attention over mere production? A team of twelve becomes a team of five. Output triples. But more importantly, the output now connects to real commercial need.

Example 2: sales enablement. The sales team asks a different question. Not how to prep better, but what the sales process actually should be. They discover that 80 percent of sales time goes to administration, follow-up, sending proposals, internal alignment. Only 20 percent goes to actual customer contact. With that analysis, they redesign. AI does all the administration and follow-up: notes after calls, follow-up emails, proposal drafts based on the calls, CRM updates, lead scoring based on behavioural data. The salesperson does different things. He has more, and qualitatively deeper, customer conversations. He reads patterns from what AI summarises and decides which deals get strategic attention. A team of twelve becomes a team of six. The number of commercial conversations per rep doubles. Average deal value rises because attention goes to the right deals.

Example 3: customer success. Customer service asks something different. Not how to answer faster, but what their role actually is in retention and growth. They discover that they spend 90 percent of their time on questions AI could answer directly, and they barely get to the conversations with customers about to leave or who are ripe for expansion. With that analysis, they redesign. AI handles all standard queries directly. The system reads churn signals weeks in advance. It identifies customers ready for expansion based on usage data. The customer success rep does different things. He has conversations only a human can have, with customers at a turning point. The team moves from customer service (reactive) to customer success (proactive). Twelve people become four. But net retention rises by 15 percentage points.

Three examples of fundamentally different work. Three times a smaller team, qualitatively heavier roles, and results the old balance couldn’t have reached. Not because AI does the work, but because the work itself is fundamentally redesigned.

What this means for team structure

A commercial team in engine mode is smaller and qualitatively heavier. The roles shift in ways not everyone enjoys.

Whoever was an executor becomes a director. That isn’t a promotion, that’s a different occupation. A content marketer who used to write texts and now writes briefs for AI, then directs the output, does work closer to that of an editor-in-chief than a writer. Not every content marketer wants that. Not every content marketer can do that.

Whoever used to analyse becomes an interpreter. A data analyst who used to build reports now receives output from AI systems that continuously analyse. Her work becomes translating that output into decisions, especially recognising what matters versus what’s noise. That requires different expertise than building dashboards.

Whoever made content becomes the editor of what AI makes. That’s a role most organisations don’t have yet. It’s a hybrid between creative director and quality controller, with core competence in distinguishing good from mediocre work across ten variants an hour, rather than two a day.

Whoever did lead qualification or administration often becomes redundant. That’s the difficult reality of an engine implementation. It isn’t cost saving as a director might imagine it (fewer people doing the same work). It’s a fundamental redistribution of roles in which some disappear because AI has made them redundant, and others get heavier because they’re dealing with more material and more decisions.

That’s exactly why most companies stay with the flashing light. A flashing light doesn’t require team redesign. An engine does. And team redesign is a political, organisational, and human challenge most boards prefer to avoid.

What AI makes visible about what real marketing work is

AI is a lever. A lever only works if someone knows where to push. And right there, something gets painfully visible in commercial organisations.

What AI exposes is how much of what we called marketing wasn’t actually marketing. It was production work we called marketing because marketing was about it. Making material, testing variants, arranging distribution, maintaining segments, producing content, qualifying leads. Important work, but execution. Not strategy. Not position choice. Not direction decisions.

The real marketing work, the strategic choices about position, audience, price, and story, was always heavier than the production work. But the production work consumed the time, and so less real marketing got done than should have. The marketing manager could always say: I don’t have time for it. And she was right.

AI takes the production work away, or at least brings it down to a fraction of the time. What’s left is the real marketing work. And that’s exactly the work that’s hardest. That demands the most judgement. That requires decisions no tool will ever answer.

The same goes for sales. AI exposes that 80 percent of sales time in most organisations went to work that wasn’t sales: administration, follow-up, internal alignment. The real sales work, the conversations that shift deals, was always heavier than the administration. But the administration consumed the time.

The same goes for customer success. For product development with AI in user research. For pricing. For nearly every commercial function.

What’s left is always some form of strategic judgement: making choices, setting priorities, building trust, having conversations, holding lines. All work AI doesn’t do, and won’t do in the foreseeable future. But that work becomes relatively heavier now, because there’s less other work to hide behind.

The question for a director, then, isn’t whether to use AI. Almost everyone uses something by now. The question is what you do with the time that frees up. That’s where flashing light and engine permanently part ways. And that’s why this conversation belongs at the board table, not in a department.

The director’s role in an engine implementation

You can’t delegate an engine implementation to a department. Not to IT, not to marketing, not to operations.

The work redesign affects organisational structure. It affects roles, budgets, departmental lines. It affects expectations people had about their careers. Sometimes it affects contracts and employment relationships. That isn’t department-manager work. That’s director work, or in a larger company, board work.

Three questions only answerable at the board table, and necessary in an engine implementation.

Which work becomes fundamentally different, and which work disappears? That’s a choice, not a discovery. The director chooses whether AI mainly becomes a speed improvement or whether the work itself gets redesigned. Both are possible, but they’re different paths with different organisational consequences.

Which new roles do we create, and which do we let go? This isn’t an HR question. It’s a commercial question about what’s needed to function in the new situation. A director who delegates this to HR gets back a job architecture that doesn’t fit where the market is going.

What’s the quality bar? AI can produce a lot in a short time. What we will and won’t accept is a board decision with major consequences. A marketing department that can publish ten blogs a week with AI can also publish ten mediocre blogs a week. The bar is set by what the board considers acceptable for the brand. Not by what production allows.

These three questions don’t get asked in most organisations, because they don’t fit the operational AI conversation (“which tool do we buy”). They belong in the strategic AI conversation (“how do we redesign the work”). And that conversation rarely happens.

What this means for the timeline

A flashing-light implementation is executed in three to six months. Buy tools, integrate, train, measure. Result: a productivity gain in existing work.

An engine implementation takes longer, because the work itself needs redesigning. But it doesn’t have to happen across all disciplines at once. What we see working is an approach that builds one engine per discipline, starting where the lever is largest.

Six to eight weeks to design and test the first engine. Then six months to run and learn from it. Then the next discipline. Anyone who continues this approach for a year has fundamentally redesigned three commercial disciplines, instead of five departments with a flashing light that has changed nothing.

It’s slower in year one. It’s much faster in year two. And, above all: it’s a fundamentally different result, not an incremental improvement.

Competitors who choose the engine get three to five years’ lead on competitors who stay with the flashing light. Not in productivity, because that becomes roughly equal for everyone. In strategic sharpness and commercial effectiveness. What an engine company produces moves in a clear direction with a redesigned role split. What a flashing-light company produces moves in every direction at once, with the same role split as four years ago.

The diagnosis

The question for a director isn’t whether to use AI. Almost everyone uses something. The question is whether AI, in their company, is a flashing light or an engine.

The difference sits in two things. What changes for the work itself? And what gets done with more weight and sharpness than ever before?

A flashing light changes little and costs little. The role split stays the same. The work stays the same. The output just goes a bit faster. For some applications, that’s enough, and that isn’t a criticism. Not everything has to be strategic.

An engine fundamentally changes the work. The team gets smaller. The roles get heavier. The director has to make choices he’d rather not make, about who leaves and what changes. And the lever he gets is disproportionately large.

Most companies unconsciously choose the flashing light because the engine creates discomfort. The redesign demands something you’d rather defer. But the market won’t wait. Competitors who do choose the engine build a commercial organisation that does more with fewer people, with more focus, with better choices. That lead grows every quarter.

It isn’t a pleasant question. It’s the right one. And it’s a question only answerable at the board table. Not by a department. Not by a tool. Not by an implementation plan.

---

Further reading

Marketing at the board table. Why strategic marketing has disappeared from the boardroom agenda, and what belongs there.

Why are we losing this. The three strategic layers underneath every lost deal.

Marketing in the wrong meeting room. Why strategic marketing ends up between the gaps.

Three excuses for a lost deal. Why the standard excuses block the learning.

AI as the engine, not the flashing light. The short version of what changes when AI redesigns the work.

Why do AI projects fail?. Three failure patterns in commercial teams.

Mental availability. Why your market share isn’t determined by your product.

Five marketing questions for the board. Five questions a director should ask, and almost never does.

Reading a marketing report as a director. Six signals that reveal what’s missing, and which questions to ask.

Win/loss analysis: how to make it work. Three common mistakes and how to set it up differently.

The commercial team under AI. Which roles disappear, which get heavier, and what the director has to do.