This article is written from Jakob Mayer, CEO of SlideLizard, who has spent the last decade building software that takes the manual work out of enterprise presentations.
Every week I talk to people who run large organisations – operations directors, heads of strategy, sales leaders. And at some point in almost every conversation, the same thing comes up. Not as a complaint, more as a resigned observation:
We still spend an insane amount of time on PowerPoint.
Not because they lack tools. They have Copilot. They’ve tried the AI deck generators. They have corporate templates, brand guidelines, a design team somewhere. And still, a senior analyst is spending three hours on a Friday afternoon updating a KPI slide because the numbers changed.
I used to think this was a workflow problem. After years of building SlideLizard, I’ve come to believe it’s something more fundamental. It’s what we now call the last mile problem of AI presentations – and it’s exactly the problem LIZ AI was built to solve.
-
Contents
1. Why AI Deck Generators Don’t Solve This
When the first wave of AI presentation tools arrived, the promise was obvious: describe your deck, get your deck. Upload a PDF, receive 30 slides. Type a prompt, watch the magic happen.
The problem is that’s not actually how enterprise presentations work.
A pitch deck for a CFO is not the same as a company overview for a new hire. A management update for the board is not the same as a project status for the team lead. Each of these follows an implicit logic – a structure, a language register, a density of information – that no generic AI tool has ever been trained to understand.
What you get instead is 30 technically correct but contextually wrong slides. And then someone still spends two hours fixing them.
We started calling this the “last mile problem” of AI presentations: the AI gets you 60% of the way there, and then you’re on your own for the part that actually matters. LIZ AI exists to close that gap.
2. The Enterprise Privacy Problem
There’s a second problem that stops most enterprise teams before they even get that far. Enterprise presentations contain exactly the kind of information that should never leave the building: revenue forecasts, M&A discussions, strategic roadmaps, unreleased product data.
The standard answer from AI vendors – “we take privacy seriously and your data is encrypted” – is completely beside the point for anyone who has sat in a procurement review. The data still leaves the building. That’s the issue.
Solving it requires rethinking where the security boundary sits architecturally, not just updating a privacy policy. LIZ AI’s approach: sensitive content never leaves the local environment. Real numbers are replaced with placeholders on-device, sensitive text is stripped out before anything reaches a model. The LLM receives a structural skeleton. The original data never travels.
3. A Different Mental Model: The Presentation Cycle
This isn’t a new observation. People have tried to automate presentation work for years. The reason it’s worth revisiting now is that the underlying technology has fundamentally shifted. Foundation models have reached a point where contextual understanding – of language, structure, audience, intent – is good enough to build domain-specific logic on top of.
The insight that sharpened how we approached LIZ AI came from watching how enterprise teams actually work with presentations.
They don’t create decks once. They maintain them. They update them. They adapt them for different audiences, different markets, different moments in a product cycle. A presentation is never finished; it’s in a permanent state of evolution.
Classic AI tools think in output. Generate. Done.
Real enterprise presentations live in a cycle: create, review, update, adapt, check for brand compliance, update again, adapt for a different audience, repeat. The moment you think about it this way, it becomes obvious why a one-shot generation tool can’t be the answer.
4. What a Large Presentation Model Actually Means
LIZ AI is built on what we call a Large Presentation Model. I want to be careful here, because the term gets thrown around loosely.
A Large Presentation Model is not a large language model that has seen lots of slides. It’s a system that understands the logic of professional presentations: how different slide types function, what structural patterns work for which use cases, how brand and content relate to each other, and how a single slide fits into the context of the whole deck.
The difference in practice looks like this: when a sales manager needs a competitive comparison slide, LIZ AI doesn’t just generate two columns of text. It understands that a comparison slide in a sales context needs a clear visual winner, a specific density of claims, and a language register that builds confidence. It knows the difference between a board-level executive summary and a project status update – not because it was told, but because it understands presentation logic.
More importantly: LIZ AI works selectively. If one element on a slide doesn’t work – a headline, a text block, a data label – you mark it, describe what you want, and only that changes. Nothing else. The rest of the slide stays untouched.
This sounds like a small thing. It isn’t. The biggest frustration with current AI tools isn’t the output you get on the first try. It’s that any attempt at improvement breaks everything else. Selective editing is the difference between a tool you use once and a tool you use every day.
5. Where LIZ AI Is Going: Agentic Presentations
Agentic AI – systems that don’t just respond to prompts but take initiative, monitor conditions, and act autonomously – is about to hit enterprise workflows in a way that most companies are not prepared for.
For presentations, this means something concrete: imagine a quarterly board deck that updates itself. Not because someone ran a macro, but because LIZ AI monitors your connected data sources, detects that the revenue figure changed, checks whether the updated number affects the narrative logic of adjacent slides, and makes the adjustment – flagging it for human review before the next meeting.
The obvious objection: the data that lives in your systems is rarely the data that ends up on the slide. Most teams don’t pull a number and put it in. They adjust it, cross-reference it, apply judgement calls that live in someone’s head, not in a database.
This is true. And it’s actually the more important problem. If the preparation step before a presentation requires manual stitching of five sources and a set of undocumented rules, that’s a sign that the system of record hasn’t caught up with how the business actually works. An AI that exposes that gap is more valuable than one that papers over it.
For teams that know exactly what that manual step looks like: LIZ AI can learn it. The logic of combining sources – this number from the CRM, that figure adjusted by the latest forecast, this text pulled from the latest strategy doc – can be defined once and then applied automatically. The human judgement doesn’t disappear. It gets encoded.
6. The Last Mile Was Always About Keeping Slides Right
The technical components exist. What’s missing in most organisations is the layer that understands presentations specifically – the contextual logic, the brand rules, the audience-specific adjustments – well enough to be trusted with autonomous action.
That’s the direction LIZ AI is moving. Not faster deck generation. A system that understands your presentation infrastructure the way a senior colleague does – and can take care of the maintenance work so that the humans in the room can focus on the thinking.
The last mile was never about generating slides. It was always about keeping them right.
