Here’s a pattern that plays out across media companies of every size. The CMS was chosen in 2016. The ad server was bolted on a year later. Analytics came from a different vendor. The paywall was somebody’s Q3 project. The recommendation engine was a startup integration that the startup no longer maintains. And the consent management layer was added in a panic right before a GDPR deadline.

Nothing was designed to work together. Everything was designed to solve last quarter’s problem.

The result is a tech stack where editorial wants speed, product wants flexibility, revenue wants performance and nobody’s getting what they need because every system has its own data, its own logic and its own version of the truth. The 2026 Publishers’ Tech Stack Study put it clearly: the modern publisher stack isn’t under-equipped. It’s under-orchestrated.

If your tech stack were a band, every musician would be playing a different song. In a different key. At a different tempo. And somebody in the back would still be tuning.

How Publisher Tech Stacks Got Here

Nobody builds a fragmented tech stack on purpose. It happens one reasonable decision at a time.

A CMS gets selected for its editorial workflow. An ad tech layer gets added for programmatic revenue. An analytics platform gets chosen because the marketing team likes its dashboards. A subscription system arrives when the business model shifts. Each tool solves a real problem. Each tool creates a new silo.

Over a decade, these layers accumulate. A Digiday survey found that 79% of ad operations professionals believe their current tools are inadequate. 55% cite system fragmentation as a major barrier to scaling operations. The tools aren’t bad individually. They just can’t hear each other. Like the Fellowship of the Ring before Rivendell — everyone’s talented, nobody’s coordinated, and the hobbits are doing their own thing entirely.

The pattern accelerates under pressure. A new revenue initiative needs a new tool. A platform migration gets started but never finished. A vendor gets acquired and the integration breaks. Legacy code accumulates because nobody has time to refactor it. And eventually the tech stack reaches a state where changing anything feels dangerous, so nothing changes. That’s technical debt compounding silently — and in publishing, it compounds faster than most industries because the content cycle never stops.

What a Fragmented Stack Actually Costs

The cost of a fragmented publisher tech stack isn’t always visible on a balance sheet. It shows up in slower places.

Revenue leakage. When ad servers and analytics platforms don’t share data cleanly, optimisation becomes guesswork. Header bidding latency kills impressions. Viewability drops because the page loads like a Tolkien chapter — beautifully written, painfully long. Multiple systems competing for the same ad slot means nobody wins efficiently. Publishers leave money on the table not because their content is weak, but because their delivery infrastructure can’t keep up.

Editorial slowdown. Storyblok’s research shows 49% of WordPress users report it takes over an hour to publish content. For high-volume newsrooms, that’s a serious bottleneck. When the CMS requires developer intervention to publish a story, add a module or update a layout, editorial velocity drops. And in media, editorial velocity is revenue velocity.

Operational fatigue. Teams spend hours reconciling data across platforms, debugging broken embeds, running manual exports and maintaining integrations that nobody fully understands anymore. That time isn’t free. It’s engineering hours that could be building features, optimising performance or reducing technical debt.

Missed audience engagement. Disconnected systems mean disconnected experiences. If personalisation, content recommendations and subscription prompts all run on different data, they pull in different directions. The reader notices. Not consciously, but in the way that makes them leave after one page instead of reading five.

What a Modern Media Architecture Actually Looks Like

A modern publisher tech stack is about having connected tools.

A unified content layer

The CMS should be the single source of truth for content — not one of several places where content lives. Modern CMS architectures (headless or hybrid) decouple content from presentation, which means the same content can serve the website, the app, Apple News, AMP and any future channel without duplication. The content layer talks to everything. Everything talks back.

An integrated ad and revenue stack

Ad serving, header bidding, programmatic and direct sales need to operate on shared data with shared goals. When the ad stack can communicate with the CMS and the analytics layer in real time, publishers can optimise ad placement, density and formats based on actual user behaviour rather than yesterday’s report.

Connected data infrastructure

Audience data, content performance data, revenue data and operational data should flow through a shared pipeline. Not identical dashboards — different teams need different views. But the underlying data should be consistent, accessible and trustworthy. When three platforms show three different numbers for the same metric, nobody trusts any of them.

APIs as connective tissue

The systems don’t all need to come from the same vendor. They need to communicate through well-defined APIs and integration layers. This is where the architecture earns its value — in the connections between components, not in the components themselves. A modern publisher stack is more like a well-connected network than a monolithic platform. Each piece does one thing well. The APIs make sure they do it together.

How We Approached This at Hearst

This isn’t hypothetical for us. Lerpal’s engineers work embedded within Hearst’s internal teams across the Digital Magazines division, a global operation reaching 400 million unique visitors through 25 U.S. brands and close to 200 websites worldwide, including Cosmopolitan, Elle, Car and Driver and Men’s Health.

At that scale, every update ripples out to an enormous audience. The work spans exactly the kind of complexity described above: an existing CMS with real limitations, a new one being built to replace it, and an extensive internal ecosystem where every new service has to fit in without breaking what’s already there.

Some of the highest-impact work:

A new CMS built around editorial needs. Working alongside Hearst’s content teams, we helped build a new CMS shaped around how their editors actually work, while managing the constraints of the legacy system in the meantime. This is the unified content layer in practice, and it only succeeds when the people using it every day shape it.

Integrating new services into a complex ecosystem. New services and technologies were brought into Hearst’s tech ecosystem with close attention to how each one connects to everything else. The result was greater operational efficiency rather than another silo. This is where APIs as connective tissue stop being a diagram and become daily engineering.

Scalable, high-performance architecture. We’ve continued to develop and sustain an architecture built to handle the demands of a global audience, balancing scale with performance so growth doesn’t come at the expense of speed.

A better search and browsing experience. We revamped on-site search, improved navigation and optimised page load times, all aimed at keeping readers engaged beyond their first page.

Continuous A/B testing. Ongoing experimentation gave Hearst clear insight into how readers behave, so teams could tailor content and adjust interfaces based on evidence. User session times increased notably, and ad revenue grew across Hearst websites as a result.

None of this would have been possible on a fragmented stack. Every piece required systems that could talk to each other, share data and operate as a coordinated architecture. The engineering work wasn’t glamorous. It was integration, architecture, testing and a lot of making sure things that should talk to each other actually do. That’s the Billie Jean of publisher engineering: technically precise, deceptively smooth, working harder underneath than it looks.

Where to Start If Your Stack Needs Work

Not every publisher needs a full platform migration. Most need targeted improvements in the right places.

Audit what you have. Map every system, every data flow, every integration point. Find out what’s connected, what’s broken and what’s held together by a script someone wrote three years ago and left undocumented. This audit almost always surfaces surprises — orphaned systems, redundant tools, integrations that nobody monitors.

Identify the most expensive bottleneck. Where does the stack slow down editorial? Where does it lose revenue? Where do teams waste the most time on manual work? Start there. The highest-impact improvement is usually obvious once you map it.

Fix data flow before adding new tools. Most publisher tech stack problems are data problems in disguise. If your content data, audience data and revenue data don’t flow cleanly between systems, adding another tool makes the problem worse. Fix the pipes before you add more water.

Layer improvements gradually. You don’t need to replace everything at once. Wrap legacy systems with APIs where possible. Modernise the most critical integration points first. Build the connected data layer that every future improvement will depend on. Each improvement compounds — and a year of steady progress beats a two-year migration that never finishes.

Invest in the integration layer. The most undervalued work in publisher tech is the work that connects systems. API design, data pipelines, event architecture, monitoring. This is the infrastructure that determines whether everything else works together or works in parallel. It’s also the work that’s hardest to justify in a budget meeting because it doesn’t have a UI. But it’s the foundation that everything visible depends on.

The Stack Isn’t the Strategy (But It Enables One)

A modern publisher tech stack doesn’t guarantee great journalism, brilliant ad performance or a loyal audience. Those things come from the people, the editorial vision and the business model.

But a connected, well-orchestrated tech stack removes the friction that prevents good work from reaching its potential. Pages load faster. Content publishes sooner. Revenue optimisation happens in real time. Teams spend their time on strategy instead of firefighting. And when the next opportunity arrives — a new channel, a new revenue model, an AI integration — the stack is ready for it instead of blocking it.

At Lerpal, we’ve spent twenty years building and connecting publisher technology. We know what media tech stacks look like under the hood — because we’ve been inside them, fixing them, modernising them and making them work at scale. If your stack feels like it’s held together with hope, that’s fixable. And it’s fixable without burning everything down and starting over.

Let’s talk about making your tech stack work as well as your content does.

Somewhere between “we should probably use AI” and “our entire team runs on it” there’s a gap many businesses are still crossing.

The tools are here.
But which ones are worth your time and which ones will show results six months from now?

Yes, artificial intelligence has moved past the novelty phase. Generative AI tools now handle content creation, data analysis, customer communication and business processes that used to require entire teams.

And here are some numbers.

Deloitte’s 2026 State of AI report found that worker access to AI rose by 50% in 2025, and two-thirds of organisations report productivity gains from AI adoption. And the Federal Reserve’s 2025 Small Business Credit Survey found that 71% of businesses using AI saw increased productivity.

So, the best AI tools for business in 2026 are the ones that embed into your existing business operations and make them faster without creating new problems.

How AI Tools Improve Business Productivity

AI-powered tools have moved far from the “interesting toy” phase. They automate repetitive tasks that used to eat hours:

• scheduling, data entry, first-draft writing, meeting notes;
• streamlining workflow management by connecting systems that previously required a human to copy-paste between them;
• helping teams with research, documentation and analysis at a speed that would have required hiring two extra people three years ago.

The bigger shift in 2026 is AI orchestration: combining multiple AI tools into connected stacks where the output of one feeds the input of another. Think of a meeting that gets transcribed by one tool, summarised by another, turned into action items by a third and pushed into your project management workspace automatically. 

That kind of pipeline used to require custom software development. Now it takes a Zapier account and about twenty minutes.

What to Look for in AI Tools for Business

Before you commit to any AI tool, evaluate four things:

1/ Ease of integration

The best AI tools connect to what you’re already using: your docs, your CRM, your productivity apps. A tool that lives in isolation creates another silo and you probably have enough of those.

2/ Automation capabilities

Can it automate workflows and handle repetitive tasks without you babysitting it? If the tool saves you thirty minutes but requires twenty minutes of prompt engineering every time, the math doesn’t work.

3/ AI model performance

Many tools now run on advanced AI models like GPT-5, Claude or Gemini. The model underneath affects how well the tool handles nuance, follows instructions and deals with complex inputs. This matters more than the marketing page suggests.

4/ Scalability

A tool that works for three people needs to work for thirty. And then three hundred. Check whether pricing, features and performance hold up as your team grows.

10 Best AI Tools for Business Productivity

1/ ChatGPT

The AI Assistant That Set the Standard

ChatGPT has more than 900 million weekly active users, and for good reason. It handles everything from writing emails and generating reports to brainstorming ideas and automating customer support. The Business plan ($25/seat/month) adds workspace collaboration, SSO and data privacy controls. Enterprise goes further with compliance certifications, analytics dashboards and custom connectors via MCP.

For most businesses, ChatGPT is the AI equivalent of a Swiss Army knife. Not always the best tool for any single task but consistently useful across almost all of them. The GPT-5 models rolling out through early 2026 have made the reasoning and output quality noticeably sharper.

2/ Perplexity

AI Search and Research Assistant

If your team spends hours researching competitors, markets, regulations or technical topics, Perplexity changes the game. It combines AI search with real-time citations, so you get answers with sources attached. Yes, finally no more opening seventeen tabs and cross-referencing manually.

For research-heavy use cases in consulting, finance, legal or content creation, Perplexity saves time that ChatGPT alone can’t.

ChatGPT generates. Perplexity finds and verifies.

Different tools, different jobs.

The free tier is genuinely useful, and the Pro plan unlocks deeper analysis and longer research sessions.

3/ Canva AI

AI-Powered Design and Content Creation

Canva became one of the most used AI tools in business by making design accessible to people who can’t tell kerning from leading. The AI features now generate AI images, resize designs across formats, write copy and suggest layouts. Yes, all inside a tool that marketing teams were already using daily 5 years ago.

For teams producing social media content, presentations, marketing materials and internal docs, Canva AI removes the bottleneck of “waiting for design”. The free plan includes basic AI features. The paid tiers unlock the full AI creative suite, brand kits and team collaboration.

4/ Notion AI

The AI Productivity Workspace

Notion AI turns your workspace into something that thinks alongside you:

• summarises docs;
• drafts content;
• extracts action items from meeting notes;
• answers questions about your own documentation.

For teams that already live in Notion for project management and knowledge management, the AI layer makes the existing workflow faster without adding another tool.

The magic is that it operates on your data:

• your meeting notes;
• your project briefs;
• your team wiki.

Tools like Notion AI work best when they understand context and context comes from the information your team already has inside the platform.

5/ Zapier

Workflow Automation Engine

Zapier connects apps. That’s the simple version. The more accurate version: Zapier builds AI-driven workflows that automate the repetitive tasks your team does between tools:

• the copy-paste;
• the “when X happens in Slack do Y in our CRM”;
• the data routing that nobody enjoys but everyone needs.

In 2026, Zapier’s AI features have expanded to include natural language automation setup and AI-powered workflow suggestions. You describe what you want to happen and Zapier builds the automation. This is the closest thing to hiring an assistant who never sleeps, never complains and connects to thousands of apps across the ecosystem.

Are these tools actually saving you time, or just adding more to manage?

AI works best when it’s connected into how you already run your business. We help make that happen.

Let’s talk

6/ Jasper

AI Marketing Content Creation

Jasper is built specifically for AI marketing:

• blog writing;
• ad copy;
• email campaigns;
• landing pages.

It’s trained on marketing best practices and optimised for the kind of AI content output that marketing workflows demand. Where ChatGPT is a generalist, Jasper is the specialist who knows your brand voice, your target audience and what performs. It even offers template libraries for common content types so you’re not starting from scratch every time.

If your content marketing operation produces volume like multiple blog posts per week, ad variations, social campaigns, then Jasper can handle the first draft faster than any human writer. The human still edits (please still edit), but the time from blank page to workable draft is way faster.

7/ Grammarly AI

Writing Assistant

Grammarly started as a spell-checker and evolved into something significantly more useful. The AI features now:

• rewrite for tone;
• suggest structural improvements;
• adjust formality level.

And our favorite: it catches the kind of awkward phrasing that makes professional communication feel off.

It works inside email, docs, Slack and browsers.

For teams where written communication is the default mode of work (which is most teams in 2026), Grammarly AI improves productivity by reducing the back-and-forth of “can you rephrase this?”. It’s one of those tools that’s hard to notice when it’s working and impossible to ignore when it’s not.

8/ Fireflies AI

Meeting Assistant

Open source tends to be the stronger choice when your company has engineering capability and wants control. For example:

Meetings happen. But as we all know – notes from those meetings frequently don’t. Fireflies solves this by recording, transcribing and summarising meetings automatically, then extracting action items and making everything searchable. It integrates with Zoom, Google Meet and Microsoft Teams.

The productivity gain is straightforward:

• nobody has to take notes during the meeting;
• nobody has to email the summary afterwards;
• nobody can argue about what was really decided.

Fireflies just records what happened and makes it available. For teams that run fifteen meetings a week (and we all know that’s a conservative estimate), this tool gives back hours.

9/ Claude AI

Advanced Reasoning Assistant

Claude is the AI assistant you reach for when the task requires reading a lot, reasoning carefully and not hallucinating a plausible-sounding answer. It handles long document analysis, complex writing tasks, enterprise workflows and the kind of nuanced reasoning that requires holding a lot of context at once.

Businesses use Claude for:

• contract analysis;
• research synthesis;
• technical documentation;
• strategic planning.

Where Claude tends to stand out is in tasks that require reading and reasoning over large volumes of text:

• upload a 200-page vendor contract and ask for clause-level risk flags;
• drop in a quarter’s worth of customer interviews and ask it to pull out the three themes that actually matter;
• give it a 50-tab spreadsheet export and ask for a structured summary.

These are the examples where Claude’s long-context handling earns its keep. The enterprise API and team plans offer the integration and security controls that larger organizations need.

10/ Browse AI

Automation for Data Collection

Browse AI automates web data collection without requiring anyone to write code. Point it at a website, tell it what data you want and it extracts, monitors and delivers it on a schedule. Competitor pricing, market data, job listings, product catalogues – anything that’s publicly available on the web.

For operations and research teams who currently do this work manually (or pay someone to do it manually), Browse AI turns hours of weekly data gathering into an automated pipeline. The use cases are specific but the time savings are significant for teams that depend on fresh external data.

Free AI Tools for Businesses

Not every AI tool requires a budget approval. Many of the best AI tools for business offer free tiers that let teams experiment before committing:

• ChatGPT free plan gives access to GPT-5 with usage limits. Just enough to test whether AI fits your workflow before scaling up.
• Canva AI includes basic AI features on the free plan, enough for simple design work and social content.
• Perplexity free tier handles standard AI search queries well, with the Pro plan unlocking deeper research.
• Grammarly offers a solid free version for basic writing assistance, with AI features on the premium plan.
• Notion AI integrates into free Notion workspaces, though the AI-specific features require a paid add-on.

Starting with free AI tools reduces risk. Test the workflow, measure the time saved, then decide whether the paid version justifies the investment.

How Businesses Should Use AI Tools Strategically

Collecting tools is a trap. The strategy is building stacks. Instead of adopting AI tools individually you can combine them into connected workflows where the tools work together:

• Marketing stack: Canva for visuals, Jasper for copy, ChatGPT for brainstorming and editing – content creation from idea to published asset.
• Operations stack: Notion AI for documentation and project management, Zapier for automation between systems, Fireflies for meeting intelligence – the operational backbone.
• Research stack: Perplexity for sourcing and verification, ChatGPT for synthesis and drafting, Browse AI for competitive data – research that used to take days compressed into hours.

The connective tissue matters. When your AI tools share data and trigger each other’s workflows, the productivity gains compound. When they don’t, you have ten subscriptions and the same bottlenecks. AI tools can help most when they talk to each other.

The Future of AI Productivity Tools

AI tools are moving toward autonomy. The latest AI development everyone’s watching is agentic AI: AI agents that perform multi-step tasks independently based on a goal rather than a single prompt. Deloitte found that 44% of companies were already deploying or assessing AI agents by late 2025, and NVIDIA’s 2026 report showed telecommunications leading agentic adoption at 48%.

We’re heading toward a world where your AI assistant doesn’t wait for instructions. It watches your calendar, reads your inbox, prepares your briefs and flags the problems before you ask. That world isn’t fully here yet, but the tools on this list are building toward it and the businesses investing in AI productivity stacks now will have a significant head start when agents become the standard.

How To Build AI Into Your Business?

AI productivity tools aren’t going away, and they’re not getting simpler either. The benefit comes when you choose tools deliberately, connect them into real workflows and treat AI adoption as an ongoing capability rather than a one-time purchase that will do all the work.

If you treat AI tools as isolated experiments, then you’ll get isolated results.

Integrate AI tools into your core business operations, connect the tools to your business processes and you’ll definitely report real productivity gains.

At Lerpal, we help organizations move from “we use some AI tools” to “AI is embedded in how we operate”. Whether you need help designing AI-driven workflows, building custom AI solutions, integrating AI tools into existing systems or developing an AI strategy that scales, we’ve been building this kind of infrastructure for twenty years. And AI is the latest (and arguably most interesting) layer.

Let’s talk about building AI into your business the right way.

AI adoption is accelerating. Good AI and AI strategies? Much less so.

McKinsey’s 2025 Global Survey found that 88% of organisations now use artificial intelligence in at least one business function. At the same time, only about 7% have fully scaled it. MIT’s research went further: roughly 95% of generative AI pilots deliver zero measurable return on P&L. Thirty to forty billion dollars of enterprise investment and the vast majority of it stuck in pilot purgatory.

So the technology works and adoption is real, but somewhere between “let’s try AI” and “AI is generating business value” something keeps breaking. Repeatedly. At scale. With impressive budgets attached.

That something is AI strategy.

Enterprise AI strategies determine whether AI becomes a capability your organisation can build on or an expensive experiment that gets shelved next quarter when someone asks “wait, what did we get from this?”.

Good AI, the kind that aligns with business goals, scales and creates measurable value that is an outcome of disciplined execution. And discipline, as it turns out, is harder to buy than a language model subscription. ​

Here are 8 enterprise AI strategies that separate the 5% from everyone else.

1. Business-First AI Strategy

Before anyone picks an AI model or evaluates AI tools there’s a question that deserves an answer: what core business problems are we solving?

Good AI strategies starts with clarity on priorities, growth objectives, margin pressures, risk exposure and customer outcomes. If an AI initiative can’t trace a direct line back to your business objectives and corporate strategy, it probably shouldn’t be funded.

That sounds ruthless, but consider the alternative: MIT’s research found that over half of enterprise AI budgets go to sales and marketing tools (the most visible category) while back-office automation consistently delivers higher ROI. 

Investment follows hype when business strategy isn’t leading the conversation. And hype, for the record, has never once appeared on a balance sheet as an asset. Align AI strategies with what the business needs and only then choose the technology. The sequence feels obvious but the number of organisations that do it backwards suggests otherwise.

2. Use-Case Prioritisation Strategy

There are always more AI use cases than time, budget or patience to implement them. Always. The 2025 AI Governance Benchmark Report found that 80% of enterprises have fifty or more generative AI use cases in their pipeline. Most have only a handful in production.

The gap between “ideas” and “shipped” is where enterprise AI goes to age.

Leading organisations sequence their AI initiatives deliberately. They identify high-impact domains, pick the ones with the best ratio of effort to business value, prove those work and build momentum from there. Scattered experimentation is how you end up with twelve AI projects, three conflicting analytics dashboards and a proof-of-concept that nobody remembers commissioning. Every one of those projects has a Slack channel, a project lead and a monthly update meeting. None of them have shipped.

Good AI is curated. Pick a boring expensive problem, automate it, measure what changed and then expand.

3. AI Operating Model Strategy

Structure determines whether AI scales or stalls, and this is where many enterprises hit a wall that has nothing to do with technology. Good AI strategies involve getting the operating model right before scaling anything.

An AI operating model needs to define decision rights, accountability, funding authority and cross-functional coordination:

• Who approves new AI projects?
• Who owns outcomes after deployment?
• Who decides when to kill something that isn’t working?

Without clear answers, AI initiatives drift into the organisational equivalent of a shared Google Doc where everyone has edit access and nobody owns the final version. And the longer ownership stays unclear, the worse it gets. Projects duplicate because nobody knew someone else was already doing the same thing. The waste isn’t dramatic but administrative, and that’s why it survives so long without anyone noticing.

The typical options are centralised (one team controls everything), federated (business units run their own AI) or hybrid. Each has tradeoffs.

• Centralised models offer consistency but can bottleneck every decision through a single team that becomes the busiest people in the building;
• Federated models move faster but risk duplication: three teams building three chatbots for three slightly different versions of the same problem;
• Hybrid approach tries to balance both and (like most compromises) works best when the governance framework around it is solid.

Pick one, define the roles, revisit as AI maturity grows. The worst choice here is no choice because ambiguity doesn’t pause while you figure it out, it just creates more pilots.

4. Governance and Responsible AI Strategy

Governance gets a bad reputation. People hear the word and picture slow approvals, thick policy documents and “compliance” repeated until it loses all meaning, but in the AI context governance is what keeps things from going wrong in expensive and public ways.

The PEX Report 2025/26 found that only 43% of organisations have a formal AI governance policy in place, while 29% have none at all. Meanwhile, Deloitte’s 2026 State of AI report found that only one in five companies has a mature governance model for autonomous AI agents, even as agentic AI adoption accelerates quickly. That’s like building faster cars while removing the brakes. Technically impressive, strategically questionable.

Responsible AI means embedding data controls, ethical AI, compliance oversight and model monitoring into how AI systems operate from day one.

• Address ethical risks before they become headlines.
• Establish data protection standards before a regulator asks about them.
• Monitor AI outputs continuously, because a model that worked well in January might hallucinate confidently by March.

Governance done well is the reason you can deploy without worrying that something is going sideways. Governance done badly (or not done at all) is the reason your competitor is in the news for the wrong reasons.

5. Data and Technology Foundation Strategy

AI cannot outperform its foundation. You can deploy the most sophisticated machine learning models available, and if they’re trained on messy, incomplete or contradictory data, the AI outputs will reflect exactly that. Garbage in, confident garbage out!

Data quality, interoperability, infrastructure scalability and platform standardisation is the groundwork that separates robust AI from expensive disappointment. And you can’t redesign business processes on top of fragmented, siloed data. If data lives in four systems and disagrees with itself in three of them, your AI will confidently act on whichever version it sees first or likes more. It won’t ask for clarification, it will just be wrong, fluently.

This also means managing data governance alongside AI governance.

• Know where your sensitive data lives;
• Know who can access it;
• Know which data your AI systems are training on.

When you skip this step, you tend to discover the problem later, usually in front of a regulator or on the front page of something you’d rather not be on.

Data preparation typically consumes 40-60% of AI project budgets and up to 70% of data science teams’ time. That ratio feels backwards until you’ve watched a perfectly capable AI system produce nonsense because nobody cleaned the inputs.

6. Talent and Capability Development Strategy

Insufficient worker skills is one of the biggest barriers to integrating AI into existing workflows. The technology is ready but the people strategy, in many cases, is only catching up.

AI maturity is organisational maturity. It means:

• AI literacy across leadership (not just the data science team) so that executives can evaluate AI initiatives without defaulting to “sounds impressive, approved”;
• Technical depth in core teams so engineers and analysts can work alongside AI systems rather than route around them;
• Continuous up skilling, because AI tools evolve so fast that last year’s training is already a historical document.

McKinsey’s data shows that high-performing organisations are three times more likely to report strong senior leadership engagement with AI. That engagement goes beyond signing off on budgets. It means leaders who role-model use of AI set the vision and make governance decisions rather than delegating them into a committee that meets every other Thursday.

Successful AI strategies were driven by empowered line managers, not centralised AI labs. The people closest to the problems usually know which problems are worth solving. Give them the skills and the permission to use AI and adoption stops being a top-down mandate and starts being something that sticks.

Build AI you trust

Talk to our experts to shape a practical, governed AI strategy that scales safely and delivers real outcomes.

Let’s talk

7. Portfolio and Investment Strategy

AI investments should be managed the way any serious portfolio is: with balance, discipline and regular rebalancing.

That means:

• Allocating across efficiency gains and growth initiatives;
• Short-term ROI and long-term positioning;
• Incremental improvements and entirely new AI capabilities;
• Courage to defund things that aren’t working which is harder than it sounds when someone’s reputation is attached to a pilot that’s been running for nine months with no clear key performance indicators.

8. Dynamic Iteration and Change Strategy

An enterprise AI strategy written in 2025 and left untouched by 2027 will be irrelevant. Markets evolve, technology follows. Your strategy has to keep pace or it becomes a very well-formatted artifact from a different era.

• Establish review cycles;
• Build performance tracking into every initiative from the start;
• Create feedback loops where the people using AI tools can report what works, what doesn’t and what’s changed since deployment;
• Build capital reallocation mechanisms so budgets follow performance rather than staying locked to last quarter’s plan.

If you still review their AI strategy annually, you are reviewing a document that was outdated by the time the meeting started.

The best enterprise AI strategies treat iteration as a feature. Getting it right the first time is a fantasy, but getting better continuously is how AI success happens.

A Practical Framework for Creating Good AI

If the eight strategies above feel like a lot (and they are), here’s the condensed version. Five steps to create an effective AI strategy:

 

This won’t win awards for novelty. But if you follow it step by step without skipping the boring parts, you will end up on the right side of that 95/5 divide.

What Comes After Strategy

Enterprise AI strategies fail because there are too many pilots without ownership, too much investment without alignment, too much technology without architecture, too much enthusiasm without a governance framework. The patterns are well-documented. That 95% failure rate is alarming, but it’s also useful because it tells you exactly what to avoid.

Good AI is the outcome of everything that comes before the model: the business alignment, the operating model, the data foundation, the talent, the governance, the portfolio discipline, the willingness to iterate. Get those right and the AI part becomes the relatively straightforward part.

At Lerpal, we approach AI strategy the way we approach everything: architecture first, execution always. We help organisations build the conditions for AI to work the operating models, the data infrastructure, the governance structures, the delivery capability so that when AI enters the picture it has somewhere solid to land.

Good AI comes from better foundations. And foundations, as it happens, are what we build.

Explore how Lerpal helps organisations create the conditions for AI that works.