Adam Stamper, The Automation Engine; Ed Sweet, Unicorn Orange; James Northcott, Toucan; Matt Anderson, SG – The Growth Ecom Co
A product’s price drops. Should its advertising stop, or should the brand spend more to support the promotion?
The price movement alone cannot answer that question.
You need to know what the brand is trying to achieve, what it can afford and which trade-offs it has agreed to make.
That is where I see the next test of relevance for Amazon agencies: access to commercial data and knowledge that Amazon does not have, and the ability to use it to change what happens next.
Our first issue asked what clients would pay agencies for as Amazon takes on more of the work. This edition takes that question into the operating model. What must an agency know, retain and be trusted to decide?
Four contributors offer practical perspectives, starting with a simple example of the same platform signal leading to opposite actions.
The knowledge beyond the account
Amazon holds extensive information about shopping behaviour, advertising and marketplace performance. Agencies should take the growing capability of its tools seriously.
Amazon’s latest Seller Assistant announcement describes persistent memory, automated workflows and connections through Quick and Claude. Quick can bring Amazon information together with other business data. Amazon Ads Agent also supports campaign decisions and actions across sponsored ads, Full-Funnel Campaigns and Display, Video and Audio.
So the agency proposition cannot rest on assuming Amazon will remain confined to account data.
The opportunity is in the relevant information the platform does not have: true product economics, cash commitments, retailer agreements, competing priorities across channels, internal capacity and the reasons leadership accepted or rejected a previous recommendation.
Some of this is recorded in other systems. Some emerges through conversations, negotiations and experience. Some is commercially sensitive and may never be shared with a platform.
That access must be earned through the client relationship, kept current and translated into decisions. A margin figure from six months ago is little help if costs have changed. A growth target is incomplete without knowing what the business is prepared to sacrifice to reach it.
The useful distinction is between the platform signal and the commercial context that determines the response.
1. The same data can demand opposite actions
Adam Stamper, The Automation Engine
Adam supplies a simple illustrative example.
One client prioritises profitability. If its Buy Box price falls below an agreed floor, the agency configures advertising to pause.
Another client prioritises promotional sell-through. A price drop below a set threshold becomes the cue to activate advertising and support momentum.
The signal is the same. The action is different because the strategy is different.
The important agency decision happens before the automation runs: agreeing the objective, establishing the threshold and determining the response. The agency sets the conditions under which the system acts and reviews whether those conditions still apply.
That example gets to the heart of the issue. More data does not settle what the business should do. The system also needs the client’s objectives, agreed thresholds and permission to act.
Organisational memory becomes useful when it informs those rules, including when a rule should be reviewed.
If the client changes its priorities, yesterday’s sensible automation can become today’s expensive mistake.
2. Better decisions need a reliable explanation
Ed Sweet, Co-founder & Managing Director, Unicorn Orange
Ed’s contribution starts with the less glamorous work: creating systems that capture and update context continuously.
At Unicorn Orange, he describes a data pipeline combining performance and descriptive information through APIs and MCP connections, with architecture files directing models towards the appropriate sources.
The important point is what that context helps the agency investigate.
A sales decline might involve stock constraints, a competitor’s price reduction and weaker creative. Identifying one plausible explanation does not tell you how much of the movement it accounts for.
Ed argues that structured historical and commercial context can help models examine those relationships and estimate the contribution of different factors. Those estimates still need scrutiny; finding a pattern does not establish its cause.
His closing distinction is worth keeping:
“It’s then up to us to decide what we should do next.”
For agency leaders, the question is whether the team is recording enough to support that decision. A dashboard shows the movement. Useful memory preserves the conditions and reasoning behind it.
3. Challenge the recommendation against the brand’s reality
James Northcott, CEO & Founder, Toucan
James sees more automated execution as an opportunity for agencies to concentrate on strategic judgement.
He argues that agencies need to retain margins, stock positions, wider marketing activity, previous tests, customer behaviour and the reasons behind past decisions.
His example concerns measurement.
A full-funnel campaign may report strong attributed performance. The agency still needs to ask how much reflects existing customers or demand that was already there, and what further analysis or testing would support a claim of incremental growth.
He points to AMC as part of that investigation. The distinction matters: attributed sales and incremental sales answer different questions.
The same principle applies to AI recommendations. A suggested increase in media spending should be assessed against the brand’s economics and objectives, with a clear reason for accepting, changing or rejecting it.
James captures the opportunity neatly:
“AI doesn’t remove the need for organisational memory. It gives agencies the opportunity to operationalise it.”
That requires context to influence the workflow, rather than sit in a document alongside it.
4. The decision may sit beyond advertising
Matt Anderson, SG – The Growth Ecom Co
Matt offers a useful challenge to the whole premise.
Agencies already hold years of bid histories, campaign tests and listing experiments. He argues that access to that history becomes less distinctive as tools make it easier to retrieve and analyse.
The commercial opportunity is in the decisions it enables.
His example is a SKU whose ACoS has worsened for three consecutive quarters. That history should prompt questions about the range, sourcing, packaging, margin and the team’s ability to spot the problem earlier.
It may justify a change to advertising. It may also reveal a product or operating problem that advertising cannot resolve.
Matt’s argument moves organisational memory into change management: helping the client make decisions, strengthen its own capability and improve how the business operates.
It also raises a question for agencies positioning themselves as strategic. Do they have the skills, scope and client permission to have that conversation?
Remembering the problem is useful. Helping someone resolve it is what gives that knowledge commercial weight.
Make the knowledge change the workflow
The thread connecting these contributions is practical.
Adam shows how the objective changes the response. Ed explains why the agency needs a reliable account of what happened. James connects recommendations to the brand’s economics and evidence of growth. Matt asks whether the problem sits somewhere advertising cannot fix.
Together, they make a stronger case for agency relevance than having more reports or running tasks faster.
But there is a catch. Information that lives in a founder’s head, an account manager’s inbox or the memory of someone who has left is difficult to apply consistently.
Organisational memory matters because it makes relevant knowledge available when a decision is made. It should preserve the reasoning, constraints and lessons that change an action, with enough structure for colleagues and tools to use them.
For one client and one recurring decision, start by recording:
The objective: what the business wants to achieve and the trade-offs it will accept.
The commercial inputs: the margins, cash position, stock commitments or channel considerations that affect the decision, including their source and when they were last checked.
The history: what was tried, what happened and why the team accepted or rejected previous recommendations.
The response: what should happen under which conditions, including when to stop or take no action.
The authority: who can recommend, approve and act.
The review: what would trigger a reassessment.
The point is to make that information influence the recommendation or automation rule.
If the client’s priority changes from profitable growth to clearing stock, the response may need to change too. Keeping the old rule running efficiently would simply make the wrong decision happen faster.
Access creates an opportunity. Authority makes it usable.
Having relevant knowledge does not automatically give an agency permission to act.
The client and agency need to agree which decisions the agency owns, which require approval and which remain with the brand. Accountability should follow that authority.
That agreement also makes the agency’s contribution easier to explain. The recommendation changed because the margin would not support it. The campaign was paused because stock was committed elsewhere. The apparent performance improvement was challenged because the evidence did not establish additional demand.
These are the decisions an agency should be able to explain and review.
The knowledge gap will also move. Clients can connect more systems, and platforms can receive more information. An agency cannot build its future on a permanent assumption that a particular dataset will remain unavailable to Amazon.
It needs to keep earning access to the client’s wider commercial reality, judge which information matters and use it to improve the response.
That is the progression I see: from carrying out tasks to deciding which actions should happen, which should not, and why.
The question for agency owners is:
What does your agency know that Amazon does not, and which decisions become better because you know it?
With thanks to Ed Sweet, Matt Anderson, Adam Stamper and James Northcott for their contributions.
ELEVATE | Strategic advisory for Amazon agency owners
Start-Up → Scale-Up → Sale