Denha NexusFeature · Approvals

Approvals: automation proposes, a person decides, the system applies

This is the limit the product is built around. The system reads data, works things out and proposes changes, but it does not apply them itself. Anything that moves money waits in the approval queue with its reason and its expected effect. Confirm it and it is applied, read back from the account, and written to the log with your name.

14 days on any plan, no card needed to start. After the click: a three-field form and the workspace.

What lands in the queue

Where from
Recommendations, background modules, the assistant in chat and in Telegram, automation rules
What
Budget, bid, campaign or ad group status, placements, a campaign launch, a publication
With what
The reason, the expected effect, the amount of data behind it, the platform and the account
Who decides
A person with the right to confirm; how much the system may do alone is set in the Autopilot section
After
Applied through the API, read back from the account, written to the log
Where
The Approvals section, the Today summary, and the client portal for actions the client signs off

The same count of waiting actions used to appear in two places and disagree with itself. Now the queue has one counter, and it is where you expect it.

01What it looks like

Approvals in ADS Beast: an action with a reason, confirmed and checked

Screen recording: an action in the queue with its reason and expected effect, a person confirming it, the change applied through the API, and the result read back from the account.

A screen recording of the product in the ADS Beast demo workspace. The clients and campaigns on screen exist for the demonstration; no real advertiser stands behind them.
02What it does

What the approval queue does

01

Every action arrives with a reason

An action in the queue is not "raise the budget". It is "raise the budget on this campaign, on this platform, because over this period and on this much data the cost per lead is under target". Without a reason and an amount of data, an action does not reach the queue at all.

  • The reason and the expected effect behind every action
  • The platform, the account, the period and the amount in dollars
  • A mark for whether the action moves money
02

Confirming, rejecting, and the trail it leaves

You can confirm or reject from the section, from the summary or from Telegram. A rejection is recorded too: the system remembers which proposals a person turns down and does not bring the same one back twice.

  • Confirm and reject buttons with a comment
  • Protection against repeating the same recommendation
  • An action log: who, what, when
03

Applied, then checked that it happened

After confirmation the system applies the action through the platform API and re-reads the object from the account. If the platform refused, the refusal text is shown inside the action. Here "applied" means "read back", not "sent".

  • Confirmation by reading back from the account
  • The platform response text when it refuses
  • The action status: waiting, applied, rejected, failed
04

How much the system may do alone is set out loud

The Autopilot section shows how far the system is allowed to act on its own: what it may only propose, and what it may apply after confirmation. The assistant in Telegram follows the same rules: any tool that moves money leads to the queue.

  • Autonomy levels by action type
  • One set of rules for chat, Telegram and background modules
  • Assistant tools with no entry in the registry are closed by default
03How it works

The path of a single action

From recommendation to log entry there are four steps, and a person takes only one of them.

  1. 01

    Proposal

    A module, a recommendation or the assistant forms an action with its reason, effect and amount of data.

  2. 02

    Waiting

    The action sits in the queue; the counter shows in the summary and a notification goes out on the chosen channels.

  3. 03

    Decision

    A person confirms or rejects it, from the section, the summary or Telegram.

  4. 04

    Applied and checked

    Through the platform API, then read back from the account and written to the log.

04How it fits together

Where the data comes from and what you get out of it.

Ad accounts hand over their statistics through your own OAuth apps. The system puts them into one picture, agents look for what to change, and only what you approve gets changed.

Platforms
  • Google AdsOAuth
  • MetaOAuth
  • TikTokOAuth
  • LinkedInOAuth
  • Microsoft AdvertisingOAuth
  • Windsor.aiAPI key
ADS Beast
Combines and checks
  • sync every 15 minutes
  • conversion to USD at the day rate
  • reconciliation with the native account
  • 90 days of history
What you get
  1. One spend summaryEvery account in one currency and one date window.
  2. Tasks from agentsWhat to change, where and why, linked to the row in the account.
  3. A person decidesBudgets, bids and statuses change only after approval.

Early access

See your own ad accounts the way the diagram shows them.

We are opening access to agencies in turn. Leave a work email and we will write when your turn comes, then help you connect the first account.

We use the email only to reply about access. Privacy policy

05Figures from the code
8
channels run through one queue
5 min
between notifications about waiting actions
4
notification channels: email, Slack, Telegram, browser
06What it does not do

What the approval queue does not do

  • It does not apply actions without confirmation, however obvious or urgent the recommendation looks.
  • It does not accept actions that carry no reason and no amount of data.
  • It does not count "sent" as "applied" - only what has been read back from the account.
  • It does not bring the same recommendation back after you reject it.
07Questions and answers

Questions and answers

Can the system raise a budget on its own when cost per lead looks good?

No. It will propose that as an action with its reason and expected effect. It raises the budget only after a person confirms, and then re-reads the campaign from the account to be sure the change stuck.

Who can confirm actions?

Users whose role in the organisation allows it. Actions the client signs off, a launch against a proposal for instance, are moved into the client portal.

What happens if the platform refuses a change?

The action gets the status failed, with the platform response text. That tells a platform limit apart from a fault in the system and points at what to change.

Does this work in Telegram?

Yes. The assistant in Telegram shows waiting actions and takes a confirmation or a rejection. It does not run money-moving tools directly - only through the queue.

Connect an account and get a recommendation with a reason, not an "apply all" button

The first 14 days are free on any plan and you do not need a card to start. The approval queue is always on and cannot be switched off.

14 days on any plan, no card needed to start. After the click: a three-field form and the workspace.