How do I show complex ideas in a presentation?
March 29, 2024 3:46 PM Subscribe
I'm really struggling with creating presentations. I need to show AI use cases (basically product recommendation engine) and other places "generative AI" can be used. The key is to sell this internally so sales, which has no understanding of AI outside of ChatGPT, can sell this to their clients who absolutely need to invest in AI. Does anyone have any ideas on how to do this in a limiting presentation environment like PowerPoint or PDF? Keep in mind it must demonstrate how AI works in one of those boring consulting type presentations.
Due to layoffs our marketing team is basically gone. The requirements sound silly, and my own intuition would be "AI powered recommendation engines result in up to 20% better conversions." There's a lot of gotcha's here:
- We have no product we sell consulting services, but as an engineer I know what's capable regardless of the final product they buy/build/can afford.
- For some reason we have a mandate everything must be AI and "show AI" so that sales can understand how to sell AI. I don't agree, it is what it is.
- Client also doesn't understand AI, but for some reason has a budget to do "something" with AI.
- When I say doesn't understand AI I mean I said exasperated in a meeting, not all AI is ChatGPT you know that right? They asked what other AI there could possibly be.
- If I show abstract things like a wireframe of a website and say "content is labeled first by the AI" they won't get that.
- It needs to be in PDF/PPT.
- I plan on buying Adobe CC just to see how far free templates and images get me.
Here's my current idea:
1. Create a fake product site so I can generate an image using (ha!) AI.
2. Create/find an icon that represents text classification and just call it GenAI, like a brain or computer-y looking thing.
3. Use slides to "animate" the experience, as in the 3-4 pages are unlabeled, next slide introduces the gear, and says in marketing-speak that it has automatically labeled the pages. Think "men's hightop sneaker" with the fake page showing something that looks like that. A "negative" product, like dress shoe, etc.
4. Next slide introduces a visitor and each slide shows him going from page to page building up a profile, "Interested in men's athletic wear: 0, Interested in dress wear 0" to start out.
5. Of the 3-4 fake pages if he keeps going to sneakers the score rises. If he goes to the dress shoe but spends little time on it, that score drops.
6. Finally last slide he's presented with a page that represents a complete outfit for men's sportswear.
That's a lot of slides and breaks every presentation rule in the book. There's no real executive slide that summarizes it. Creating a fake website that showcases real products I'm hoping can be done (poorly) by AI. There's a weird line because the client will say "we don't sell clothes that won't work for us" and if I don't put something tangible like a widget they will be confused.
I have to create a lot of use cases and I found out there's a ton of SaaS products that just create an interface off a model found of Hugging Face and then put optimal use cases in their marketing. That's actually easier for me to verify its possible. I'm hoping Adobe CC will have enough AI/stock images/icons/infographs to pull off 80% of the creative part of this, is that correct? Also is there a better way to show this except by using slides as keyframes basically?
I can't find any presentations that demoes stuff like this. Putting it online using InDesign's fancy export tool won't be possible. Trust me I tried explaining we need to not give them a book of use cases and that's a lazy way of selling, etc. and we should be using the Pareto Pyramid and structured problem solving. But it is what it is.
Due to layoffs our marketing team is basically gone. The requirements sound silly, and my own intuition would be "AI powered recommendation engines result in up to 20% better conversions." There's a lot of gotcha's here:
- We have no product we sell consulting services, but as an engineer I know what's capable regardless of the final product they buy/build/can afford.
- For some reason we have a mandate everything must be AI and "show AI" so that sales can understand how to sell AI. I don't agree, it is what it is.
- Client also doesn't understand AI, but for some reason has a budget to do "something" with AI.
- When I say doesn't understand AI I mean I said exasperated in a meeting, not all AI is ChatGPT you know that right? They asked what other AI there could possibly be.
- If I show abstract things like a wireframe of a website and say "content is labeled first by the AI" they won't get that.
- It needs to be in PDF/PPT.
- I plan on buying Adobe CC just to see how far free templates and images get me.
Here's my current idea:
1. Create a fake product site so I can generate an image using (ha!) AI.
2. Create/find an icon that represents text classification and just call it GenAI, like a brain or computer-y looking thing.
3. Use slides to "animate" the experience, as in the 3-4 pages are unlabeled, next slide introduces the gear, and says in marketing-speak that it has automatically labeled the pages. Think "men's hightop sneaker" with the fake page showing something that looks like that. A "negative" product, like dress shoe, etc.
4. Next slide introduces a visitor and each slide shows him going from page to page building up a profile, "Interested in men's athletic wear: 0, Interested in dress wear 0" to start out.
5. Of the 3-4 fake pages if he keeps going to sneakers the score rises. If he goes to the dress shoe but spends little time on it, that score drops.
6. Finally last slide he's presented with a page that represents a complete outfit for men's sportswear.
That's a lot of slides and breaks every presentation rule in the book. There's no real executive slide that summarizes it. Creating a fake website that showcases real products I'm hoping can be done (poorly) by AI. There's a weird line because the client will say "we don't sell clothes that won't work for us" and if I don't put something tangible like a widget they will be confused.
I have to create a lot of use cases and I found out there's a ton of SaaS products that just create an interface off a model found of Hugging Face and then put optimal use cases in their marketing. That's actually easier for me to verify its possible. I'm hoping Adobe CC will have enough AI/stock images/icons/infographs to pull off 80% of the creative part of this, is that correct? Also is there a better way to show this except by using slides as keyframes basically?
I can't find any presentations that demoes stuff like this. Putting it online using InDesign's fancy export tool won't be possible. Trust me I tried explaining we need to not give them a book of use cases and that's a lazy way of selling, etc. and we should be using the Pareto Pyramid and structured problem solving. But it is what it is.
Response by poster: I should add some clarification I have familiarity with the Adobe suite but am primarily using it running under the assumption their AI tools, stock templates, images, etc. working in PowerPoint directly is horrible. I’m familiar enough with layers and grouping and frankly any photo editing beyond MS Paint will pay for itself.
The secondary question is really showing something as complex as scoring by visiting pages. Feels really like it needs to be a demo with a ranking in a video or showing maybe lines on one slide hoping they get the concept the user is visiting pages and their recommendation ”weight” has changed? That seems like a very busy and confusing slide.
posted by geoff. at 4:35 PM on March 29
The secondary question is really showing something as complex as scoring by visiting pages. Feels really like it needs to be a demo with a ranking in a video or showing maybe lines on one slide hoping they get the concept the user is visiting pages and their recommendation ”weight” has changed? That seems like a very busy and confusing slide.
posted by geoff. at 4:35 PM on March 29
I see two ways to go with this challenge.
1) This is a case where user centered design can be pointed at your client -- in this case, that's the internal sales team. You want to find, say, 3 problems they actually feel pain from. They might sound like "it's hard to sell our services because they're so ambiguous" or "I wish we could find the decision-makers who get our value proposition" or "I just don't see how AI could be valuable in my customer's key industry". Whatever they are, that's your research to do. Collect more pain points than you're going to use, because you want to find ones that help you demonstrate your point.
Then you create stories that demonstrate how AI can be used to solve those problems. Fictitious stories are the same kinds of tools sales people use (in addition to data about outcomes and features, and case studies from past customers) in their sales process. Speaking their language will reach them in ways that speaking your own language to them will not.
2) Alternately, you could just make an "AI for dummies" style deck (don't name it that) and share it at a Lunch & Learn or some kind of everyone-welcome session. You could become legend at your company, for helping everyone understand. You could also, in the process, position yourself as the expert on the Key Hyped Technology Everyone Needs, and offer your opinions on what's cool about AI and what its drawbacks are.
posted by nadise at 6:04 PM on March 29 [4 favorites]
1) This is a case where user centered design can be pointed at your client -- in this case, that's the internal sales team. You want to find, say, 3 problems they actually feel pain from. They might sound like "it's hard to sell our services because they're so ambiguous" or "I wish we could find the decision-makers who get our value proposition" or "I just don't see how AI could be valuable in my customer's key industry". Whatever they are, that's your research to do. Collect more pain points than you're going to use, because you want to find ones that help you demonstrate your point.
Then you create stories that demonstrate how AI can be used to solve those problems. Fictitious stories are the same kinds of tools sales people use (in addition to data about outcomes and features, and case studies from past customers) in their sales process. Speaking their language will reach them in ways that speaking your own language to them will not.
2) Alternately, you could just make an "AI for dummies" style deck (don't name it that) and share it at a Lunch & Learn or some kind of everyone-welcome session. You could become legend at your company, for helping everyone understand. You could also, in the process, position yourself as the expert on the Key Hyped Technology Everyone Needs, and offer your opinions on what's cool about AI and what its drawbacks are.
posted by nadise at 6:04 PM on March 29 [4 favorites]
Does the presentation have to be about recommendation engines? It may be a bit of a tough sell to convince people that Netflix style user profiling ("because you watched X you might like Y') is AI, because it predates the current AI hype cycle by quite a lot, even if some of the underlying ML concepts are the same.
posted by BungaDunga at 7:06 PM on March 29
posted by BungaDunga at 7:06 PM on March 29
This is a use case for embedding animated GIFs into your PPT. Using AI to help you generate the fontent for your embedded GIF is a good idea.
posted by shock muppet at 8:15 PM on March 29
posted by shock muppet at 8:15 PM on March 29
There are two things here: message, and medium. It sounds like both are up in the air, with the overall goal being a bit unclear. Are you fully in control of both? Like your goal is to both come up with prospective use cases and communicate them?
As to medium, embedded videos or GIFs may be helpful. In presenting on ChatGPT uses, I found video demonstration was much more visceral in making people see the possibilities. Particularly, anything showing streaming text (word by word) at a readable pace gave the "vibe". You could mock this up with pre-determined text to give the impression and convey the idea of dynamic interaction.
posted by lookoutbelow at 9:24 PM on March 29 [2 favorites]
As to medium, embedded videos or GIFs may be helpful. In presenting on ChatGPT uses, I found video demonstration was much more visceral in making people see the possibilities. Particularly, anything showing streaming text (word by word) at a readable pace gave the "vibe". You could mock this up with pre-determined text to give the impression and convey the idea of dynamic interaction.
posted by lookoutbelow at 9:24 PM on March 29 [2 favorites]
You can do a live demo where you pose the ML model this problem and -- subject to you developing good prompts for the sales journey you want to tell -- have ChatGPT or other respond like it has the data and the recommendations.
Something like: "Please describe an example in sales language where we are able to optimise sales recommendations in $client_market using AI."
posted by k3ninho at 2:13 AM on March 30
Something like: "Please describe an example in sales language where we are able to optimise sales recommendations in $client_market using AI."
posted by k3ninho at 2:13 AM on March 30
Response by poster: Thanks all. ChatGPT or Dall-E hallucinates too much, but gave me some ideas. A recommendation engine is hard and there’s no one way to do it. Furthermore I can broadly explain features/hyperparameters that would go into it but I never visualized generated parameters. Thats the whole point of deep learning right? But ChatGPT did give me a spider graph I can make up and not get bogged down in figuring out how far I can go explaining clustering and the problem of people buying just the same things it won’t recommend anything outside of that (there’s a term for that I forget). Really Netflix recommendations vary so much from Amazon in a simple executive slide not knowing the product it looks like pre-deep learning just normal algorithms.
Cleverly using Adobe and generating a generic product site which I used AI to filter with fake data (ha!) I can just go down a list of popular SaaS AI products for sites that I don’t get why it’s such a big deal (replacing background on product images?!), and ignore it only works under optimal conditions. A lot of these are for sites like Etsy or Amazon that have millions of products generated by users and seemingly want a more professional photo shoot look to take the background and put it one shot by a photographer. Cleverly placed like a cloud inset have so many offerings which are 90% models you can find on HuggingFace with a nice SaaS interface they get big name clients for their “photoshop replacement tools” for lack of a better word and wow them with work background replacement. Nike might use it for these purposes or to rotate a shoe in 3D bit like Chanel isn’t going to sacrifice a photo shoot. A lot of the gee whiz stuff is for sites with 10+ million user generated content. And of course with some image trickery and picking the background that matches the lighting works sort of. Not knocking the idea I didn’t realize a of SaaS platforms just wrapped things in a nice interface and threw so many products together they attracted such diverse clients mid range clients who don’t have strong marketing or ML teams get enamored.
So I’m taking one use case making a couple of variations of different technical ability trying to explain multi text classification, image classification and determining model fot at a high level just to see how that plays and others create UX widgets for lack of a better term. I’d really like to reframe this into finding a solution and defining the problem but there’s nothing these SaaS platforms do that Photoshop couldn’t do especially for internal marketers. Still AI.
Getting them (sales, management) to think that AI does not include just ChatGPT and you can’t just throw data at a model to “retrain” it is not the ask but at a high level I hope that clicks.
Frankly a lot of sites that are big B2B brands (think large chemical companies) have no KPIs defined and frankly when you’re talking about an industrial used in regulating some obscure manufacturing process the best they could do is have a simple chat prompt because I’m guessing the sales process is so complex you don’t just happen to be building a billion dollar factory without knowing about it. More likely you’re downloading datasheet to plugin numbers to your estimate. There might be a photo of a guy in a hard hat spraying foam, followed by 200 pages of obscure engineering documents and data. AI could be really useful if the front page was just a chat prompt where you could go “how fast does H1b342 core at 90 C” instead of a giant meganav and blog posts.
But I digress, thanks all. A combination of ChatGPT, very simple product site to show realism and some UX widgets along with Adobe CC icons and stuff go a long way.
Nadise’s first point is great but I can’t talk to sales until I have use cases which is backwards. I’m fully in control but not as with any big organization.There’s a ton of politics.
I’d like to do embedded gifs but I get awfully close to to creating a UI in the gif the client and sales gets stuck on using gradio or whatever. Keep in mind they are used to selling SAP or salesforce that have polished product demos. I was hoping to keep this as agnostic as possible.
I think the fake website with end result screenshots plus Adobe and ChatGPT is my best bet. Again AI makes a lot of decisions for you so a UI with a slider graph isn’t necessary. Really AI in marketing is limited to meeting KPIs that traditional non-ML excels at without marketing intervention or basically CSR replacement force-commerce. Maybe better content generation and tone along with generating social posts, etc from one source of content. Then there’s a whole host of domain specific things that demo well for certain clients but aren’t applicable to traditional brands. If you’re not measuring KPIs or something it is very hard to say AI works.
Companies that are not tech centric don’t realize you can’t tell if something works without defining the problem and coming up with criteria of what working is (especially corporate sites with no ads and a lot of direct sales. There’s ways that fall under strategy but it is a larger non-AI issue.
Bottom line they were expecting an SAP or Salesforce approach that had a working product with use use case and could “train” 20k “AI Engineers” on whatever it is these places use for AI. I’m on the custom side an am used to well defined problems where you may intuitively know something resembling ML + stats will work. So while I’m used to sales and marketing this is… odd to me. And I was given a month to come up with a marketing and implementation strategy.
Sorry to complain you’ve all helped a lot. Of they don’t like what I come with so be it. I knew this wasn’t going to be true AI and just take existing solutions and plug them in but I should have looked at big enterprise players and looked at how they marketed it.
posted by geoff. at 2:07 PM on March 31
Cleverly using Adobe and generating a generic product site which I used AI to filter with fake data (ha!) I can just go down a list of popular SaaS AI products for sites that I don’t get why it’s such a big deal (replacing background on product images?!), and ignore it only works under optimal conditions. A lot of these are for sites like Etsy or Amazon that have millions of products generated by users and seemingly want a more professional photo shoot look to take the background and put it one shot by a photographer. Cleverly placed like a cloud inset have so many offerings which are 90% models you can find on HuggingFace with a nice SaaS interface they get big name clients for their “photoshop replacement tools” for lack of a better word and wow them with work background replacement. Nike might use it for these purposes or to rotate a shoe in 3D bit like Chanel isn’t going to sacrifice a photo shoot. A lot of the gee whiz stuff is for sites with 10+ million user generated content. And of course with some image trickery and picking the background that matches the lighting works sort of. Not knocking the idea I didn’t realize a of SaaS platforms just wrapped things in a nice interface and threw so many products together they attracted such diverse clients mid range clients who don’t have strong marketing or ML teams get enamored.
So I’m taking one use case making a couple of variations of different technical ability trying to explain multi text classification, image classification and determining model fot at a high level just to see how that plays and others create UX widgets for lack of a better term. I’d really like to reframe this into finding a solution and defining the problem but there’s nothing these SaaS platforms do that Photoshop couldn’t do especially for internal marketers. Still AI.
Getting them (sales, management) to think that AI does not include just ChatGPT and you can’t just throw data at a model to “retrain” it is not the ask but at a high level I hope that clicks.
Frankly a lot of sites that are big B2B brands (think large chemical companies) have no KPIs defined and frankly when you’re talking about an industrial used in regulating some obscure manufacturing process the best they could do is have a simple chat prompt because I’m guessing the sales process is so complex you don’t just happen to be building a billion dollar factory without knowing about it. More likely you’re downloading datasheet to plugin numbers to your estimate. There might be a photo of a guy in a hard hat spraying foam, followed by 200 pages of obscure engineering documents and data. AI could be really useful if the front page was just a chat prompt where you could go “how fast does H1b342 core at 90 C” instead of a giant meganav and blog posts.
But I digress, thanks all. A combination of ChatGPT, very simple product site to show realism and some UX widgets along with Adobe CC icons and stuff go a long way.
Nadise’s first point is great but I can’t talk to sales until I have use cases which is backwards. I’m fully in control but not as with any big organization.There’s a ton of politics.
I’d like to do embedded gifs but I get awfully close to to creating a UI in the gif the client and sales gets stuck on using gradio or whatever. Keep in mind they are used to selling SAP or salesforce that have polished product demos. I was hoping to keep this as agnostic as possible.
I think the fake website with end result screenshots plus Adobe and ChatGPT is my best bet. Again AI makes a lot of decisions for you so a UI with a slider graph isn’t necessary. Really AI in marketing is limited to meeting KPIs that traditional non-ML excels at without marketing intervention or basically CSR replacement force-commerce. Maybe better content generation and tone along with generating social posts, etc from one source of content. Then there’s a whole host of domain specific things that demo well for certain clients but aren’t applicable to traditional brands. If you’re not measuring KPIs or something it is very hard to say AI works.
Companies that are not tech centric don’t realize you can’t tell if something works without defining the problem and coming up with criteria of what working is (especially corporate sites with no ads and a lot of direct sales. There’s ways that fall under strategy but it is a larger non-AI issue.
Bottom line they were expecting an SAP or Salesforce approach that had a working product with use use case and could “train” 20k “AI Engineers” on whatever it is these places use for AI. I’m on the custom side an am used to well defined problems where you may intuitively know something resembling ML + stats will work. So while I’m used to sales and marketing this is… odd to me. And I was given a month to come up with a marketing and implementation strategy.
Sorry to complain you’ve all helped a lot. Of they don’t like what I come with so be it. I knew this wasn’t going to be true AI and just take existing solutions and plug them in but I should have looked at big enterprise players and looked at how they marketed it.
posted by geoff. at 2:07 PM on March 31
« Older Restaurant in New Haven, CT that served sandwiches... | Can He Keep the Man From Taking His Money? Newer »
You are not logged in, either login or create an account to post comments
posted by underclocked at 4:30 PM on March 29 [1 favorite]