Confessions of a conversation designer

5 Things I wish I knew before becoming a CxD

elaineinthebay
Bootcamp
10 min readMar 10, 2023

--

meme of two passengers on a train. one passenger is sitting on the side of car with a beautiful sunrise mountain view, while the other passenger is depressed looking at the mountain-side itself, with no light, only rocks. on the sunny side, the text overylayed reads: “me learning about conversation design” on the sad side, the text reads “me interviewing for convo design jobs”
Yes, I’ve posted this meme already. Yes, it’s still relevant :)

Last month, I unceremoniously acknowledged my 2 year anniversary as a full-time conversation designer. I write “unceremoniously” because I’m usually the kind of person to cling to every milestone and anniversary like cat fur on black clothing. I love milestones, and I’ve celebrated them before by writing blog posts 6 months into my role at NLX and at the 1 year mark as well. Although, a lot has happened since— obvious example: I don’t even work at the same company anymore— so it was hard for me to really capture everything I’ve learned about this career path in a formulated (mildly braggy) list of achievements. The way I feel about conversation design is not even remotely the same as before and I’d rather not give the impression that I have everything figured out now.

Let me put it this way: it takes one moment to get married, and an entire lifetime to stay married— that’s how I feel about being a conversation designer. It’s all fun and games until you realize Conversational AI and the design that comes with it is a constantly-evolving field with its underlying technology so much in flux, a principle or tool you hold dear now might become irrelevant 5 years later. Don’t even get me started about how we’re vetting and interviewing this new generation of conversation designers. While I can’t speak to the industry-leader super senior (maybe director) level, applying for conversation design jobs is a harrowing experience I know most people would want to speed through as quickly as possible. That said, I still enjoy my job and continue to find pleasure advocating for my craft, but as far as the rest of the blog post goes, I’ll do my best not to sugarcoat this field, which I’m worried I’ve done too much of.

Below are a series of inner thoughts I’ve had at some time or another in reaction to this industry.

in-game capture from the game Among Us, where one of the members who is actually an imposter gets thrown out into deep space. instead of the usual game text, the text reads, “they never ask how is the imposter”
Also enjoy the series of homegrown memes I’ve made these past 2 years

“Where the hell are all the junior roles?”

Before “breaking into tech”, I had the enormous luxury of gaining seniority the traditional way: start from the bottom and work my way up over time, proving I could take on more responsibilities by performing tasks that required compounded knowledge gained only through experience. During and after college, I started as a student transcriptionist, then translator, then interpreter, then language services lead and conference organizer! I consistently put in the work and the time, and promotions came.

In tech, or more specifically, in conversation design, that luxury doesn’t exist— or if it does, it’s extremely rare. In my 3+ years of snooping around the field, I’ve only met 1 designer who got their start in conversation design through an apprenticeship. I don’t even know of any company other than Meta or Nuance Communications that has offered conversation design-specific internships annually (FYI to anyone applying for summer internships, Disney just opened one).

As a consequence of this quirk of zero junior roles, there are more imposters in the field than there are designers. I don’t mean “imposters” literally because there are a lot of extremely talented designers who’ve made it work, but rather, there is now an entire industry of designers who were thrown into the role that now suffer from imposter syndrome and have not been given the necessary tools to onboard or up-skill successfully. There are no recent grad roles, and even if you do stumble upon a job post with entry-level qualifications, it’s highly likely the role will require the designer to perform beyond their level (being asked to own their own research, field their own requests and timelines, run their own data analysis, present to their own clients, etc).

For that reason, many junior conversation designers in search of their first role might find the CxD interview process extremely difficult. Companies will sometimes narrow their search for that one CxD unicorn that can literally do it all. By the time an opportunity rolls around that junior CxDs might actually have a chance to pass the interview, some of them have already pivoted into other tangential fields— UX writing, content strategy, digital media operations, or product management (to name a few real life examples). It’s an extremely sad moment to lose a potential conversation designer, but what’s more distressing is knowing that realistically, they would’ve broken into the field as a solo conversation designer, with few chances for direct mentorship.

Conversation design simply isn’t for everyone. If you’re someone who is not easily scared of having to fight it out, survive the interview process, AND on top of that, perform at your max for your first year, then have at it!

“Wait, you’re a senior CxD after one year?”

It’s kind of the unspoken rule that if you’re able to complete one entire year of full-time work (or contract work with full-time hours), then you’ve officially graduated from conversation designer to senior conversation designer. There’s no true mid-level conversation designer out there because, by default, the role requires to perform your day-to-day responsibilities as well as evangelize the value of your profession within your workplace. As senior conversation designer Christy Torres puts it: “Where [the work got] a little more complicated was once we started implementing, helping to convey why things needed to be done a certain way. So [there is] a lot evangelization necessary to really break down [to your stakeholders], ‘These are the best practices in UX in order to achieve the goal that you would like[.]’”

Unlike product design team structures or career levels (IC3, IC4 or L3, L4, L5, etc), a lot of conversation design teams behave under a consulting firm-like operating model, wherein all individual contributors work on their own projects, possibly with their own clients, with few points of collaboration with their fellow designers, except where high-level guidelines are involved. The team itself usually sits outside of or isolated from the company’s design team (if one exists) and is called upon from various different departments across the company for (ad-hoc) consulting work. Again, to survive in this kind of environment, the conversation designer must acquire the skills and knowledge to perform their duties, but also be a fast learner (in regards to learning about their building platform), expert communicator (when presenting to stakeholders who are not familiar with conversation design), and educator.

After working in this operating model for at least a year, it’s no wonder that conversation designers quickly level up in their career. One year also gives the outward credibility that you now know what you’re doing (because you weren’t fired or your contract didn’t expire which means you might’ve helped make *some* money for the business), so at this time a lot of companies and 3rd party recruiters will start picking candidates from designers with at least one year of work experience.

screenshot of a web chat interaction with the UPS chatbot. the user chat bubble reads “you suck” while the chatbot bubble reads, “I’m sorry I haven’t been able to help you. As a virtual agent, my reason for existing is to help, but I can’t answer everything… yet”.
We break bots to learn about bots 💅🏼

“Conversation design isn’t the same everywhere?”

No, it’s not voice designers on one side and chatbot designers on the other. Voice and chat are first and foremost modes of delivery. The UX strategy can sometimes be the same. In fact, when I was working in my previous role at NLX, I’d often take inspiration and learning from chatbot designs or chat-focused designers, even though the bulk of my projects were voice or multimodal. The *real* differentiator is: do you work in customer service (enterprise, B2B) or consumer product?

Customer service and consumer product are vastly different spaces and it’s a big part of the reason why the UX behind something like Siri looks and feels so different from something like a banking IVR. Customer service is all about cutting inefficiencies in a current process and/or automating manual tasks to help the process work at scale. Consumer product is about getting a customer invested and hooked enough to buy your product and keep using it.

For customer self-service design, you might optimize an experience to help a customer achieve a goal quicker than they would without your conversational product and thus, you may choose to track the success of your experience by tracking the time spent on the task or the platform. If your user spends a REALLY long time with your experience, it might mean something is broken in the flow. For consumer product design, you’re all about enticing your users to stay and (hopefully) love your product, so some of your success metrics might be more related to retention and looking at daily usage count. Granted, we now have the cautionary tale of Alexa: a former Alexa AI engineer explained, “[w]hen the business cares more about traffic/number of interactions, teams create chatty experiences to improve those metrics, which doesn’t necessarily mean added value to the user.”

That is not to say that these tactics always carry across different modalities or look entirely the same for multimodal experiences. For me personally, as soon as I stepped into consumer product, a lot of what I knew about conversation design went out the window. It’s a whole different ball game to work on a voice assistant! I wish I could write about it, but for the sake of my NDA’s, I’ll just say it’s more entertaining and liberating (in terms of what you can consider “moments of delight” in a conversational experience).

two PET scans side by side. on the left is a depressed brain, with mild neural activity. on the right, is a human brain “after updating months worth of conversation design documentation.” the PET scan on the right is aglow with massive amounts of neural activity and stimulation.
Idk about y’all but I love documentation

“Why is documentation so hard as a CxD?”

Conversation design and the “conversation design” role was standardized in large part due to Google’s design team. Thanks to them, we now have an umbrella term for the people working to improve the inputs and outputs of these large Conversational AI experiences. However, an umbrella term is still an umbrella term, and ultimately, CxDs cannot agree on a uniform style of documentation because sometimes there literally is no “one” way to do conversation design. Not all projects use the same underlying technologies (rule-based vs. machine-learning AI models), or cover the same interaction models (single-turn vs. multi-turn dialogues).

It’s always interesting to hear conversation designers talk about the tools of the trade or the format of their deliverables because it’s (generally) always different. Some designers might be die-hard flowchart fans because of its ease of shareability, others might look to more dynamic-style flows and get really in the weeds with their platform to ensure smoother handoff.

The one aspect about documentation that does seem to unite us all though is the issue around versioning. Depending on the format of the CxD documentation, it might be hard to easily access all version histories of a design or keep track which prompts are true for which dates. If you’re working straight from a platform, then you’re reliant on that particular software to have versioning included. If you’re working from a flowchart or spreadsheet, then it becomes a manual process to update and revise all possible prompts in an experience, which is hard to maintain at scale.

photo taken from the Cannes film festival of a cast for a movie where all the actors have varying levels of importance and reputation. on the left is actor timothee chalamet described by overlayed text as “multimodal experiences”. to the right of him stands director Wes Anderson with the overlayed text, “chatbots”. next is actress Tilda Swinton described as “voice assistants”. finally, is actor Bill Murray, likened to IVRs.
A very niche meme

“Sounds like there is no ‘Voice-First’ future”

This one was a bummer to realize, but ultimately, what we build with voice as designers and what users actually use voice for are very distant, separate things in today’s world. I do believe in voice as a modality, but I also recognize even with widespread adoption or high-levels of comfort using voice does not guarantee that voice is the best input for a particular interaction. From the accessibility angle, or from the “sensory design” angle (who are we leaving out of the conversation by forcing voice as an input?), voice should not be the only way users should be allowed to interface with their digital products, and by extension, their world.

I cannot imagine a voice-first or voice-only future, but I do see us approaching a reality where we extend the capabilities of our senses with digital assistance, in whichever form that may be. We need to be flexible in our design, not rigid. When we propose voice to a new interaction, we must also consider if the barriers to voice or its constraints are outweighed by the value it brings to an experience. We also need to consider the consequences of our designs should they include voice— where that voice data gets processed and who is allowed to access that information. And (though not directly voice-related), we need to question hype around our industry and the new wave related to things like ChatGPT where, as linguist Emily M. Bender states, “there’s too much effort trying to create autonomous machines, rather than trying to create machines that are useful tools for humans.”

As designers in the room where the magic happens, we owe it to our users to uphold ethics in our design process.

Final Thoughts

There will always be a demand for conversation designers because there will always be a demand for automation and innovation. If you’re still interested in conversation design as a career, then let me say: working in this field has been equal parts terrifying and rewarding. I don’t always do as much design as I’d like (because of the evangelism part of the job), and I certainly don’t do as much writing as I initially thought I would (this really isn’t a creative writing job), but it’s still what I want to do.

meme template changed into conversation design humor. on the left a woman is blushing and saying, “thank you for changing my life.” on the right, is the Siri voice assistant, responding with, “I didn’t get that. could you try again?”
Not the meme I shared to my cohort right before presenting my capstone project

Elaine Anzaldo is a seasoned Conversation Designer, having worked on voice technologies at companies such as Meta, NLX, Apple, and SRI International. As a designer for both influential voice assistants and the customer self-service industry, she has created natural conversational artifacts for voice, chat, and multimodal interfaces.

Follow on: LinkedIn | Instagram | YouTube

--

--