AI and automation
Give your team back the hours they spend on busywork.
Triage, drafting, extraction and reporting built into software you own. A person reviews what matters before it goes out.
Digital Potter builds AI into your own systems, not a chatbot on top of them. Your data stays yours.
- A personreviews what matters
- You decidewhat every AI feature can see
- 100%of the code and prompts in your name
Sound familiar?
Your team reads, sorts and retypes the same things all day.
Requests arrive by email, form and phone, and someone sorts them by hand. Invoices and forms get retyped into another system. Replies follow the same pattern a hundred times a week. A chatbot demo looked impressive, but it never touched the real work.
Inboxes sorted by hand every morning
Documents retyped from PDF into a system
The same reply written again and again
AI pilots that never reached daily use
Where it saves time
Automation aimed at real work, inside your own systems.
Intake triage
Every request read, labelled and routed to the right person.
Drafted replies for review
A first draft waits for your team to edit and send.
Document extraction
Invoices, forms and contracts turned into clean records.
Reports written for you
Weekly numbers pulled together and summarized in plain language.
Review screens for the decisions that matter
Cost limits and usage reports from day one, with code and prompts in your name
How we build it
Tested like the rest of your software, because it is.
Every AI feature starts with real examples from your work. We measure accuracy against them before launch, and again after every change. Anything uncertain goes to a person instead of out the door. Senior engineers review every change before it ships.
A test set built from your real cases
Accuracy measured before and after every change
Low-confidence results sent to a person
Spend limits and alerts on every model call
Every result logged so you can check it later
How it runs
From one slow task to one that runs itself.
The usual shape of an automation project. Yours may be shorter.
Find the hours
Week one
We watch the work and count where the time goes. You get a ranked list of what to automate.
Agree on the first automation
Week two
One task, a fixed scope and a fixed number. Plus the accuracy it must reach.
Build and test on your examples
The middle weeks
Your team reviews the results in staging every week. We tune it until it meets the bar.
Launch with a person in the loop
The final weeks
It starts with review on every result. Review steps back only where the numbers earn it.
Not sure where to start?
Start with a review or an embedded team.
Forward-deployed engineers
Forward-deployedSenior engineers inside your team, shipping one important system.
A straight answer
When AI is the wrong call.
You want a chatbot on the homepage
That rarely saves anyone time. We automate the work behind the site instead.
There is no process to automate yet
If the work changes every week, write the process down first. Then automate it.
There is no data to learn from
AI needs real examples of the work. Without them, start by collecting them.
Investment
Automation projects from $20k.
You get a fixed number for the first automation after the call.
recommended
AI and automation build
from $20k
One automation, built into your systems, tested and launched with review.
- Fixed scope and price
- Accuracy agreed up front
- Cost limits from day one
- Code and prompts in your name
Questions
What owners ask before trusting AI with real work.
Where does AI actually save time?
In work that is repetitive and text-heavy. Sorting requests, drafting replies, reading documents and writing summaries. It rarely helps with decisions nobody can explain.
What happens to our data?
Your records stay in your database and your cloud account. An AI feature only sees the data its task needs, and you approve that list. We choose model providers with you, under terms your team has read.
Which AI models do you use?
The one that fits the task and the budget. We test a few on your real examples and pick on accuracy and cost. You can switch later without a rebuild.
What does it cost to run?
Usually a monthly model bill that grows with use. We estimate it before the build. Spend limits and alerts keep it from surprising you.
What happens when it gets something wrong?
A person catches it. Uncertain results go to review instead of out the door. Every result is logged, and mistakes become new test cases.
Can it connect to our existing systems?
Yes, when they have an API or an export. Email, forms, storage and most business tools do. The automation reads from them and writes back.
Next step
Bring the task your team is tired of doing by hand.
In thirty minutes you learn whether AI can take it on and how we would test it. And roughly what it costs.