Uplify /
UA RU EN
> contact

AI-first means the agency process architecture is rebuilt around AI: AI prepares drafts of reports, RCA and action plans; a human makes the decision through approve-flow with a pre-state snapshot, audit log and rollback option. At UPLIFY four AI agents (Lead Hunter, Merchant Analyst, SLA Watcher, Content Editor) work under this control 24/7.

AI-firstUPLIFY OSAI agentsapprove-flowaudit logLead HunterMerchant AnalystSLA WatcherContent Editorstate machinerollback
● UPLIFY / AI-first ● methodology ● UA / EU / WW

AI-first is an
architectural decision,
not a buzzword.

"We use ChatGPT" is not AI-first. AI-first is when the process architecture itself is rebuilt around AI: data syncs itself, anomalies are caught automatically, AI drafts the analysis, a human makes the decision — through approve-flow with full audit log.

see the commits → how it looks in OS
4 AI agents on staff · 27 sync tasks before 08:00 · 23 alert types · 0 AI actions without approval
/ what we mean

AI-first is about the process, not the tool.

Tools are easy to swap. ChatGPT, Claude, Gemini — any agency can plug them in over a weekend. That does not make the agency AI-first. AI-first means the process architecture itself is rebuilt around AI: what the machine does, what stays with the manager, where the human sits in the decision loop.

At UPLIFY, AI-first means four rules:

  • → AI prepares drafts — reports, RCA, action plans, ad copy. The manager reviews and approves them.
  • → Humans make decisions — about bids, budgets, strategy, creatives. AI suggests; humans decide.
  • → Approve-flow — every important AI action goes through a state machine with a pre-state snapshot, audit log and rollback.
  • → No black boxes — the client sees what the AI did, when and why, in the client portal.
/ philosophy · architecture

why AI-first, not "an agency that uses ChatGPT"?

AI-first isn't about a tool. It's about process architecture: what the system does vs what stays with the manager. Below — 4 commits already running inside our UPLIFY OS. All of it in numbers.

d4f2a1 · sync.py 01 · data+1 −1
manual: each morning the manager logs into 7 systems and hand-merges data
uplify: 27 sync tasks done by 08:00 — Google Ads, Meta, GA4, Merchant, Clarity, GSC, PageSpeed, banks
client portal is ready before the manager opens the laptop
8b7e30 · alert_detector.py 02 · alerts+1 −1
manual: ROAS dip noticed Friday, when the report is built
uplify: 23 alert types — spend spike, conv drop, MC disapproved, hygiene, asset quality, GA4 outage… every day at 08:00
critical ones become an auto-task for the manager + an AI RCA draft in plain English
f91c55 · reports.py 03 · reports+1 −1
manual: copy-paste screenshots into PowerPoint Friday evening
uplify: weekly report draft at 09:30 Thursday — narrative + screenshots + findings. Manager reviews, sends
weekly automatic · monthly report on the 2nd at 09:00
a0e312 · ai_action.py 04 · control+1 −1
manual: AI proposes — "let the manager decide if it's right"
uplify: every AI action goes through a state machine approve→execute + pre-state snapshot + role-based access + audit log
AI drafts — yes. AI actions without oversight — no. rollback — when needed
$AI-first is an architectural decision, not a buzzword. The tool will change — we'll rebuild it. The process stays.
/ approve-flow

how every AI decision gets made.

01 · DRAFT
AI prepares the draft
RCA, action plan, report copy, creative variant.
02 · REVIEW
Manager reviews
In the client portal, with full context — sources, metrics, before-state.
03 · APPROVE
Human hits approve
Or edit → approve. Without approval the action does not run.
04 · EXECUTE
AI executes
With an audit log entry: actor, timestamp, before/after snapshot.
05 · ROLLBACK
Always reversible
The pre-state snapshot lets us roll back to the state before the AI action.
/ what AI does NOT do here

where approve-flow ends.

Unsupervised actions are the main worry whenever a client hears "AI is running my ads". So we drew a hard line on what AI deliberately does not do at UPLIFY:

  • — does not change bids in Google Ads / Meta without manager approval
  • — does not publish creatives or ad copy without moderation
  • — does not modify targeting, audiences or exclusions automatically
  • — does not move ad spend — that stays with the platform under approved strategy
  • — does not reply to clients on Telegram in our name without a human read
  • — does not train on your data — we use APIs, never fine-tuning datasets
/ AI vs automation vs scripts

where the cron-job ends and AI begins.

automation

if-this-then-that

Hard rules: "if ROAS < X → pause". Works until the condition shifts.

  • fast
  • predictable
  • blind to context
AI-first

context + narrative

AI sees the full context — why ROAS dropped, what's happening in Merchant, whether it's seasonal — and proposes a plan.

  • contextual
  • explainable
  • approve-flow gated
plugins / scripts

one-off tasks

"Generate a description", "find keywords". Useful, but not integrated into the system.

  • spot-only
  • no memory
  • not tied to data
/ faq

frequently asked questions about AI-first.

01Does AI make decisions instead of the manager?

No. AI prepares the drafts — reports, RCA, action plans, ad copy. The human makes the call via approve-flow. That is the core architecture of UPLIFY.

02What happens if the AI makes a mistake in a draft?

The manager reviews — and does not approve. Mistakes surface immediately because we show contextual data: which metrics, dates and previous values the AI based its conclusion on. If the AI did execute something before approval, the pre-state snapshot allows a 1-click rollback.

03Will my data train your models?

No. We use provider APIs (OpenAI, Anthropic, Google) in no-data-retention mode. Provider contracts confirm: prompts and responses are not used for model training.

04How is this different from a regular agency that uses ChatGPT?

Regular agency with ChatGPT: a manager opens ChatGPT once a week, copies context in, gets some text, pastes it into a report. UPLIFY AI-first: AI is wired to live data 24/7, drafts on its own, runs through the state machine, owns four roles (Lead Hunter, Merchant Analyst, SLA Watcher, Content Editor) with KPIs. The difference is "we use a tool" vs. "we rebuilt the process".

05Can I see what the AI is doing?

Yes. The client portal has a section listing every AI action in the last 24 hours: timestamp, which agent, which metrics it used, which drafts it produced, what the manager approved or rejected.

consult

see AI-first in code.

We will show how approve-flow and the audit log work in production — using your own account after a free audit.

see the OS → send a brief