AI automation for businesses
There is a lot of noise in this area and very little of it survives contact with a real business.
We build automation where it removes genuine repetitive work and where a person remains responsible for the result. We are going to be specific about what we have actually put into production, because the distance between a demo and something that runs reliably every day is where most of these projects fail.
Our position
The test we apply is straightforward: is a human still doing this because their judgement genuinely matters or because nobody built the alternative. Where the answer is judgement, a person keeps doing it. Where the answer is that nobody built it, that is worth automating.
Anything customer facing or anything that spends money keeps a person in the loop. Not as a formality, as an actual approval step that things wait behind.
What we have actually built and run
Automated campaign building
We build Meta advertising campaigns through our own system rather than by hand. We set the specification, the system builds it and a person reviews and approves before anything goes live. For a multi-location client this turns a week of repetitive building into a morning and removes the copy and paste errors that come with doing the same thing twelve times. See Six by Nico.
Automated email campaign building
The same approach for email production. The build is automated. The approval is not.
A customer enquiry agent on WhatsApp
Running live for a client, answering common enquiries, handling questions about services and pricing and passing anything it should not handle to a person. It works from a locked set of approved information rather than making things up and it is explicitly prevented from inventing prices. We have had to harden it against people trying to manipulate it, which is a real and ongoing part of running something like this.
Cross-channel reporting automation
Pulling Meta, Google, Klaviyo, Google Analytics and booking or ecommerce data into one live view, continuously, instead of someone assembling a spreadsheet each month. This is what Holisticly does.
Internal operations automation
We run a significant amount of our own agency on systems we built, which is where most of this capability came from. We built it for ourselves first because that is the honest way to find out whether something works.
How we build it
Start with the process, not the technology.
What actually happens now, who does it, how long it takes, how often it goes wrong.
Work out whether automation is the right answer.
Sometimes the process should just be deleted or simplified. Automating a bad process makes it fail faster.
Build the smallest useful version.
Running against real work, with a person checking every output.
Widen it only once it is reliable.
Reliability is measured over weeks of real use, not demonstrated once.
Keep the human checkpoint.
Wherever the output reaches a customer, spends money or cannot easily be undone.
Who is responsible
You are and so are we. Automation does not transfer accountability to software. In practice that means:
- A named person approves anything customer facing
- Outputs are logged so you can see what happened and when
- There is a way to stop it
- Someone is responsible for noticing when it stops working, which is where most automation projects quietly die
Where this goes wrong
Models make things up confidently.
In a customer facing context that is a commercial and sometimes a legal problem, which is why our customer facing agents work from a fixed approved source rather than generating answers freely.
People will try to manipulate anything public facing.
We have dealt with this directly and it requires ongoing attention rather than a one off fix.
Automation breaks silently.
An integration changes, a credential expires and it stops working without announcing it. Monitoring is not optional.
An automated bad process is still a bad process.
Just faster and harder to see.
What we will not do
- Put an unsupervised AI in front of your customers making commitments on your behalf
- Automate something where a mistake is expensive and hard to reverse
- Tell you a prototype is a product
- Build something requiring ongoing maintenance without being clear about who maintains it and what that costs
What is roadmap, not service
In production, running daily
- Campaign building. Email building. Reporting automation. The customer enquiry agent. A large amount of our own internal operations.
In development or internal only
- Deeper decision support. Automated content workflows. Broader cross-client intelligence. Several agent workflows we use ourselves but would not yet sell as a client service.
If something is on the second list and you ask for it, we will tell you it is on the second list.
Who this suits
Suits you if
- You have a genuinely repetitive high volume process. Staff time is going on assembling information rather than acting on it. You have a real appetite to change a process, because the technology is rarely the hard part.
Does not suit you if
- You are looking for AI because it sounds modern, without a specific process in mind. We will ask what you want to stop doing and if there is no clear answer we will say so rather than invent a project.
Common questions
Tell us what your team should not be doing
Answer a few short questions so we can help you decide whether there is a useful project here.