Building AI Generated Presentations
Using generative AI to create quick and compelling narratives
Why is this important?
Earlier this year, we saw the release of ChatGPT, a prompt based search / text generation tool, that reached 100 million users in two months. It’s the fastest that any platform has scaled to this number of users. While ChatGPT has showed incredible success so far, it’s an indication of market maturity for AI generating tremendous value at various levels, all the way from the chip / wafer level to the application layer. We’ve reached an inflection point where the innovation on a chip & foundational model layer has turbo-charged the capability for the application layer to iterate quickly to create value to end consumers. One such use case that stands out is digital presentations. In this article, I’ll dissect this specific application layer use case, to build a balanced hypothesis on these products and how they boost productivity across the workflow, how they position against traditional tools, evaluate total addressable market and provide my perspective on what could differentiate them from one another.
A bit of history…
Microsoft Powerpoint was created back in 1987 by Robert Gaskins and Dennis Austin for the American computer software company Forethought, Inc. The program, initially named Presenter, was released for the Apple Macintosh in 1987. Before PowerPoint, there were slides. Real ones. They were tiny, and tactile, and delicate, and kind of delightful. Making slides was a trained profession for highly skilled designers and technicians
Generative AI & Relevance to Visual Storytelling
AI generated presentation tools engage in visual & textual storytelling by automating the core “plumbing“ & the baseline creative judgement needed to put together a narrative. A user can generate a full blown narrative by:
Inputting an initial prompt on the problem statement or theme.
Feeding the tool, a detailed document which enlists the narrative is a semi-structure format
Leveraging a “co-pilot” while building the narrative in an existing tool, using visual and textual content recommendations, auto-complete & peer-review style functionalities
What’s under the hood?
AI generated presentation tools operate on an application layer level as a consumable abstraction of:
Large Language Models (LLMs) like ChatGPT that have the capability to crawl the internet based on a prompt, extract the right level of information and articulate it in a semi-structured way.
Generative Image models like DALL-E that intake a natural language prompt to create images that visually contextualize those prompts.
Such tools culminate AI generated visual and text outputs in a more structured and compelling way rather than a blob of text or standalone images in a prompt driven chat style product design. They thematically structure the text based responses & generated images into a coherent storyboard, thus creating what we call a “presentation”.
Cost of building presentations
Making a simple presentation with a few slides can take anywhere between a few hours to days depending on the complexity of the narrative. Here’s the cost of building a presentation:
Defining the narrative: As a consultant, I spent 40% of my time building narratives on new proposals, services offerings & existing project updates. A typical storyboarding exercise involves construction of the problem statement, understanding the end audience & applying frameworks to explain the solution.
Research & Data Augmentation: Harnessing facts, metrics, or abstracting raw data to create a compelling story via a data visual such as a chart, trend-line or data tables.
Text & Visuals: Adding succinct bullet points & compelling visuals that can articulate the message in a minimally verbose fashion. The best decks / slides are the ones that lean more on light-weight visuals & minimal words.
Beautification: While this may seem like a counterproductive exercise to some, corporate teams focus heavily on beautification & alignment. Tools such as Power point & Google Slides provide great features to distribute assets, align text / images, yet, this is largely a time consuming process
Brand Compliance: Brand compliance is a mandatory step for established enterprise companies that aim to achieve consistency across all their external and internal narratives. It involves pre-defined color pallets, chart styles, iconography, product abstractions etc
Prominent Players
In the recent past, the number of companies offering AI driven presentation building tools have sky rocketed. Having said that, the market can be split into two major categories:
Existing tools bundling an “AI co-pilot”
This comprises of companies that have offered tradition presentation building tools and are now bundling what most people are calling a “co-pilot”. A couple of representative companies include:
Microsoft Co-Pilot for PowerPoint, which embeds the Microsoft 365 co-pilot built on the OpenAI services, across all Office 365 tools
Prezi, which has been an established presentation tool for quite some time, but is now bundling AI capabilities to enhance user experience
Canva which has been an incredibly beneficial tools for graphic designers. Canva has recently rolled out a “Magic Write” functionality to write text responses & “Text to Image” to generate images.
AI-native presentation tools
Most of these tools have been released in roughly the past 3 - 5 years and are built natively on the GPT & DALL-E / Stable diffusion frameworks, thus providing slightly different user workflows.
The more prominent companies include Tome.app & Beautiful.ai that have a substantial user-base built over the last few years. Tome.app claims to have grown to 1 million users in 2022.
The up & coming companies include Designs.ai, Presentations.ai, Kroma.ai, DeckRobot, Slidebean, Pitch, Vengage
This list is representative and the number of new companies is ever-increasing, given advancements such as GPT-4 and newly released image generation models.
What typically makes a good presentation tool?
Regardless of whether a tool is or isn’t AI driven, following are some key features that create a coherent user experience for a typical presentation tool.
Pre-Built Templates: Pre-built templates have traditionally provided a great head start to building coherent presentations by providing the user with a skeletal storyboard, that they can then build upon.
User Experience: A sizable potion of the world has moved on to web native tools with very responsive UIs and simplistic designs to facilitate ease of use. Tools such as Prezi, Tome.app & Beautiful.ai are a testament to this. On the other hands, tools such as Powerpoint are the 800 pound gorillas with a ton of feature-sets but are relatively hard to access and use.
Integration with Modern Design Applications: Specific tools such as Tome.app & Beautiful.ai provide a deep integrations with design & white boarding tools such as Figma & Miro which are used by modern project teams to visually and textually articulate ideas.
Integrations with Analytics Tools & Databases: Every presentation tool should be able to connect to a database to generate visual representations such as charts or succinctly display tabular data on a slide.
How is AI being injected into different parts of the user workflow?
Automated Visuals & Styling: AI co-pilots speed up the process for researching visuals that are relevant to the scope of information highlighted in the slide, which would otherwise require manually crawling Google Images. This is being injected as prompts that a user can input while building such narratives.
Automated Speaker Notes: Based on the context of the slides, such tools can automatically generate speaker notes to add further relevant information about the topic at hand, while keeping the slide succinct.
Fine-Tuned Storyboarding: Since the underlying LLMs may be pre-trained on thousands of similar storyboards on the same topic, such tools can often provide “SME” level comments on the coherence of the storyboards, and help with enriching the content, which essentially solves for what one calls “Peer review”
Automated Styling & Alignment: Finally, the last few bits of plumbing being automated is the iconography, graphic styling, alignment, which otherwise is a manual and counter-productive process.
Addressable Market
While the the total addressable market for presentations is roughly estimated at 5+ billion USD in an overall workplace productivity tool market of 50+ billion, it’s hard to segregate a specific productivity tool such as a Google Slides or Powerpoint from a broader set of productivity suites such as Google Workplace & Office 365. It is however important to understand the various functions or personas where such tools add tremendous value:
B2B / Enterprise Companies
Management Consulting: Consulting firms spend 20 - 30% of the total time in a project on building narratives in the form of decks. Given that the typically margins for a services engagement can be 25 - 50%, any productivity boosts in streamlining the 20 - 30% effort spent in building presentations helps increase bottomline and improve outcome quality. The predictive & repetitive nature of such storyboards is what makes AI generated presentation tools, a compelling fit for this industry.
Sales Teams: Sales teams spent roughly 50% - 80% of their time in either outreach (typically done by BDRs) and conducting customer demos. Out of this chunk of time, roughly 50% is being spent on crafting presentations. The capability for AI generated presentation tools to create highly contextualized narratives that speak to a customer’s problem through competitive analysis & industry specific research can turbo-charge an AE’s productivity. This indirectly reduces the cost of acquisition for a customer, thus increasing the overall value achieved from the deal.
Product & Engineering Teams: Product managers typically build the vision for a product, translate it into clear product features, prioritize them based on needs of customers & business value, and finally work with engineering teams to execute and implement those features This is articulated in what they call PRDs (Product Requirement Documents) which are essentially narratives aimed at highlighting the core problem, proposed solution and benefits. Some also involve summarizing A/B testing results, generally available user research studies & cost benefit analyses. Building compelling decks from verbose PRDs speeds up the product planning process, thus increasing product velocity.
Early Stage Companies: Early stage companies index heavily on speed to execution, given limited access to capital, ambiguity of the market which they’re trying to capture and the need to iterate quickly on customer feedback. Most efforts around building presentations are spent by extremely nimble sales & management teams (typically founders and early execs) on acquiring new customers and investors to spearhead the company forward. Any tool that boosts productivity by automating the “plumbing” is beneficial for such companies to achieve a faster speed to market.
Business to Consumer
While I’m not an expert at business to consumer markets, the key personas for tools includes (but not limited to) students, job seekers, independent freelance artists and graphics designers, each of whom are looking to showcase their work or credentials to potential customers or companies.
What could differentiate these tools?
Almost every “AI driven” presentation tool may end up using the same set of Model as a Service APIs for information retrieval and search, including GPT, Bard, Cerebras and other companies building foundational models. However, there are moats that each of these businesses can develop:
A two sided economy for training datasets: Data is the goldmine for each of these models and the expansive “Reinforcement Learning from Human Feedback (RLHF)” is what eventually fine tunes accuracy and depth of information. Building a 2-sided incentive model for customers and partners to contribute to the core models by providing the ability to add new data, update existing data or reporting inaccuracies is what could turbocharge the accuracy and stickiness of such products. In return, such customers and partners can earn credits, that discount the total price they pay for using such services via credits. David Sachs from Craft Ventures talks about this in detail in his post on the Give-To-Get Model for AI.
Privately Hosted Models: In extremely regulated industries such as banking, healthcare & pharmaceutical, a lot of narratives or storytelling involves consumption of proprietary data such as drug testing results or medical procedures that are still un-approved or are being tested. In such scenarios, corporations may have a ton of existing content or content frameworks which are often rinsed and repeated to create compelling narratives. The possibility of being able to train a privately hosted model for specific corporations or fine-tuning a privately hosted version of the existing generic model with company specific data can turbo-charge adoption of such tools.
Plug-Ins to Existing Presentation Applications: While natively built AI tools have taken a stab at re-designing the workflow of building presentations from ground up, it is important to acknowledge that a vast variety of users in the world still leverage the existing players in the market such as Powerpoint, Google Slides & Canva. These 800 pound gorillas with their massive distribution advantage can essentially bundle functionalities such as “Co-Pilot” for free, thus putting the distribution channel for the new AI-native presentation tools at jeopardy. While this is an incredibly hard problem to solve, building a “Co-Pilot” style plug-in for such tools which integrates into existing productivity suites may put such tools on an equal footing from a distribution perspective. It still however, it tough to sell a custom co-pilot for a non-zero dollar amount while the core product offers a free version.
Citations: A large variety of presentations, especially the ones that delve into deep market research about an industry, a product or a service require a user to cite sources for the data. While the GPT framework as of the date of writing this article doesn’t expose the citations, it will become increasingly important to quote data sources to avoid usage of any copy-righted or prohibited content in such presentations. This is particularly important for traditional or regulated enterprise businesses that have strict guidelines around what data they can or cannot use.
Brand Compliance: As mentioned before, brand compliance is crucial particularly for big enterprise companies to create consistency, visually & sometimes even from a textual messaging perspective. Brand compliance involves using specific icon styles, deck themes,
Security: Provided that such products focus largely on harnessing public data-sets, adoption of such tools in enterprise businesses requires intense information security reviews. It’s hard to predict the security framework that will be set forth by regulators (if at all), but the companies that achieve mass adoption will have to come up with a transparent framework on what datasets have been used, and providing the customers secure architecture models to host the tools as well as the underlying models.
Who wins the race?
The human brain is not wired to think in exponential terms. In an era where research is being re-written every year (even months), It’s incredibly hard to predict which player wins in the market. Incumbents such as Microsoft & Google could continue to bundle such co-pilots and maintain their position, or some nimble player could come up and disrupt the market. While there may not be a winner-takes-all in productivity tools that provides an application layer on top of the foundational models, the companies that innovate on a chip-set / wafer level such as NVIDIA, Cerebras as well as the ones who build the foundational models, such as OpenAI might end up building the biggest moats. At the end of the day, there is still enormous value creation for such tools, even though the broader market may be highly commoditized in the near term.
References & Notes:
Thanks to ChatGPT and DALL-E for helping me with content review and a title image :)
Generative AI Is Exploding. These Are The Most Important Trends To Know