AI Analyst

Evidence, before the decision.

A Decision Brief is only worth reading if you know what it was allowed to see. This is the context Vendlen builds first, and the two scopes any decision can be judged at.

Workspace
Ads Manager, in the App
Scope
An individual ASIN, or a Product Group read as one context
Evidence
Consecutive periods, read against each other and against your change history

Four of the decisions a brief can carry.

Illustration — products and figures are invented. Not the whole set either: a brief can also move a placement adjustment or an audience boost.

  1. Remove negativeExact

    Unblock “puppy training treats” on this ASIN

    It was blocked in this ad group on 4 clicks and no orders. The negative only applies here, and the same ASIN has kept taking orders on that query in the ad groups where it still runs. Its variations take far more.

    This ad group, before the block
    4 clicks · 0 orders
    Same ASIN, ad groups still running it
    26 orders · 29–47% ACOS
    Its variations, same query
    49 orders · 42% ACOS

    365-day lookbackProduct Group scope

  2. Harvest keywordExact

    Create the keyword “freeze dried salmon treats for puppies”

    It converts in two independent records: the auto target that surfaced it, and the search query report for the destination ASIN.

    Auto targeting, 90 days
    37 clicks · 7 orders · 44% ACOS
    Search query report, 84 days
    5 clicks · 1 order
    Opening bid
    $1.46, inside your $4.50 limit

    90-day periodIndividual ASIN scope

  3. Negative product target

    Block this product target

    The target is a freeze-dried salmon fillet for people and the ad is for dog treats. It does not belong here on relevance alone. The clicks it has taken confirm that rather than being the argument for it.

    Relevance
    A different product category
    This target, 91–120 days
    84 impressions · 2 clicks · 0 orders
    Variations on the same target
    Nothing to compare against

    Days 91–120Individual ASIN scope

  4. Bid change

    Trim this bid from $1.13 to $0.95

    ACOS stayed above your ceiling across the long period, and the most recent one barely moved. The trim is sized to the gap it has to close, not cut flat.

    60 days
    23 clicks · 3 orders · 82% ACOS
    The most recent period
    $41.23 at 78% ACOS
    Trim
    15.9%, inside your bid limits

    60-day periodIndividual ASIN scope

Relevance

It knows what the target actually is.

The hardest waste to find is the target that was never right for the product. It does not look unusual in a report, because it behaves exactly like a target that simply needs more time.

Your ad

Freeze-dried salmon dog treats, 4 oz

The target, as it appears

asin="B0EXAMPLE01"

Nothing here says what it is

The product behind it

Freeze-dried salmon fillets, 8 oz

A different category, and not a substitute

Propose a negative product targetBefore it spends, not after it has spent enough to prove it

Illustration — The product, the ASIN and the wording are examples, and no account data is shown.
  1. 01A code is not a decisionAn ASIN target arrives as a string of characters. Nothing in the row says what the product is, what category it sits in, or whether it has anything to do with what you sell. AI Analyst reads the listing behind it and puts the product back into the sentence.
  2. 02Relevance is about the product, not the metricA keyword either belongs to what you are advertising or it does not, and no threshold on clicks or ACOS answers that. The brief states the relevance judgement in words, so you can disagree with the reasoning rather than with a number.
  3. 03Which means it can act before the spendA rule has to wait for proof, and proof is expensive: clicks that returned nothing. When a term never belonged in the first place, waiting for it to demonstrate that is paying for evidence you did not need.

What a brief is made of.

Six parts, every time. A brief you cannot argue with is a brief you cannot trust, so each one carries the evidence and the period it was read at.

What it reads, and what it does not.

AI Analyst works from your own advertising record and the relationships between your products. It does not read Inventory, so no brief is ever argued from how much stock you happen to hold. What you are about to run out of is a real question, and it is answered on its own side of the product rather than smuggled into an advertising one.

How it reads timeWhatever periods the question needs

A week, a fortnight, something longer — chosen for the question rather than fixed in advance, and read one after another instead of folded on top of each other. That is what makes a break visible instead of averaged away, and it is why AI Analyst can reach back further than the Data Window an Automation rule is held to.

  • Product Listings
  • Advertising keywords
  • Search terms
  • Search query reports
  • Ads performance
  • Change History
  • Product relationships

And then you act on it.

Apply it as writtenThe action goes to the Amazon Ads account you connected, and the change is recorded with its type and source, so what happened is readable afterwards.

Or repeat the judgementWhen the same reasoning keeps being right, it becomes an Automation rule with its own criteria, its own limits, and a preview before anything applies.

Or disagree with itLeave the proposal unapplied when its reasoning does not hold. If the term should be protected from Automation too, add it to the ASIN's Whitelist.

Working context

What a decision is allowed to see.

Vendlen builds the Product Line itself from what your account already contains. You can add business context when it helps, but you are never asked to model your catalog before the product becomes useful.

Vendlen reads

  • Product Listings
  • Advertising keywords
  • Search terms
  • Performance history
  • Seasonality it has observed

and maintains

Product LineHydration GearCategory working context and the observations that accumulate with it
  • Individual ASINInsulated Travel TumblerCarries its own keyword set, so it is judged on its own evidence

    Decision scope: this ASIN

  • Product GroupCycling Water BottlesASINs that share one meaningful keyword set: compact, standard, and large
    Read as one aggregated context, never sibling by sibling

    Decision scope: this Product Group

Fictional product example used to show the structure.

How a brief gets written

Find the break, then find out what caused it.

AI Analyst is not a scoring pass over your campaigns. It works down to the individual search term and the individual targeting, looking for the ones that stopped behaving the way they used to, and it goes to your own change history for the reason before it proposes anything.

  1. 01Read every target on its ownEach search term and each targeting is read individually, not folded into a campaign average. A single term going wrong inside a campaign that still looks healthy is exactly the case an average hides.
  2. 02Compare periods that do not overlapA period is read against the periods before it, rather than against one long cumulative range that already contains it. The lengths are not fixed — a week, a fortnight, something longer, whatever the question calls for. What that surfaces is the break: the point where a target stopped behaving the way it had been.
  3. 03Look in change history for the causeA break usually has a reason on the record. Change history is searched around the date it happened — a bid that moved, a negative that was added, a target that was enabled or paused — and that becomes the evidence the brief argues from.
  4. 04Write the decision down, with the actionThe result is a Decision Brief: one action, the reason, the rows it rests on, the period they were read at, the scope they were read at, and an Apply control. You can act on it, turn it into a rule, or disagree with it.

This is not the same thing as an Automation Data Window. A rule runs on a window you configure, and that window has a hard ceiling of 180 days no matter which plan you are on, because a rule has to be predictable to be safe. AI Analyst is doing the opposite job: it is answering a question, so it takes the periods that question needs and can reach back further than a rule ever will.

Beyond about a year the surviving record is thin, and a brief leaning on it says so rather than presenting it as a strong reading.

Why the scope matters

One row rarely settles the question.

Average six weeks into a single number and the fortnight that changed disappears inside it. A query you blocked on one ASIN can still be earning across the rest of its Product Group. Reading every target this way by hand, one at a time, does not scale.

Same target, three consecutive periods

  1. Days 42–299 orders, ACOS inside your ceilingBehaving
  2. Days 28–158 orders, ACOS inside your ceilingBehaving
  3. Last 14 daysSpend held, orders fell awayBreaks the pattern

Then, in your change historyDay 26 · bid raised $1.20 → $1.90The cause the brief argues from, rather than a target that simply “stopped working”.

The periods do not overlap: each is read against the ones before it, so a change in behaviour shows as a break instead of being averaged away. Bar lengths are a relative drawing, not measured performance.

Same query, two scopes

Selected ASINQuery blocked hereNo traffic since it was negated
Remaining Product GroupRead as one contextOrders continue against the same query
The remaining ASINs are always aggregated into a single context. Vendlen does not compare siblings one by one.

Read the reasoning before you apply it.

Every brief names its action, its reason, the period the evidence came from, and the scope it applies to, so you can disagree with it in the same place you would accept it.