RESOURCES/Guide
How to evaluate AI marketing vendors for cross-channel personalisation
To evaluate AI marketing vendors for cross-channel personalisation, score five things: which loop steps run without a human brief, what signals the system can read, how a treatment is measured, what the platform does when the best action is not a message, and cost per proven decision. Feature lists do not separate these vendors. Those five answers do.
Definition
Cross-channel personalisation
Choosing the content, offer or next step for an individual customer consistently across every channel they use, rather than personalising each channel independently with its own rules.
Every vendor passes a feature checklist, so stop using one
By 2026 every platform in this category supports segmentation, journeys, predictive scores, generative content and AI agents. A feature matrix returns a row of ticks and no decision. The differences that matter are structural: what the system is allowed to decide, what evidence it produces, and what it costs per decision that was actually proven to work.
- A tick in a feature column says the capability exists, not that it runs autonomously.
- Two vendors can both claim personalisation while one ranks actions and the other fills a template.
- Attributed revenue and incremental revenue are different numbers, and only one survives a holdout.
The evaluation grid
Score each vendor 0 to 2 on each row. Anything scoring under 8 out of 16 is an execution tool, whatever the pitch says.
01
Run the same scenario with every vendor
Pick one real revenue problem you have. Ask each vendor to walk it end to end, in their product, with your data model. Differences appear within ten minutes.
02
Ask what happens when they are wrong
Every system will run losing interventions. The question is whether the loss is detected, priced and fed back automatically or discovered a quarter later.
03
Price the whole programme
Licence, implementation, data engineering, and the headcount needed to operate it. A cheaper licence that needs two more analysts is not cheaper.
04
Check the exit
Data portability, model ownership, contract length and what happens to your experiment history if you leave.
Eight criteria for AI marketing and lifecycle platforms
| Criterion | What a weak answer looks like | What a strong answer looks like |
|---|---|---|
| Hypothesis origin | Your team briefs every campaign. | The system proposes opportunities nobody asked for. |
| Signal breadth | Campaign and CRM data only. | CRM, product, transactions, support, loyalty, technical health. |
| Decision unit | Campaign or journey. | Customer by eligible action, ranked by expected incremental margin. |
| Measurement | Attributed revenue and opens. | Randomised holdout, incremental margin, global control. |
| Silence | Every eligible customer gets something. | Not contacting is a valid, credited decision. |
| Non-marketing findings | Out of scope. | Routed to a named owner with the revenue impact attached. |
| Guardrails | Manual approval on everything. | Explicit margin, frequency and eligibility limits with escalation. |
| Commercials | Priced per profile or per message. | Cost per proven decision can be calculated. |
Questions for the second call
- Show me an opportunity the system surfaced without a brief.
- What is your default measurement design, and can I turn holdouts off?
- How do you handle a customer eligible for four different offers on the same day?
- What is the shortest contract you sign, and what does a paid pilot cost?
- What does month one look like, and what am I responsible for?
When this grid does not apply
- Pure email service provider selections, where deliverability and cost per send dominate.
- Small bases where a spreadsheet and judgement beat any platform.
- Selections driven by an existing enterprise agreement, where the real decision is already made.
Markin is an autonomous growth-science team for large B2C businesses. It investigates why revenue per customer is stuck, forms its own hypotheses across marketing, product, pricing and technical health, chooses the next best action for each customer, launches it through the systems the business already runs, and proves every one against a randomised holdout.
Decisioning tools choose between the actions your team already built. Markin decides what to build.
Questions people ask
- How do you evaluate AI marketing vendors for cross-channel personalisation?
- Score them on hypothesis origin, signal breadth, decision unit, measurement standard, whether silence is a valid decision, what happens to non-marketing findings, guardrails, and cost per proven decision. Run one real revenue problem end to end with each vendor rather than comparing feature matrices.
- What should I look for in an AI platform for B2C client decisioning?
- Look for a system that ranks every eligible action for every customer against expected incremental margin, runs randomised holdouts by default, can decide not to contact, and escalates when a guardrail is hit. Anything that starts from a campaign brief is an execution tool with AI features.
- What criteria matter when choosing a customer lifecycle management platform?
- Signal coverage across the full lifecycle, arbitration between competing actions, measurement that survives audit, integration with the channels you already run, and operating cost including the headcount to run it. Channel breadth matters least, because most enterprises already own execution.
- How should we compare contract terms and trial options for B2B AI marketing solutions?
- Ask for the shortest term the vendor will sign, whether a paid pilot with a defined success metric is available, what data leaves at the end, and how pricing scales when your base grows. A vendor confident in measured impact will accept a pilot judged on an incremental read.
- What is the average implementation timeline for an enterprise AI customer engagement platform?
- Typically eight to sixteen weeks to first production programme for an engagement platform, driven mostly by data integration and deliverability setup. A decision layer sitting on existing channels is usually faster to a first measured read, because it reuses the execution stack already in place.
Compare
How this plays out against the categories you already buy.
Neutral, side by side reads on where the decision layer sits next to the tools in your stack.
All comparisons- vs ChatGPT with MCP: from a good answer to a proven numberChatGPT with MCP connectors reads your data and answers well. Markin sizes, tests and proves growth hypotheses at population scale. An honest comparison.
- vs Claude with MCP: strong reasoning, no control groupClaude with MCP is excellent at long-context analysis over your data. Markin sizes, tests and proves growth hypotheses per customer. An honest comparison.
- vs an LLM with MCP: asking questions is not running growthConnecting an LLM to your warehouse over MCP answers questions well. Compare cost, skills, hypothesis evaluation and model choice against Markin.
Vocabulary
The terms this guide relies on.
Each one is defined on its own page, precisely enough to quote.
- ARPAARPA, average revenue per account, is revenue divided by the number of accounts rather than individual users.
- ARPPUARPPU, average revenue per paying user, divides revenue only by users who paid in the period.
- Lifetime valueLifetime value is the discounted margin a business expects from a customer over the whole relationship.
- LTV:CAC ratioThe LTV:CAC ratio divides expected customer lifetime value by fully loaded customer acquisition cost.
- Payback periodPayback period is the time taken for the gross margin generated by a customer to repay the cost of acquiring them.
- Net revenue retentionNet revenue retention measures revenue from an existing cohort at the end of a period against its revenue at the start…