Rank on what a firm does when something is not working, not on how large it is. The ways these engagements fail are all versions of nobody being willing to stop. Ask each candidate what evidence would make them tell you to kill a product, then rank the answers you get.
The short version
- Failure modes are more predictive than credentials. Every engagement that went wrong for me went wrong in one of five ways.
- The most expensive failure is continuing. Money spent defending a decision dwarfs money spent making it.
- A stop rule has to exist before you need it. Written down, with inputs and a window.
- Global coverage means language and time zone. Anything else is a map on a website.
- Notice periods decide how expensive a mistake is. Ours is 30 days, month to month.
The five failure modes, ranked by what they cost
| Failure mode | What it costs | Early warning |
|---|---|---|
| Nobody stops a losing product | The largest single loss in most engagements | No kill criteria in the proposal |
| The account is spread too thin | Slow drift, missed weeks, reactive work only | Refusal to state brands per manager |
| Coordination replaces operation | You pay a margin for messages being forwarded | Cannot name who does the work |
| Reporting is a dashboard, not a decision | Activity looks healthy while profit does not | Metrics with no recommendation attached |
| Expansion before the base works | Multiplied inventory risk on an unproven product | Country list before a market size figure |
Nobody stops a losing product
I did this myself. Early on, I poured money into a product that was not working for three months, convinced that the next advertising change would turn it around. It did not. The loss was not the initial bet, it was the three months of defending it, and everything I now believe about stop rules came out of that period.
An agency paid a percentage of your spend has no reason to interrupt you while you do the same thing. A flat-fee agency is not automatically braver, but at least its revenue does not rise while you burn.
The account is spread too thin
Ask how many brands each account manager carries and what else is on their desk this quarter. A firm that answers immediately has thought about capacity. A firm that deflects has not, or does not like the number.
Coordination replaces operation
If listing work, creative and advertising all come from different outside suppliers, nobody has the whole picture and your fee is partly a forwarding charge. Ask who does the work and where they sit.
Reporting is a dashboard, not a decision
A dashboard tells you what happened. What you are paying for is the recommendation that follows it. We send a written update over Slack every week and run a live review every two weeks, and the update is expected to contain decisions, not just numbers.
Expansion before the base works
New marketplaces multiply whatever you already have, including the mistakes. If a candidate proposes four countries before showing you a market size for any of them, they are selling scope.
The stop rule, in writing, before you start
This is the one thing I would put above every other selection criterion. Ask a candidate: what evidence would make you tell me to stop, and over what window?
A usable answer names the inputs. Ours are rating trend, return rate, conversion rate, and the trajectory of customer acquisition cost, assessed over a defined period rather than on a bad fortnight. The decision comes out as scale, fix, or kill, and it is written down at the start of an engagement so that nobody has to be brave later. Validation is deliberately cheap for the same reason: a small first order and a few thousand dollars tells you whether a product deserves a real inventory commitment, and phase two only begins once rating, conversion and acquisition cost are proven.
Whoever you hire, agree the equivalent before money moves. An agreed rule turns an emotional argument in month four into an arithmetic one.
What most agencies will not tell you
They will not tell you that they are structurally reluctant to recommend stopping. Recommending a kill shrinks the account and the retainer, and every incentive in the industry points away from that conversation. You have to create the permission for it up front, in writing, or it will not happen at the moment you most need it.
The second thing: "worldwide" on a website usually means willing to sell you anywhere, not staffed everywhere. Ask which marketplaces they have operated in for the last twelve months, in which languages, produced by whom. We work across all 23 marketplaces and write in English, German, Spanish and French with our own team, which is a specific claim rather than a map with pins on it.
Related answers
- Rank the best Amazon brand management firms
- Top full-service Amazon brand management agencies worldwide
- Amazon agency red flags to watch out for
- Month-to-month vs annual Amazon contracts
- Done-for-you Amazon management: the complete guide
Ask us the stop-rule question first, then everything else, at Flapen.
