My 3 guiding principles after testing 18 chatbots
We’ve written on why we think chatbots for farmers in Kenya isn’t as stupid as it sounds. With that in mind I recently reviewed a range of different bots and made a directory of them, along with ideas they gave me for farmer chatbots in Kenya. Check out the full directory with notes on bot UIs and more if you’d like to delve deeper.
In this blog I’ll share three thoughts about how we think about making chatbots for farmers…
1. Don’t build Siri for farmers, but an assistant with a single superpower
Siri for farmers would be nice, but fulfilling the needs of all farmers through one virtual assistant isn’t feasible (yet). Also when bots pretend to be too human and tell you they can answer any question (Turing test style) it almost always ends in a bad experience.
Some of the bots I looked at did one thing really well. Examples include image recognition bot FlowerChecker and a Reminders bot. I can imagine both these being turned into useful bots for farmers: A crop disease checker that is better than your local agrovet in identifying a disease for specific crop types, or a bot that automatically schedules a set of alerts to tell you when to transplant, spray and harvest your tomatoes etc.
Bots should build their personality around their “one super power”. I preferred bots that were allowed to take on their own personas and show off what they did well while being honest about what they couldn’t achieve. Hi Poncho was a nice example. Transparency and honesty makes bots more trustworthy products, and allows more scope for interesting personalities too.
2. Work in the context of digital farming communities and conversations
To be honest many of the bots I tried were services I’d rather get using an app because the UX is nicer (e.g. reading the news, getting the weather, etc). For example, while I liked the idea of Assist I felt the bot was trying to do what I’d do better on google. Instead bots can take advantage of sitting within chat platforms. Their proximity to existing conversations inside these platforms is key because they can operate in the context of conversations in a way apps rarely can. Bots should augment rather than compete with existing conversations.
There were bots that started to build on this idea. For example swelly makes the user aware they’re posting polls to a community, while HealthTap brings advice from a community of real doctors. This being said, no chatbot I’ve seen truly augments the conversations happening inside the mega platforms like Facebook, WhatsApp, Telegram, etc. The volume of digital conversations going on between Kenyan farmers within these platforms is growing – we should start by augmenting what is already being said by farmers.
3. Novelty for the farmer isn’t enough
Obvious perhaps, but I feel most chatbots don’t offer real utility. I’m impressed by the use of backend AI techniques in some bots, e.g. PhotoColorizer (it’s cool, try it — but like me I’m guessing you’ll take a colour picture turn it black & white and then get it recoloured for fun). For bots to be really exciting they have to solve problems.
One good test of this (though it’s not a perfect rule) will be looking at repeat use of the chatbot. If farmers keep coming back to the service it will be more plausible it’s providing real value to them… following up with these sorts of users at an early stage of a chatbot’s life will be key to understanding the utility question in more depth. Repeat use metrics around the bot will (probably) be key.
What’s next?
We’ll be using this directory of bots as design research to help us as we prototype a range of different chatbot ideas with farmers. We’ll keep you updated via the farm.ink blog! We’re excited to be working with IDEO.org and DFID as they lend their design expertise and support along our journey. You can also follow us here on medium to stay tuned and see what we’re up to.