Choosing an AI marketing app: what to check in the demo

Bring one real campaign to the demo

Choose a product you sell and a promotion you could actually run. Ask the app to work with that example. A store-specific example makes missing controls easier to spot than a polished generic demonstration.

Follow the whole path: draft, audience, trigger, schedule, offer, and approval. Ask the demonstrator to change a product, exclude a group, and pause the campaign. Write down which actions you could complete yourself.

Find the boundary between suggestion and action

“AI-powered” can mean many things: suggesting a subject line, drafting a campaign, configuring a flow, or deciding when to send. Your choice should reflect how much decision-making you want to delegate.

Shopify's early-access Campaign Autopilot documentation is a useful example of why details matter. It describes previewing subject, content, audience, and schedule, with approval activating or scheduling the work. Some fields can be edited while others, including the audience, have constraints. Availability is limited to selected merchants. Read the documented controls.

Use that as a question for any app: exactly which parts can you approve or edit, and which parts are decided by the product?

Review the message your customer will see

• Product availability and links to the correct product.
• Discount value, expiry, exclusions, and landing page.
• Personalization fallbacks when a name or product is missing.
• Sender details, unsubscribe controls, and the phone layout.
• The recipient list and channel-specific subscription status.

Ask how subscription changes synchronize with the store and other tools. A generated message does not establish that its recipients should receive it.

Map the automations already running

List your current welcome, abandonment, and post-purchase messages, including the tool that sends each one. Before enabling a new flow, decide which existing flow it replaces and how the transition will work.

A useful trial is one controlled workflow with a clear owner. Record its trigger, exclusions, frequency, and stop conditions. That gives you a way to investigate unexpected messages without guessing which app sent them.

Ask what the revenue number means

Attribution assigns credit using rules. Incremental lift asks what changed because the marketing happened. They answer different questions.

Omnisend describes configurable attribution windows and notes that refunded or canceled orders remain in attributed sales. That is a reason to understand the report's definition before comparing totals, not a claim about campaign quality. See its attribution rules.

Ask about windows, opens versus clicks, refunds, and experiment options. A small store may not have enough data for a conclusive lift test. Keep the trial's operational findings—accuracy, control, and workload—separate from claims about revenue causation.