↓Skip to main content

AI Products that Small Businesses Want

Table of Contents

These are a few boring small needs that businesses have. They won’t change the world, but LLMs could be used for some of these to make products that may save a little bit of time and make a little bit of money.

Faster review of candidates in ATS #

Some companies are using ChatGPT to help them review candidates but it’s a bit clunky.

They go into their ATS, open the applicant’s PDF resume, copy out the text into ChatGPT, get it summarized for key points, then copy that text back into the ATS for other team members to review.

That could be automated. This will probably be a feature of existing ATS companies like Lever, Ashby, and Greenhouse, but it could also be a small standalone ATS.

Could also imagine some kind of AI powered ATS: Talent outreach, better application form for applicants, better candidate review, better candidate communication / scheduling.

Content businesses want “ChatGPT for my content library” #

For businesses that have a lot of content, some of them want to add a ChatGPT style thing that allows their users to chat with that content.

I think there’s a risk that this is mostly a gimmick, and that end users won’t actually care and stick with it. That said, it’s still something that some businesses want to buy.

For now they should probably use an off the shelf tool like one of these.

ChatGPT for our data #

Small businesses (and big businesses) want something that’ll be a perfect internal knowledge base.

Instead of employees needing to spend time manually searching for info or asking other employees, they’d be able to ask the bot.

The key here, as with most things, is quality. It’s easy to make a not-so-good version of this. Making a great version with lots of integrations (build ingestors for one-off and ongoing content eg from Slack, maybe messages, Google docs, email, local files) and that works really well (clear citations, give an answer, make it easy for the user to see if the answer is correct by showing them the quotes it’s based on).

ChatGPT as a customer support bot #

As an assistant to existing support reps. GPT can have a great memory and be massively parallel. It hallucinates, so likely better as a tool to augment existing reps that allows them to support more customers faster, rather than a replacement.

Themes from these #

  1. Businesses only know they want something if they can imagine it. Many CEOs have used ChatGPT personally, so they can easily imagine how a ChatGPT for internal data or customer support would be helpful. There are other things that’d help their business that don’t map onto anything they’re familiar with, so for those, they won’t be able to imagine it until they see itl

  2. Businesses want things that help make them look good to customers (or peers), that save them time, or that save them money.

  3. Businesses are using ChatGPT, so the most promising use cases may be the things that they’re already doing manually with ChatGPT every single day but that kinda suck to need to do repetitively.

What are some limits to LLMs being used in more products? #

  1. Cost (GPT-4 is too expensive for many production use-cases)
  2. Speed (of GPT-4)
  3. Reliability (in terms of errors and timeouts)
  4. Reliability (in terms of consistently getting a useful result to the prompt)

For open-source LLMs, the main limit is that they are not yet high enough intelligence for many production use cases.

↑