Reviews
Part of Strong AI marketing tools reviews start with a written scoring rubric
Principles for reviewing AI marketing tools
A practical framework for reviewing AI marketing tools, covering five scoring areas, evidence types, governance checks and the limits of desk-based assessment.
What to take away
An AI marketing tools review methodology is the written set of tests, sources and scoring rules a publication uses to judge a marketing platform. It exists so two reviews can be read side by side, and so a reader can see where judgement ends and evidence begins.
- Every review names what was assessed and what was left out.
- Vendor claims and independent evidence sit in separate columns.
- Scoring weights are fixed before a tool is examined, not adjusted afterwards.
- No review claims hands-on testing that did not take place.
What the methodology assesses
Five areas carry a score: data handling and permissions, workflow fit, reporting and attribution, integration, and pricing and exit terms. Weights are agreed in advance, so a long feature list cannot lift a weak score elsewhere.
Five scored areas
- Data handling and permissions
- Workflow fit
- Reporting and attribution
- Integration
- Pricing and exit terms
Scores rest on documents any buyer could obtain as well: public pricing pages, product documentation, security pages and release notes.
If a supplier publishes nothing, we record a gap, not an assumption. Each review keeps a source log, so readers can trace any mark to the page it came from. A mark without a traceable source is not published.
The shortlist grid brings the individual marks together, giving buyers a way to compare platforms against their brief.
What the review does not cover
There is no test lab here. This desk does not open trial accounts, does not sit through sales demonstrations on a vendor's behalf, and does not quote customers it has never spoken to. Those tasks belong to a procurement team working to its own brief.
A review also cannot verify claims about model accuracy, inbox placement or cost savings unless supporting data is published. Where a claim arrives without evidence, it is labelled a vendor statement.
What a review can do is shorten the reading list. It points a buyer at the questions worth asking and the documents worth requesting before a trial begins.
Vendor claims and independent evidence
Keeping the two apart is the most useful discipline in the method. A claim tells you what a supplier wants buyers to believe. Evidence tells you what a buyer can check without the supplier in the room.
Claims versus evidence
Vendor claim
- What it is
- What supplier wants believed
- Feature exists
- Logged as claim
- Regulator guidance
- Not applicable
- Vendor references
- Named use cases
- Published pricing
- List bands
- Disagreement
- Noted in text
Independent evidence
- What it is
- What buyer can check
- Feature exists
- Reproducible by buyer
- Regulator guidance
- Pass or fail checks
- Vendor references
- Context, never proof
- Published pricing
- Compared like for like
- Disagreement
- Governs the score
| Evidence type | What it can support | How it is used |
|---|---|---|
| Vendor documentation and release notes | That a feature exists, and its stated limits | Logged as a claim until a buyer can reproduce it |
| Regulator and standards guidance | Legal and governance duties | Applied as pass or fail checks |
| References chosen by the vendor | Named use cases | Read as context, never as proof |
| Published pricing | List bands and contract terms | Compared like for like where units match |
| Editorial judgement | Fit for a stated brief | Scored against the published rubric |
Judgement still carries weight. A tool built for one channel may score badly overall and yet be the right answer for a narrow brief. The total is a summary of fit, not a verdict on quality. Where a claim and the evidence disagree, the evidence governs the score, and the disagreement is noted in the text rather than smoothed over.
Governance, data and ethics checks
Risk questions follow the NIST AI Risk Management Framework, which describes how organisations govern, map, measure and manage AI risk. A tool that cannot answer basic governance questions loses ground before any feature comparison begins.
Email automation faces a separate test. Consent and soft opt-in rules decide whether a sequence can be sent at all, so platforms are checked against ICO guidance on direct marketing using electronic mail.
Ethical questions, including how contact data was sourced and how opt-outs are honoured, are judged against the DMA code. A tool that automates a practice the code discourages will not score well on governance, however tidy its interface.
This rubric covers software rather than agency services. Buyers comparing agencies should assess pitches, references and contract terms against their own brief.
How often the method is refreshed
Rubrics are revised each quarter, and sooner if a regulator publishes new guidance. A refresh does not rewrite past scores. It changes the questions asked from then on, so readers can see if a review predates a rule change.
Every review carries the date of the rubric version used. That stops an older piece being mistaken for a current judgement.
Common questions
Who sets the scoring weights?
The editorial desk sets them and publishes them before a review begins. Weights can change between review cycles, but never mid-review, and any change is noted in the piece.
Does a vendor briefing count as evidence?
No. A briefing is logged as a claim from a named source. It becomes evidence only when a buyer can corroborate it, for example through a published subprocessor list.
Can a low score still be a sensible purchase?
Yes. A tool marked down on integration may suit a team running a single channel. The rubric rewards fit to a stated brief rather than size or reputation.
