[shash7]

Can relate, I've been in a similar boat running a small B2B Saas over the last 2 years. It does get easier over time.

I've learnt a few tricks for managing early stage pain points.

- You need to develop a polite but curt tone of voice for customer support.

- Once your core product is built, its worthwhile spending some time automating the heck out of everything. This will save a TON of time in the near future.

- Invest in good docs, even if you're not running a api saas. Good docs + consistent ux + rock solid support will solve most of your support issues.

I think a lot of literature around running a online biz has been boiled down to rather basic advice and its hard to find anything solid in this area. I've been running a small blog where I document these issues(operational.co) if anyone wants to check it out.

[duxup]

>You need to develop a polite but curt tone of voice for customer support.

And very focused responses in terms of action items.

You might think of 3 things to say, check, but sadly 90% of the people you respond to with a list will behave like they read just one of them. Sadly this also leads to dragging things out for everyone who can handle more than one thing at a time :(

[Vegenoid]

Yep. All the time when I worked in IT:

Me: Please try these 3 things and let me know how it goes: (list of 3 things with instructions)

Them: I tried (thing #1) and it didn’t work.

Me: Thank you, please try these 2 things and let me know how it goes

Them: I tried (thing #2) and it didn’t work.

Me: Thank you, please try this thing and let me know how it goes

Them: (no response)

Me: Just checking in to see if this is resolved?

Them: (no response)

Me: I’m closing this ticket as I haven’t heard back, let me know if this is still a problem and I can reopen it

Them: Don’t close the ticket, I’m still having this issue

Me: No problem, can you try this thing and let me know how it goes?

Them: (no response)

[wruza]

I’m observing this for many years and it feels like there are two types of people. Those who perceive lists as a whole and those who list.shuffle().pop(). Try asking your colleagues/etc three semi-related questions in one message and you’ll get only a partial answer in a significant number of cases. When confronted (constructively, much later) they usually get evasive and can’t explain. I could theorize it’s a learned behavior to avoid threaded pedantry or something, but my messages aren’t even long and other people share my frustration too (we communicate 10x faster and clearer between us). I’d write it off to attention capacity issues, but these people often aren’t even busy at all.

[suprjami]

Giving people a bullet point action plan at the end of an update helps with this.

## Action Plan

- Read my comment

- Try it

- Comment on your experience

[philsnow]

> You need to develop a polite but curt tone of voice for customer support

If this makes you uneasy, it can be easier if you sign initial support replies under another name.

  Hey, this is John, I'll be happy to help you with that.
  
  <Blah blah blah.>
  
  Let me know if that helps.
  --John

[lolinder]

I worked at a tiny startup with a single customer support plus sales person—she had an array of fake names that she'd use for different purposes (billing, sales, tech support, etc) to try to simulate a larger org. She actually kept them all straight and apparently it worked.

[eps]

   - Hey, this is John, I'll be happy to help you with that.
   + Hello Xxx,

No need to fake being happy (or sorry). Just provide an actual answer.

[Terr_]

Or in some cases, be "Ed Chambers" from Silicon Valley.

https://m.youtube.com/watch?v=y-CA2EW4Z_U

[whatshisface]

"I'm escalating you to tier seven..."

[RachelF]

"- Invest in good docs..."

In my experience, end users don't read the docs or FAQ's or help search - they send you a question.

[jkukul]

I think in the era of LLMs good docs/FAQ are of an even greater value.

You can write a support bot that sends a user's question + docs/FAQ to an LLM to automatically deal with the basic questions and only involve a human in the loop once a question goes beyond what's in the docs.

[kadoban]

Even if users don't read them of their own volition, they're still valuable. You can copy/paste responses from them, link to them, or hell just use them to remember the answer yourself later without having to confirm it or even think about it.

[vonunov]

Run stats on support requests and make canned/auto replies for the common ones?

[thehappyfellow]

Do you have an example of “polite but curt” tone? I’m struggling to see what you mean.

[shash7]

Both xyproto and Gustomaximus have solid examples.

Here's more:

- Be direct, Hi, the xyz feature is available on the PRO plan. You can upgrade to the PRO plan at app.saas.com/billing

- Be brutal, Hi xyz, your card couldn't be charged for your Saas subscription, and hence your subscription has been deactivated. To reactivate, enter your card details app app.saas.com/billing

- Be honest, Hello xyz, thanks for the feature request. We'll put it in our wish list but can't guarantee it will make the cut.

- Be generous, Hey xyz, thanks for pointing that out. We have identified that as a bug and have pushed a fix for it. In the meanwhile, I've extended your trial by 7 days, on the house.

Couple of other tips:

- Dumb down your reply as much as possible. If you can't, throw your reply through chatgpt and make it dumb down.

- Unless a support issue is very basic, reply after a few minutes if you're near your computer. Usually users figure out things on their own if given some time.

- But don't allow issues to go stale. To really wow customer service, reply as humanely quick as possible, especially for existing customers.

- Make your support timelines clear somewhere in your product, eg: Our support will respond within max 48 hours, but most responses take 2-3 hours.

- Make your terms and privacy policy pages clear. People do read this. getharvest.com is a gold standard in this area.

[xyproto]

Thanks for reaching out. The issue you’ve described seems to be on your end. Please check your settings or consult our docs for further guidance. If the problem persists, feel free to get in touch.

[benatkin]

From the sound of it, the politeness is the shallow politeness that you can easily get with ChatGPT. The curtness is defending from the users expecting too much, which can include delaying handling issues before properly checking if they're real. I experienced this with Vercel and it probably makes economic sense for them. (BTW I really should cancel my Vercel account but haven't decided to take the time to migrate yet.) https://x.com/search?q=vercel%20benjaminatkin_&src=typed_que...

The reason it can be framed as curtness is because they're being curt about the expectations, and the real expectations are pretty low. "Sure, I can delay really addressing the issue for a couple weeks. You're only paying me 40 bucks a month, why would you expect more? The goal of responding within two days is just for a canned response." See, they were curt and didn't let me demand something more than I deserved, like being able to use the product I'm paying for in the next several days!

[blantonl]

"Check yourself before you wreck yourself"

[guzik]

> - Once your core product is built, its worthwhile spending some time automating the heck out of everything. This will save a TON of time in the near future.

Interesting. Anything you've automated successfully that you can share? I've heard so many times that you should hold off on automating too early because constant pivoting and refining can end up making you spend more time fixing the automation than actually doing the work itself, so I kinda paused on it. I can see how it would make a big impact on my marketing outreach, which I'm doing manually right now with not much success.

[remoquete]

Docs... And a technical writer that'll tend to them.

https://passo.uno/signs-need-tech-writer/

[withinboredom]

Spending some time to learn technical writing (if you want to bootstrap a saas) is worth its weight in gold (same with marketing, business admin, accounting).

[Loxicon]

> automating the heck out of everything

What kinda things did you automate?

[andrewljohnson]

Make sure not to apply this polite but curt tone to consumer apps.

[lostemptations5]

Wow! This is a great blog. Thanks for putting it out :)

[samstave]

Ive had tremendous success submitting files to AI and just having it produce a structured readme.md for the logic within the code, I'll do a thing where I say Give me a readme, include description of functions, logic, how its invoked, dependencies, monitoring and give me a mermaid and swim diagram of the workflow.

Its great.

Another thing I do with an AI helper is when I have it write out a function for me I have it write out the descriptor that can go in the readme for that function. I also have it write a header with version description path etc.

It was an experiment which started with the goal is to have all the code for a simple project with its associated readme functions loaded into a textai workflow and postgres DB and I can dynamically call the readme functions for everything by having a bot yank all the MD table for all the functions and just put together a real-time readme. As I add txtAI workflow scripts, they put their MDs into the DB, I try having it spit a JSON of its MD to a mongo?

The point is playing with ways to have the system self document as I/it develops.

Because one of the ways I have been using Claude to learn is through a F-ton of iterating on an initial thought and bird-walking through implementing a tool wrapped around it.

This way, as I iterate through many many version of a concept or script/function/api/crawler - it keeps a reminder readme for when I want to know what the heck was I thinking (I have had super awesome moments of brilliance, and then after a sleep or lengthy distraction - totally loose the Mode and have no idea what I was doing, or how I came up with BrilliantIdea(TM)

ADHD is a helluva drug :-(

[samstave]

EDIT: One thing I attempted was to have claude keep a running artifact for a readme while I worked out a problem - but that didnt work so well. Also, as it hallucinates, one of the first indicators that its losing context is that I keep a header section on artifacts it makes and a version and description ad path for the artifact...

when iterating and the header gets stripped in the response, the bot is taking a turn...

Sometimes ask it to spit out the full context and its understandings in a manner I can use to re-prompt itself for best efficiency, and it gives a nice summary of the concept I am thinking through, and I use that with vary degrees of satifying results.