← All articles How it works

How the recommendations queue works: from proposal to decision

What a recommendation in InsightAdScale contains, the five kinds it covers, how a team reviews, accepts or dismisses each one, and why every decision is kept.

A dashboard tells you what happened. The harder question on a Monday morning is what to change, and who agreed to change it. InsightAdScale's Recommendations screen is a queue for exactly that: specific proposed changes, each with its reasoning, waiting for a person to say yes or no.

In short

Each recommendation names one entity (a campaign, audience or creative), the change, the reasoning, a confidence and the projected impact. The team works the active queue by priority, accepts or dismisses each item, and every decision lands in History with who made it and when.

What a recommendation contains

Every item in the queue carries the same fields, so a buyer can judge it in seconds and an analyst can audit it later:

FieldWhat it holdsExample (made-up data)
TypeOne of five kinds of change (below)Creative rotation
PriorityHigh, medium or lowHigh
EntityThe campaign, audience or creative it applies toSocial Stories: Flash Sale
TitleThe change, in one lineRotate fatigued social creative with a new UGC variant
RationaleWhy: the evidence behind itRunning 75 days, fatigue score 78/100
ConfidenceHow sure the proposer is, 0 to 100%91%
Projected impactThe expected effect, as named measuresEngagement, CTR, fatigue

The five types match the levers a media team actually pulls: budget shift (move money between campaigns or channels), audience refinement (narrow or split a segment), creative rotation (replace a worn variant), geo expansion (add markets) and bid adjustment (change caps or dayparts).

The Recommendations Engine screen with counts of active, accepted and dismissed recommendations and cards showing priority, type, rationale, confidence bar, projected impact and Accept and Dismiss buttons.
The active queue: six proposals for a made-up agency, filtered by priority, each with a confidence bar and its projected impact.

Where recommendations come from

A recommendation is a record, and it can be created by whoever has the evidence: an analyst who spotted an overlap problem, a bidding model your data science team runs, or a scheduled pipeline that writes proposals through InsightAdScale's API. Creating recommendations through the API is restricted to administrators, so the queue stays a list of things someone stands behind. Reviewing them is open to the wider team.

Why a queue and not auto-apply? Budget and bid changes have consequences that a model cannot see: a client commitment, a launch next week, a platform minimum. Keeping a person in the loop, and keeping the record of their decision, is the point of the screen.

Working the queue

The active tab counts what is pending, accepted and dismissed. Filter chips narrow the list to high, medium or low priority, and a search box finds proposals by name. Each card shows the evidence and the projected effect, and a tag links it to the campaign, audience or creative it is about, so you can open that record before deciding.

Two buttons decide it. Accept marks the recommendation accepted and stamps the time and the person. Dismiss does the same for a no. Either way it leaves the active queue and moves to History, newest first. Accepting is a decision, not an automatic change: the budget shift or bid change is then made in the ad platform or in the campaign record, by the person who owns it.

Writing a recommendation that gets decided

Whoever creates recommendations, through the API at /api/adscale/recommendations, the quality of the queue depends on the quality of each item. Only type and title are required; the rest is what makes an item decidable in a minute rather than parked for a week.

  • Name the entity. Fill in the entity type, its ID and its name, so the card's tag points at the exact campaign, audience or creative, not "social".
  • Put the evidence in the rationale. "Fatigue score 78 after 75 days; CTR down since week six" can be checked. "Underperforming" cannot.
  • Use named measures for impact. Projected impact is a small set of named values, such as a ROAS change, a cost saving or a reach increase. The card shows each one as a chip, formatted as a percentage, a currency amount or a multiple according to its name.
  • Give confidence on the 0 to 1 scale. The card turns it into a percentage and colours the bar: green from 80%, amber from 50%, red below.
Example · one made-up recommendation, start to finish

A geo expansion for a search flight

On Thursday an analyst's script writes a medium-priority geo expansion for a fictional garden-supply advertiser's search flight: add two neighbouring markets, rationale "conversion rate in the current market has held for six weeks; the two markets are in the same delivery zone", confidence 0.64, projected impact as a reach increase. On Monday it sits in the active queue with an amber confidence bar. The buyer opens the campaign from the card's tag, checks the advertiser's delivery map, and accepts. The card moves to History with her name and the time, and she adds the two markets in the search platform that afternoon.

Reading the confidence and impact

Confidence is how sure the proposer is, not a guarantee, and projected impact is an estimate. Treat a high-confidence, low-cost change (rotating a creative that has run for months) differently from a high-impact, hard-to-reverse one (moving a large share of budget off a channel). A sensible rule many teams adopt is to act on high-priority items the same week, and to ask for a second pair of eyes on any budget shift above an agreed size.

How to decide, quickly

If the change is…And confidence is…A sensible default
Easy to reverse (creative rotation, small bid change)HighAccept this week; the owner makes the change
Easy to reverseLow or mediumAccept if the rationale checks out on the linked record; otherwise dismiss with a reason
Hard to reverse (large budget shift, a market exit)AnyCheck the Budget & Bidding tab and any client commitment first; get a second reviewer
Blocked by something outside the data (a contract, a launch)AnyDismiss, and write the reason where your team keeps client notes

These are starting rules, not product behaviour: InsightAdScale records the decision, and your team sets the policy.

Budget context, one tab away

The Campaign Hub's Budget & Bidding tab puts proposals in context: budget and spend by channel side by side, the pacing of every active campaign, and a short list of budget observations. Before accepting a budget shift, it is worth a look.

The Budget and Bidding tab of the Campaign Hub showing budget allocation by channel as stacked bars and a list of budget observations.
Budget & Bidding: allocation by channel and a list of budget observations, for the same made-up portfolio.

Why keeping the history matters

Six months on, the most useful question is often "why did we move that money?" History answers it: what was proposed, the evidence given, the confidence claimed, and who accepted or dismissed it on which date. That record also lets you look back at which kinds of proposal tend to be accepted, and which are usually dismissed, and tune whatever generates them.

Questions teams ask

Does accepting change anything on the ad platform?

No. Accepting sets the recommendation to accepted and records when and by whom. The change itself is made by the person who owns the campaign, in the platform or the campaign record.

Can a decision be undone?

Not from the screen in the current build: there is no "reopen" button. If circumstances change, create a new recommendation that says so; the History then shows both, which is usually the more honest record anyway.

Who can accept or dismiss?

Any signed-in user. Creating recommendations is limited to administrators, so a team can open review to buyers without opening the queue to anyone who wants to add to it.

Is there a measure of how useful the queue is?

The Accepted tile shows an acceptance rate across decided items. A rate that falls over time is a signal to look at whatever is generating the proposals, not at the buyers.

See it in a working week: a made-up agency's Monday budget review.

See it on your own campaigns

Bring one quarter of campaign data

We'll walk through the command center, attribution and recommendations with your channels and your team, as a hosted trial or in your own cloud.

Request a demo