EVERY STEP AI-DRIVEN
The Neste app campaign may not have seemed extraordinary for the average Finns who saw the eventual ads. The way the campaign was created, however, was extraordinary.
Neste and Sherpa joined forces with AI to create a campaign in which every single step was AI-driven. From contract to media plan, from concept to visuals, from copywriting to communication.
As far as we know (and yes, we did ask the AI, too), this is the world’s first campaign that is AI-driven from A to Z.
Done with Docue. Even though there was no ready template for exactly this kind of cooperation, the basic contract template was easily modified. Within one hour, the first draft of the contract was ready for client evaluation. No AI is a match for lawyers, so after small modifications made by the Neste legal team, the contract was ready to be signed.
AI greatly reduced the work needed.
Done with ChatGPT. How to make people in Finland download the Neste app?
Apparently we are somewhat stereotypes for AI, as the first suggestions revolved around elements from nature ‘because Finns are known to value nature’. Thanks to our resident AI nerd Jukka, we were familiar with AI tendency to “hallucinate” – to provide false information that appears correct, and we were prepared not to believe everything at face value.
AI wanted to know more about the goals and target groups, and through discussions and evaluation, the focus became clear. The primary incentive to download would be a discount, supported by responsibility, as the app features help drivers to drive more economically, thus helping both their budget and the planet. It became obvious that the ones asking the questions decide the conclusion to a great length, and it is hard to say how much of the final result can be credited to AI and how much to our strategists.
AI could deliver to some extent, but needed human coworkers to guide it.
Done with ChatGPT. How to reach the audience?
AI was more than willing to assist, but the media plan it provided appeared to be vague and unrealistic at best. Given the limited time and budget available, AI struggled to make realistic budget allocations and media suggestions. There are a lot of current limitations but also plenty of future potential – once the info and data can be trusted, AI can make media planning much easier. Eventually, these channels were selected: Facebook, online displays, DOOH (Digital out-of-home).
AI proved risky at best, and a lot of human coworker guidance was required.
Done with ChatGPT and Copy.ai. What kind of concept would convince Finns to download the Neste app?
We asked for a concept including visuals, headlines, and CTAs – we got disclaimers and a promise to provide ideas for a starting point. (This was most human: AI politely telling us our brief was crap.) We also noticed that we needed to specify Neste a bit better. It is most useful to realise we do not know what AI does not know, and a lot of human experience and general knowledge is missing from the equation. As soon as we explained that basically everyone in the target group knows Neste, and that Neste ranks among the most responsible companies in the world, AI adjusted its approach in text and visuals. The language gap was apparent, and we ended up prompting in English while asking for Finnish copy. Styles, moods and registers proved challenging.
Concerning visuals, several AI solutions were not solutions but briefs: AI suggested a visual that would indicate this and that but not actually define what kind of visual would be able to do this. AI is based on a pool of data, so it provided a lot of ideas that could clearly be labelled as advertising – thus, creativity, wit, humour and emotions were not its core strengths. However, it was able to create generic content with given parameters easily and with good variation – in case you are a bulk provider, it is time to get scared.
AI is not yet terribly creative nor good at thinking different, but very efficient with generic content.
Done with Midjourney Pro, Topaz Gigapixel AI and ChatGPT. We created the visuals using Midjourney with prompts provided by Chat GPT, while final versions were enhanced with Gigapixel AI for higher resolution.
First attempts proved that AI has no respect for names or logos, and wants to recreate and combine everything. It turned Neste into Nsta, Jesta and NSSTia. And once we got a symbolic image of a rather cool and artsy phone with Neste colours and asked AI to add Neste logo, it went all Lovecraft on us and combined the elements just like the transporter pod does to Jeff Goldblum and the fly in The Fly.
The key visual we chose based on AI’s suggestions was a car driving along a forest road. To set this up in a traditional way would have required more time and budget than what we had, but AI handled it nicely, with flares and reflections in all the right places. Obviously AI adds democracy to production: very nice pictures can be produced fast, without a major budget. The Midjourney signature style may feel dated already as you are reading this, but the sheer speed of AI evolution means one can apply any imaginable (or AI-imagined) style or approach convincingly very, very soon.
AI provides quality visuals with minimal time and budget but a human coworker is needed for the right prompts and evaluating the end result.
A couple of words from AI’s human coworkers
AI is a tool that will replace some of the things we do today but also create new things to do instead.
AI is a powerful equaliser on visual production, putting the value on production value.
AI is a good coworker but not a great leader.
AI proves, once again, that data does not equal understanding or insight.
AI does not do everything for us but it does a lot of things smoothly with us.
Toimitusjohtaja / New business ja tarjouspyynnöt
+358 40 746 0427