Is AI Automation the Key to Business Survival?

[BGARK]

I write this out of genuine curiosity: why isn't AI automation at the absolute top of every business owner's priority list?

I'm not a genius by any stretch. I'm in my 50s, started from zero, and am extremely persistent with tasks I attempt. I learn as I go along, make lots of mistakes, and eventually something sticks. This approach has somehow kept me selling fully British manufactured products for over 20 years, with growing sales into Europe and the USA.

After two decades, I now have a decent business and a team that runs it. This gives me the freedom to play around doing whatever interests me while trying to remain productive. I've discovered that AI automation is where I can make the most impact.

In the background, I spend most of my days building AI workflows and working with developers on systems to reduce or soon eliminate basic administrative tasks entirely. We've implemented automation for emails, bookkeeping, customer service questions, and much more.

The results have been transformative. Tasks that once required expensive professionals (solicitors, accountants) can now be handled through well-designed AI systems. I've even created a health workflow for analysing my blood tests that's probably kept me alive!

My suggestion: please learn this technology before it's too late. At minimum, follow key people doing YouTube podcasts and test all of the latest LLM models for basic conversational reasoning.

I'm curious - how many other business owners here are actively implementing AI automation? If you're not exploring this yet, what's holding you back?

Ideally, I'd like to help others or share knowledge, though I'm not naturally a social person. If there's interest, I'm happy to share more specifics about our automation journey.

[Venisonpie]

BGARK said:

I write this out of genuine curiosity: why isn't AI automation at the absolute top of every business owner's priority list?

I'm not a genius by any stretch. I'm in my 50s, started from zero, and am extremely persistent with tasks I attempt. I learn as I go along, make lots of mistakes, and eventually something sticks. This approach has somehow kept me selling fully British manufactured products for over 20 years, with growing sales into Europe and the USA.

After two decades, I now have a decent business and a team that runs it. This gives me the freedom to play around doing whatever interests me while trying to remain productive. I've discovered that AI automation is where I can make the most impact.

In the background, I spend most of my days building AI workflows and working with developers on systems to reduce or soon eliminate basic administrative tasks entirely. We've implemented automation for emails, bookkeeping, customer service questions, and much more.

The results have been transformative. Tasks that once required expensive professionals (solicitors, accountants) can now be handled through well-designed AI systems. I've even created a health workflow for analysing my blood tests that's probably kept me alive!

My suggestion: please learn this technology before it's too late. At minimum, follow key people doing YouTube podcasts and test all of the latest LLM models for basic conversational reasoning.

I'm curious - how many other business owners here are actively implementing AI automation? If you're not exploring this yet, what's holding you back?

Ideally, I'd like to help others or share knowledge, though I'm not naturally a social person. If there's interest, I'm happy to share more specifics about our automation journey.

I'd be very interested given a move from a safe career to a partner in a small retail business is iminent. As well as enabling tech to replace "admin tasks" as you've achieved I'm also interested in modular production that's scaleable across small independent units.

[BGARK]

Venisonpie said:

I'd be very interested given a move from a safe career to a partner in a small retail business is iminent. As well as enabling tech to replace "admin tasks" as you've achieved I'm also interested in modular production that's scaleable across small independent units.

Random example: https://youtu.be/9FuNtfsnRNo?si=2pj1kRj8_0kao1Df

Subscribe to all of the n8n or make.com workflow tutorials in YouTube, basic accounts to test are practically free, a few tutorials will handle running most your common tasks.

Make sure your using the PAID versions of Claude, OpenAI, and as of this week, GROK3 is now my go to for live search and some deep search tasks. I also have the $200 a month Pro plan for GPT, its out of this world for heavy research tasks. I was sceptical but its pretty nuts, Grok3 will probably outdo GPT very soon at a fraction of the cost.

I am also coding by copy/paste. take a look at www.bolt.new for example, simply ask it to build you an app or a game.

[Simpo Two]

Excuse my ignorance but what's the difference between 'writing software' and 'building AI workflow'?

[BGARK]

Simpo Two said:

Excuse my ignorance but what's the difference between 'writing software' and 'building AI workflow'?

Probably not a lot under the bonnet, however being able to drag and drop nodes on a visual workflow (like a flowchart), is easy for pretty much anyone to handle, hence why these "no code" solutions are growing so rapidly.

[Simpo Two]

BGARK said:

Simpo Two said:

Excuse my ignorance but what's the difference between 'writing software' and 'building AI workflow'?

Probably not a lot under the bonnet, however being able to drag and drop nodes on a visual workflow (like a flowchart), is easy for pretty much anyone to handle, hence why these "no code" solutions are growing so rapidly.

I see, so instead of writing software to create an AI 'machine', it's the other way round - using AI to write software, is that it?

[Newc]

I'm interested in what you have done exactly, eg in 'automation for bookkeeping'.

Is this something where you had a screen-based manual process before, like doc A comes in, gets copy-pasted to doc B in a slightly different format, and then sent on ?

Or is it a fuzzier thing where you've got some software making its own assessments of potential ageing debtor problems, and independently starts to chase the customer ?

[BGARK]

Simpo Two said:

I see, so instead of writing software to create an AI 'machine', it's the other way round - using AI to write software, is that it?

Not exactly - it's using visual workflow tools (like n8n or Make.com) that let you connect different services together without coding. These tools now integrate with AI services like GPT or Claude, so you can build complex automation systems visually by connecting 'nodes' rather than writing code.

Think of it like building with Lego blocks. The AI provides the intelligence within these workflows - answering questions, making decisions, or generating content - while the workflow tool handles the connections between systems. Plus you control what each node does with a prompt that the LLM uses for instructions.

[skyebear]

BGARK, this is very useful and interesting, thanks for taking the time to share and I will look at those sites you mention.

[BGARK]

Newc said:

I'm interested in what you have done exactly, eg in 'automation for bookkeeping'.

Is this something where you had a screen-based manual process before, like doc A comes in, gets copy-pasted to doc B in a slightly different format, and then sent on ?

Or is it a fuzzier thing where you've got some software making its own assessments of potential ageing debtor problems, and independently starts to chase the customer ?

We haven't actually completed this yet - we have all these parts in progress for testing soon. Small sections work, but we're still learning. When fully implemented, it will handle everything I described.

This example below is primarily focused on handling supplier invoices and the messy paperwork coming into a business - the accounts payable side rather than customer billing.

Think of it like a set of bullet points - if you write out every single step that you take. This is what we're testing, one step at a time:

1. An email arrives (or the AI system monitors the inbox)
2. AI reads the contents and attachments
3. It scans the text and attachment to extract key info - invoice amounts, PO number, description, tax info, etc.
4. This is automated via OCR or other means
5. The system sends/adds this into a Google Sheet or Airtable
6. An AI model examines and compares to historical info, or looks at the original source data (e.g., the PO you made) - you prompt it in how you want it to operate
7. The data is saved to a vector database for unlimited future retrieval and comparisons
8. This syncs via API into our accounting system (like Xero)
9. A human checks the final reconciliation process

Everything is automated up to the last point.

One of the key goals is to have the system Red Flag unusual patterns for small and recurring costs, such as monthly subscriptions and all those small expenses that just get paid and we forget about. These can add up significantly over time, and sometimes increase without notice. The system would point out things we might need to look closer at, like "This software subscription has increased 15% since last quarter" or "You're paying for three similar services that might be redundant."

Note that some of this is just a sequence of events that aren't intelligent - it's like you scanning a piece of paper. But the results of that scan then digitise and follow a bunch of steps. Intelligence is only needed at a couple of steps, and you control what happens, like a conductor in an orchestra.

The end goal is both the mundane data entry parts (invoices being processed, categorised and entered automatically) and more sophisticated functions:

1. Flagging unusual transactions by comparing to historical patterns
2. Prioritising which invoices to chase based on customer payment history and current cashflow needs
3. Drafting customised payment reminder emails that maintain the right tone based on our relationship with that customer
4. For financial reporting, pulling data from multiple sources and generating insights in plain English rather than just numbers

The key breakthrough is combining traditional automation (API connections between systems) with AI that can understand context and make judgement calls that previously required human intervention.

[akirk]

sounds very interesting - do you have any stats yet on accuracy?
For me the biggest concern would be that a lack of accuracy creates risk - and one of the biggest issues I currently see with AI is the assumption that computer= 100% accurate.

from the sound of what you are doing this is not really AI - more logic engines / rules systems (it is amazing what has rebranded as AI) but still a very good area to be looking at. We do a lot of business process systems development - often for small businesses for whom this approach should be very appealing, the biggest issues we find are:
- systems are not documented accurately
- systems don't do what they claim
- systems don't talk to each other
- systems make assumptions

the advantage of the approach you discuss is the lack of technical knowledge you need
the disadvantage is perhaps the lack of technical knowledge to see whether it is working or not...

will be interested to see more about your experiences

[BGARK]

akirk said:

sounds very interesting - do you have any stats yet on accuracy?
For me the biggest concern would be that a lack of accuracy creates risk - and one of the biggest issues I currently see with AI is the assumption that computer= 100% accurate.

from the sound of what you are doing this is not really AI - more logic engines / rules systems (it is amazing what has rebranded as AI) but still a very good area to be looking at. We do a lot of business process systems development - often for small businesses for whom this approach should be very appealing, the biggest issues we find are:
- systems are not documented accurately
- systems don't do what they claim
- systems don't talk to each other
- systems make assumptions

the advantage of the approach you discuss is the lack of technical knowledge you need
the disadvantage is perhaps the lack of technical knowledge to see whether it is working or not...

will be interested to see more about your experiences

One of the features is HITL (human in the loop) it can message you questions to get answers, or when it gets stuck it stops and waits. If built correctly its far more accurate and reliable that any human can be.

When you go an look in a filing cabinet for something, are you using intelligence?

[Simpo Two]

BGARK said:

When you go an look in a filing cabinet for something, are you using intelligence?

Eyeballs and memory I'd say. Not much thinking required.

Thanks for the earlier explanation BTW.

[Nightmare]

Really interesting thread - thanks!

We’re using AI extensively in the business but not yet for this. Got our weekly AI meeting (in VR lol) today so will bring this up. Automating the accounting side would definitely be handy!

One concern we’ve had with a fair few of the AI tools is data ownership/privacy etc. (Eg. If you create a video editor using HeyJen, any videos you then create can be used by them for whatever they like which could be interesting if you’re actually using it for work stuff. Also the ability to use chatGPT to find out what others are using it for in a specific field has led to some interesting competitor intelligence in Pharma).

Any concerns there?

[akirk]

BGARK said:

One of the features is HITL (human in the loop) it can message you questions to get answers, or when it gets stuck it stops and waits. If built correctly its far more accurate and reliable that any human can be.

When you go an look in a filing cabinet for something, are you using intelligence?

Yes - huge amounts of little bits of intelligence, and as we all know there is no intelligence in AI
To get the same, the process system has to think ahead for all the scenarios that might happen and rarely is that thinking complete or accurate...
For example, looking for stock issues in a warehouse - a human might notice things visually / by smell / by sound / by feel / etc. To put all of that into an AI process is tricky, especially as some of it may be unquantifiable... not everything human can be replaced by computers - and sure, I appreciate that this is about replacing the mundane, but all we have seen in the past is that this means driving logic down to a lowest common denominator, and when using process systems plugged together in this way it often means changing the business processes to match the software logic - in some cases that can improve the business, in others damage it.

c. 40 yrs of coding has taught me that:
- coders make assumptions
- users assume that code is perfect and infallible
- there is no such thing as a piece of code that a user can't find a way to break or use differently

with all that you are plugging together - you are making assumptions about its accuracy now and into the future - yet you are taking on risk that e.g. one dev. in one of the companies makes a mistake and that enters your logic process...

There are so many variables in business processes that I for one still retain cynicism around likely success - but I am open-minded and intrigued to see examples where it does work - just not seen any yet which are total game-changers...

will follow with interest...

[BGARK]

Simpo Two said:

BGARK said:

When you go an look in a filing cabinet for something, are you using intelligence?

Eyeballs and memory I'd say. Not much thinking required.

Thanks for the earlier explanation BTW.

Very true, one way to think of this is how would you give instructions to a new member of staff, if you simply said go to the filing cabinet and use your eyeballs, they would fail without clear instructions.

Your effort is all about creating a process, for either a human or the workflow to follow.

Write it all down, you need it anyway, and it helps improve areas you might take for granted. Even the smallest details are needed.

[BGARK]

Nightmare said:

Really interesting thread - thanks!

We’re using AI extensively in the business but not yet for this. Got our weekly AI meeting (in VR lol) today so will bring this up. Automating the accounting side would definitely be handy!

One concern we’ve had with a fair few of the AI tools is data ownership/privacy etc. (Eg. If you create a video editor using HeyJen, any videos you then create can be used by them for whatever they like which could be interesting if you’re actually using it for work stuff. Also the ability to use chatGPT to find out what others are using it for in a specific field has led to some interesting competitor intelligence in Pharma).

Any concerns there?

AI just just a tool, you might upload or attach a large document and ask GPT, to summarise. I do with legal jargon.

This is doing that, except it automatically uploads the document, asks what's going on and send you a summary (one simple example) and it could do that hundreds of times an hour whilst you play golf. eg send me a one sentence summary of emails received to my WhatsApp, organise in order of urgency, etc.

For private data, run an LLM on your own server. I am not an expert but lots of tutorials online where people are discussing use with private client info.

[akirk]

BGARK said:

Simpo Two said:

BGARK said:

When you go an look in a filing cabinet for something, are you using intelligence?

Eyeballs and memory I'd say. Not much thinking required.

Thanks for the earlier explanation BTW.

Very true, one way to think of this is how would you give instructions to a new member of staff, if you simply said go to the filing cabinet and use your eyeballs, they would fail without clear instructions.

Your effort is all about creating a process, for either a human or the workflow to follow.

Write it all down, you need it anyway, and it helps improve areas you might take for granted. Even the smallest details are needed.

the difference though is the intelligence bit...
assuming some intelligence in the member of staff - if you told them to go to the filing cabinet and find the file ABC123 - they might also come back with the right file even though there was a typo on the reference number / a bit is torn off / it was filed incorrectly under ABD / the actual reference is 123ABC and you got it wrong / etc.

that approach is built into most humans - but in data terms you might have to cater for a large number of scenarios - then understanding whether the alternative scenario is correct etc.

so if AI found 123ABC but not ABC123 it would be discarded - would it consider checking all other records to see if they are numeric-alpha rather than the the entered alpha-numeric? The human in flicking through the filing cabinet might notice the visual pattern and realise the mistake...

[BGARK]

akirk said:

the difference though is the intelligence bit...
assuming some intelligence in the member of staff - if you told them to go to the filing cabinet and find the file ABC123 - they might also come back with the right file even though there was a typo on the reference number / a bit is torn off / it was filed incorrectly under ABD / the actual reference is 123ABC and you got it wrong / etc.

that approach is built into most humans - but in data terms you might have to cater for a large number of scenarios - then understanding whether the alternative scenario is correct etc.

so if AI found 123ABC but not ABC123 it would be discarded - would it consider checking all other records to see if they are numeric-alpha rather than the the entered alpha-numeric? The human in flicking through the filing cabinet might notice the visual pattern and realise the mistake...

I had a similar issue with product names, we use some unique and odd names that we made up, and if we didn't spell correctly it wouldn't find them, so we wrote an LLM prompt that simple said, look for similarities or patterns in text or spellings and make some assumptions that the humans asking might get things wrong, using your example, it would then work.

I am not really trying to discuss "what is intelligence", maybe one for a new topic?

[Frimley111R]

I use ChatGPT and ImageFX but beyond that there is a lot of chat about AI apps etc but I can't really understand how I would apply it to marketing for instance, and I'm sceptical about how it can magically generate leads for me. I've seen examples of AI comms on LinkedIn posts and they all highlight how poor AI outreach is, although garbage in - garbage out. I'm also not sure how it integrates with existing software.

I think companies don't always know where to look for AI tools and how to use them. It's fascinating but it seems like a big commitment or time to understand it all on top of doing everything else. Perhaps I'm just looking in the wrong places...