What to Test During an Omnichannel Inbox Free Trial: A 14-Day Checklist

IN THIS ARTICLE

1. 01Before you start: set the trial up to fail honestly
2. 02The five tests that actually predict regret
3. 03The eight things vendors hope you will not check
4. 04A 14-day plan
5. 05What a good trial offer looks like

Short answer. During an omnichannel inbox free trial, test five things in this order: cross-channel identity (does one customer on two channels produce one record?), AI answer grounding (does it refuse when it does not know?), the real all-in price at your actual team size, data export, and handoff quality. Run real conversations through it rather than demo data, get at least two agents using it daily, and do the export test before you fall in love with the interface. Most trials are 14 days, which is enough if you start on day one.

The reason this needs a checklist is that trials are designed to show you the good parts. The interface is the first thing you see and the least important thing you are buying. The things that determine whether you are still happy in a year, data model, pricing behaviour at scale, and how gracefully the thing fails, are all discoverable in a trial but none of them are on the welcome tour.

Before you start: set the trial up to fail honestly

1. 1Write down your three most common conversation types before you log in. Order status, refund request, and a pre-sales question, for example. These are what you will test against, and deciding them in advance stops you grading the tool on whatever it happens to be good at.
2. 2Connect real channels, not a sandbox. A trial run entirely on test data tells you nothing about how the tool handles a real WhatsApp window or a real Instagram comment thread.
3. 3Get two or more people in. Single-user trials hide every collaboration problem: assignment, collision detection, internal notes, and the question of what happens when two agents open the same conversation.
4. 4Note the start date and the cancellation deadline. Most trials are 14 days with a card required. Put the deadline in a calendar on day one.

The five tests that actually predict regret

Test 1: the cross-channel identity test

This is the single most predictive test on the list, and it takes four minutes. Message the trial account from your own phone on WhatsApp. Then message it from your own Instagram. Then email it. Now open the contact record.

If you see one contact with three conversations attached, it is genuinely omnichannel. If you see three separate contacts, it is a multichannel tool with tabs, and every promise about unified customer history is marketing. This distinction is almost impossible to reverse later, because it is a property of the data model rather than a feature that can be switched on.

Test 2: the AI grounding test

Load five real knowledge base articles. Then ask the AI agent three kinds of question: one it can answer from those articles, one that is adjacent but not covered, and one that is plainly outside your business entirely.

What you are testing is the second and third. An AI that answers the uncovered question anyway, confidently, is going to invent policies for your customers. The correct behaviour is to say it does not know and hand off. Ask the vendor directly whether grounding is enforced in code or requested in a prompt, because those are very different guarantees, and only one of them survives an unusual phrasing.

Then test the language behaviour if you serve more than one market: ask a question in a second language and see whether the reply comes back in that language or in the language of the source article.

Test 3: the real price test

Not the price on the pricing page. Build the actual number, at your actual team size, on the plan that actually includes your channels. Then build it again at twice your current headcount and twice your current volume.

• Which plan includes the channels you need? WhatsApp in particular is frequently two or three tiers above the advertised entry price, so the number you compared may not be a number you can buy.
• Is it per seat or flat? At one agent these look similar; at fifteen they diverge by several hundred dollars a month.
• Is there a per-resolution or per-conversation fee? If so, model it at your real volume. At 3,000 AI resolutions a month, a $0.99 fee is $2,970 on top of seats.
• Is the advertised price the annual rate? Monthly billing is often around 20% more.
• Does anything meter by contacts? Contact-based pricing means a successful marketing campaign raises your support bill.
• What does AI cost? Included, a per-agent add-on, or per resolution? Can you bring your own model key, and at what markup?

Test 4: the export test

Do this during the trial, not at renewal. Export your conversation data and look at what comes out. Is it complete, including attachments and internal notes? Is it structured usefully, or is it a single unparseable blob? Is it self-service, or does it require a support ticket and a wait?

How hard it is to get your own data out is exactly how much leverage you will have at every future renewal. It is also a GDPR question if you operate in the EU, because access and erasure requests are your obligation to fulfil, not the vendor's favour to grant.

Test 5: the handoff test

Have the AI handle a conversation, then escalate it to yourself as a human agent. Check what arrives: the full transcript or just a summary, whether identity verification persists, and whether actions the AI already took are visible.

Then do the harder version. Have the AI answer a question, close the conversation, and come back the next day on a different channel about the same issue. Does the tool recognise you and surface the earlier exchange, or does it treat you as a new contact? This is the continuity case, it is the most common real-world pattern, and it is the one most tools fail.

The eight things vendors hope you will not check

A 14-day plan

What a good trial offer looks like

Fourteen days is the standard and is enough. What matters more than length is whether the trial gives you the real product. Watch for trials that exclude the AI features, cap channels, or require a sales call before you can connect WhatsApp, because you cannot run any of the tests above on a crippled version. A card requirement is normal and not a red flag, provided cancellation is genuinely self-service.

Delyt's trial is 14 days on any plan, card required, cancel before day 14 and you pay nothing, with trial AI credits included so the AI tests above are actually runnable: 2,000 credits on Solo, 3,000 on Starter, 5,000 on Growth. Every plan includes the full API, webhooks and the MCP server, so the developer checks are available on the entry tier rather than gated behind Enterprise.