Another title for this could have been "how a failure to communicate cost us 6 figures"

We had been building a relationship with a PR company for a while when they came to us with a problem. Or what we thought was a problem (maybe it was a feature).

They have a client who wants a weekly media report for specific topics and bills that are getting attention in Washington D.C. The client wanted to track 100+ niche sources plus all the big media outlets. So the PR firm did what all PR firms love to do. They threw people at the problem. A small army checked all the news sources daily and put relevant news into a fancy summary for the client to read every week.

My business partner and I saw dollar signs.

We had generated PDFs before. We are experts in scraping, so we knew we could get all the data. Robots are WAY more efficient than low-level PR people, so we knew we could offer cost savings. Add GPT-3 to the mix, and we knew we could create the summary with minimal manual touches.

So... playing it cool... we said we'd build the system (we call it Pings) for $30k and credit that towards the first year of expenses.

You're probably seeing the first mistake now: we never discussed monthly costs or an ironed-out ownership structure. But I'll get to the ownership part later.

And just like that -- our team got to building. We were sprinting. Not in the agile sense of the term, but we were building fast.

1. We scraped all the sources and started populating our database.

2. We set up a simple UI in Bubble so the PR company could start to see and use the data (and give us feedback)

The PR company was delighted. It took their hundreds of boring hours to just a couple of hours a week. Yay #automation!

Between the scraping, database, and GPT-3 credits, our cost to run Pings was ~1k per month. We were targeting a price of $2k-5k a month to charge them. But we still didn't discuss pricing with them.

Next, we started building an actual UI. It was great, and we enjoyed using it. Once we began using Pings, we realized that the volume of news from the list of sources we were collecting was extremely high. 

All the topics the client cared about were really specific. Because of that, I knew we could filter out 98% of the news from all sources. But how? Keyword filtering is trash and doesn't work effectively.

GPT-3 to the rescue again -- this time with a clustering feature. 

1. We let users "star" and categorize "golden" articles. These are articles that measure against every report that comes into Pings.

2. Every time a new article comes in, we measure the relevancy distance across 2048 dimensions to each different topic (which is a cluster in AI/ML terms)

3. Save the relevancy score to the database

4. Auto-categorize the article if the relevancy score is under a certain threshold (for us— .15. 0 means the articles are duplicates). If the relevancy score is higher, we mark it as uncategorized.

5. We let users filter by a category, relevancy score minimum, and day. 

After a few days of saving and categorizing articles, Pings surfaced 99% of everything that'd go into the report in the first 50 articles. This added automation took the low-level PR person's time from hundreds of hours a month to just a few hours. 

However, we didn't want to stop there. Our goal was to generate the report in the same format as the PR company. We didn't want the PR company to have to write up a summary for every article they were adding to the report. How do we solve that? You guessed it, GPT-3 again. We trained a custom model to generate exactly what the PR company was already generating. The result: The PR company went from scrolling thousands of articles to looking closely at 50. When they wanted to add an article, they didn't need to come up with a witty caption.

Our team is super proud of what we built in Pings. The PR company was super happy with what we created. We had been collecting payments of 10k at certain milestones. The process from start to finish of building Pings took ~3 months.

Finally, we can't put off the payment conversation anymore. We tell the PR company to continue the service, it'll cost $2.5k a month, and the 30k will cover the first 12 months. Long silence from the PR company.

When we hear from them, we find out they were under the impression that the 30k they paid was for complete ownership of Pings...

In the software business, anyone who sees a 30k price tag on an app of this scale will either assume it's a scam or a subscription. In our case, it was a subscription. The PR company wouldn't wrap its head around SaaS (like GSuite, or any other software tools they use). Instead, they insisted on "owning" the software.

Long story short, we refunded the PR company their 30k and moved on (pivoted).

Now, Pings creates political news briefs. 

One problem -- neither my business partner nor I want to be in the political news briefing business. We could quickly pivot again and do media monitoring for Fortune 5000 companies. Or we could target PR companies. Seems like a simple sale. They get to outsource their media monitoring and digest capability to us. In return, we wouldn't have to make as many outbound sales calls.

But I don't think we will. Because we don't want to be in the media monitoring business. The only reason we built this in the first place was that we were excited to help these people we had built a relationship with. The money and value of a SaaS app w/ 30k in ARR was just a nice bonus.

I took 2 lessons away from this experience, and I'm sharing them so you hopefully don't make the same mistakes:

1. Don't get into something just for the money (if you don't need the money). It's not worth it, and situations can change fast.

2. Overcommunicate instead of undercommunicate regarding business. My default is undercommunication because I feel like overcommunication gets annoying. But the tail risk of a colossal misunderstanding is too high to not be a little bit annoying sometimes.

I'm not sure where we'll go with Pings. We have too much going on to do what we really need to do to get MRR & sell the biz. I send a drone newsletter to a small list of subscribers powered by Pings.

Maybe just chalk it up to lessons learned...