Home / Our approach / How we use AI
◆ How we use AI
AI rides shotgun.
Golden Digital is in the driver seat.
Every night, our AI reads every number, ticket, and message in every account we manage. It learns what works and what doesn't for each client. Every morning, our team has access to these powerful tools to help your business grow.
Nothing goes to a client — not a report, not a recommendation, not a budget change — unless someone here has put eyes on it and approved it.
Daily recap21 Aug · 06:04
A home & garden brand
A sample read — the format is real.
Revenue · month to date$284,110
Meta return · 30 days3.9×
Email share of revenue24%
Conversion rate · 30 days2.4%
◆ Worth a human’s look
Email’s share of revenue has slipped three months running — 31% in May, 24% now.
Nicole has it in Thursday’s plan.
◆ Golden Digital · StaffNights
LUCILLE
The system that reads everything, every night.
RoleOperating system
Built byUs, in-house
ShiftNightly, from 02:00
CoversEvery account we manage
Reports toThe GD team
■ The night worker
Lucille isn’t a chatbot.
She’s the operating system we built to run the agency’s background work — one system, built by us rather than a subscription we resell, reading every account we manage every night.
There’s no app for you to log into and nothing new for you to learn. You will never talk to her. You’ll just notice that the people you do talk to walk into every call already knowing your numbers.
We gave her a name and a face for the same reason there’s a dog on the homepage: she’s genuinely part of how the work gets done here, and filing that under “automation” felt like the dishonest version.
■ One night, start to finish
Here’s a Tuesday.
One night across the whole book of business — sample numbers, real routine. Most nights are uneventful, and that’s the point — the work is the reading, not the reacting.
02:10
Machine
Pulls the numbers.
Shopify, Meta, Google, Klaviyo, GA4 — every connected platform, every account. Not a sample. All of them.
02:21
Machine
Reads them against last month, last quarter, and last year.
Not against yesterday — a single day tells you almost nothing, and chasing it is how agencies waste your money. What matters is the slope over weeks and months, which is invisible when you’re looking at one account on the day you happen to open it. Every account gets this, including the quiet ones. That’s the part a human team genuinely cannot do at this size, and the honest reason we built her.
02:44
Machine
Notices what’s drifting.
Not alarms — patterns. Six things across the whole book of business were worth a person’s attention this week, and every one of them had been building for a month or more.
Worth a look · week of 18 Aug
01Email’s share of revenue down three months running — 31% → 24%
02Ad spend pacing 22% over budget with nine days left
03Repeat-purchase rate down four months in a row
04No test has run on the product page in eight weeks
+2two more, lower priority
05:30
Machine
Writes it all up.
A read per account, with the trend lines attached. A brief for every call happening today — your numbers, your open items, what we said last time. First drafts of anything we asked her for last night. All of it internal. None of it has left the building yet.
08:20
Human
Nicole decides what any of it means.
This is the judgment call, and it’s the reason a person still has the job. A trend is only a problem in context — and the context is the thing Lucille doesn’t have.
WorkEmail. Three months of slippage isn’t a bad send — two flows haven’t been touched since last year. Scoping a rebuild.
WatchAd pacing. It’s a launch month. Over budget is the plan. Check again on the 1st.
DropRepeat rate. They moved subscriptions to a new platform in April. It’s a counting artefact, not a churn problem.
+3Two watched, one dropped.
She writes down why for each one — including the drops. That’s what stops the same question being asked again next month, and it’s where house rules come from.
09:04
Human
You hear about the one that mattered.
◆ The message you actually get
“Morning — your email share of revenue has been sliding since May, 31% down to 24%. It isn’t the campaigns, it’s two flows nobody has touched since last year. I’ve scoped a rebuild and put it in Thursday’s plan. Nothing on fire, but it’s the biggest single thing on the table this month.”
One message. One human. Already thought about.
■ The job description
The line she doesn’t cross.
✓ On her own
Pull your numbers from every platform, every night
Read them and spot what changed
Raise a flag when something looks wrong
Draft a report, an email, an ad, a call brief
Remember what we decided, and why
✕ Never without a person
Send you anything at all
Change an ad budget, or turn a campaign on or off
Publish, push, or change anything on your store
Make a recommendation you’ll act on
Decide anything
■ The question nobody else answers
What happens to your numbers.
Do you train AI on my data?
No. Your numbers are never used to train a model — ours, or anyone else’s. We pay for the tier where that’s contractually off.
Who can see it?
The people on your account. They’re named, with faces, on the About page — not a pool of contractors you never meet.
Where does it actually go?
Every processor we use is named in the privacy policy — the hosting, the database, the model provider. If we add one, that list changes first.
Can I have it deleted?
Yes. Ask, and it’s gone — and we’ll write back to confirm when it’s done.
■ Why it compounds
What one store learns, the next one gets.
Most agencies keep their experience in people’s heads, where it leaves when they do. Ours gets written down — every decision, what we expected, what actually happened.
When enough of those point the same way, one of us turns it into a house rule and signs it. Then it shows up in the next store’s plan automatically. That’s the whole trick, and it’s why year three with us doesn’t look like year one.
House rule№ 0042
Don’t scale a winning ad until the product page can take the traffic.
Written afterstores did it the other way around
SignedThe paid-ads team
In force onEvery ad account we run
So what does that buy you.
Answers in hours, not next week.
No “let me pull a report and get back to you.”
Every account, every day.
Not just the loudest one, or the biggest.
Advice backed by every store we run.
Not one person’s instinct on the day.
You don't have to keep up with any of this.
The field moves weekly. You have a store to run.