B2B SaaS · AI Product Strategy · Digitalog · 2024
Conma
An integrated marketing platform with no AI in it and a board that wanted some. I set the strategy and picked the two least glamorous features on the list. Both shipped and are still in use.
- Role
- Product Manager, contract
- Org
- Digitalog · Seoul
- Product
- B2B integrated marketing platform
- Timeline
- Sep 2024 to Nov 2024
The situation
Conma is an IMP, an Integrated Marketing Platform: one place for a business to plan and run its marketing rather than four tools that do not talk to each other. It was called Digitalog while I was there and has since been renamed. I came in as contract Product Manager at an early stage start up on government seed funding, and owned the AI strategy and roadmap alongside the beta and feature launches.
The pressure with a first AI capability is to pick something impressive. The risk is that you ship a demo, it gets used twice, and the next AI proposal is harder to fund than the first one was.
The first AI feature decides what a company thinks AI is for.
What I chose, and why those two
Auto generated hashtags and written content for a social post, and interaction analytics for marketing performance. Neither is a headline. Both sit on a task a marketer already does every day, which means the feature is judged on whether it saved time rather than on whether it was clever.
I worked the generation feature out with the developers instead of handing it over. We went through how the model would be prompted, what the platform had to send it, and what came back. That set the scope against the API that existed, not against a wish.
How the generation feature works
- LLM
- Large language model. Software that predicts text, so it can be asked in plain language to write something and will produce a plausible draft of it.
- Prompt
- The instruction sent to the model. Its wording largely determines the quality of what comes back, which is why we built it rather than asking the marketer to.
- OpenAI API
- The paid service the platform sends prompts to and receives generated text from. API means the way one piece of software talks to another.
Alongside it
The second capability was interaction analytics, reading engagement back as marketing performance. Generation without measurement leaves a marketer unable to say whether any of it worked, which is the question their own client asks them first.
Both shipped and are in use. What made the AI worth adding was the thing underneath it. The platform builds marketing calendars against a specific country, so a campaign lands on Chuseok in Korea and not on a date that means nothing there.