Comparison
How Refik differs, and when it does not.
Every row carries the date it was measured, because a competitor’s price and feature set change and an undated claim quietly becomes a false one.
Method
Two sources, two dates, three cell states.
Nothing below is a marketing claim. The field was read once, on 18 September 2026: fifteen apps against fifty-six capability rows, cell by cell, from each app’s own store listing and help pages. Our own column was measured separately, and in two passes — the rows that study already answered carry 15 September 2026, and the thirty-two it left blank were read out of our source code on 4 October 2026. So every row carries the date its Refik cell was measured, and the column heading carries the date the field was read.
A filled dot means yes. A dash means no. An empty cell means neither, and that is a third state on purpose: it marks a row where more than four of the fifteen apps documented nothing either way. Printing a dash there would turn “we could not find out” into “they do not have it”, and those are not the same sentence.
Twenty-nine of the fifty-six rows are here. Twenty-six are here because the field is documented well enough to count, at most four unknowns out of fifteen. Three more are here although the field column on them is empty, because on those our own answer is a refusal rather than a gap — and keeping that apart from a gap is most of the reason this page exists. The other twenty-seven rows are left out, for the reason the field study itself writes down: a row with ten unknowns says “undocumented”, not “absent”.
The grid
Twenty-nine capabilities, counted.
The right-hand column is not one rival. It is how many of the fifteen measured apps have the row at all, which is the question a buyer actually has: not “who wins” but “is this normal in this category, and do you have it”. Eighteen rows are a yes on both sides. On five the field has it and we do not. On three neither side mostly has it. On the last three the field is undocumented and our own no is a decision, written out underneath.
| Capability | Refik | The 15 apps |
|---|---|---|
| Storefront | ||
| A chat bubble on the storefront | Yes | 14 of the 15 |
| Proactive trigger | Yes, with a sentence below this table | 13 of the 15 |
| Selling | ||
| Product recommendation and comparison | Yes | 15 of the 15 |
| Upsell and cross-sell | Yes | 11 of the 15 |
| Support | ||
| Order tracking | Yes | 14 of the 15 |
| Starting a return or an exchange | Yes | 9 of the 15 |
| Handover to a person | Yes | 14 of the 15 |
| Channels | ||
| Web | Yes | 15 of the 15 |
| No — not yet | 10 of the 15 | |
| Instagram DM | No — not yet | 10 of the 15 |
| Messenger | No — not yet | 8 of the 15 |
| No — not yet | 10 of the 15 | |
| SMS | No — not yet | 3 of the 15 |
| Knowledge and training | ||
| Catalogue sync | Yes | 15 of the 15 |
| FAQ, URL and document upload | Yes, with a sentence below this table | 10 of the 15 |
| Correcting an answer so it carries over | Yes, with a sentence below this table | 12 of the 15 |
| Brand voice and persona setting | No — not yet | 13 of the 15 |
| Analytics | ||
| Conversion and attributed revenue | Yes, with a sentence below this table | 13 of the 15 |
| Integrations | ||
| Klaviyo | Yes | 10 of the 15 |
| A ticket in your helpdesk on handover | Yes, with a sentence below this table | 12 of the 15 |
| Review app | Yes, with a sentence below this table | 9 of the 15 |
| Subscriptions | No — not yet | 5 of the 15 |
| Customisation | ||
| Widget theme, colour and position | Yes, with a sentence below this table | 14 of the 15 |
| Greeting message | Yes | 11 of the 15 |
| Quick reply buttons | Yes | 12 of the 15 |
| Onboarding | ||
| A setup-time claim | No — by choice, with a sentence below this table | Empty: 6 of the 15 are undocumented on this row, so there is no figure to print |
| Done-for-you setup | No — by choice, with a sentence below this table | Empty: 11 of the 15 are undocumented on this row, so there is no figure to print |
| Trust | ||
| Built for Shopify badge | No — not yet | 6 of the 15 |
| A compliance certification claim | No — by choice, with a sentence below this table | Empty: 6 of the 15 are undocumented on this row, so there is no figure to print |
- Yes. In the right-hand column, eight or more of the fifteen.
- No. In the right-hand column, fewer than eight.
- An empty cell is not a no. It means we have no figure to print.
- Proactive trigger. Exit intent and two time axes are wired. Scroll only resets the counter, and the page axis is fixed to product pages.
- FAQ, URL and document upload. Only question-and-answer pairs typed in by hand, capped at three hundred entries. Crawling a URL, uploading a PDF and pulling your own store pages are all absent.
- Correcting an answer so it carries over. Two separate mechanisms. The carry-over is lexical rather than learned: a correction reaches wordings that overlap the one you fixed.
- Conversion and attributed revenue. One store at a time, counting the orders Refik touched. There is no combined view across stores.
- A ticket in your helpdesk on handover. One provider, with a live call, an encrypted key and an hourly retry. The second provider named in that matrix row is not in our code.
- Review app. It reads the product rating field Shopify itself stores, rather than connecting to a review app. The rating arrives; the reviews do not.
- Widget theme, colour and position. Colour is complete. Position is half: phone bottom spacing only, with the side edge fixed. There is no light or dark theme switch.
- A setup-time claim. The setup page counts the steps and refuses to put a number of minutes on them.
- Done-for-you setup. Nobody sets it up for you. The word “agency” on our pages means your agency, given access by you.
- A compliance certification claim. We hold no SOC 2 report and no ISO 27001 certificate, and the security page says that in those words — what is wired is the behaviour the certificate usually stands for: the deletion webhooks are live.
Source for the right-hand column: our own competitor study, 15 apps across 56 capability rows, read . Source for the Refik column: the pricing study of for the rows it covered, and a read of our own source code on for the thirty-two it did not. Both are documents in this project, not vendor material; a test rebuilds every cell above from them and fails if the page and the measurement disagree.
The field, named
The fifteen apps, by name.
This is the only page on this site that names them, and it names all fifteen rather than the two or three that flatter us. The two facts in the table are the two the study documented for nearly every app, which is why they are the two we print: the Built for Shopify badge, and whether the app answers on WhatsApp. Where the sources we read said nothing, the cell is empty — one of the fifteen had been pulled from the store at the time of reading, and one is an enterprise tool with no store listing at all.
One more number, because leaving it out would be the dishonest choice: where the store still listed them, their review counts ran from 5 to 5,794. Refik has none, because it is not listed yet.
| App | Built for Shopify | |
|---|---|---|
| Zipchat | Yes | Yes |
| VanChat | Yes | Yes |
| Rep AI | No | Yes |
| Tidio | No | Yes |
| Gorgias AI Agent | No | Yes |
| Shopify Inbox | Yes | No |
| Manifest AI | No | Yes |
| Octocom | No | Yes |
| Verifast | No | Yes |
| Chizy | Yes | Yes |
| Buddy AI | Yes | No |
| Carti | No | No |
| Alby | Empty: the sources we read did not say | Empty: the sources we read did not say |
| Siena AI | Empty: the sources we read did not say | Yes |
| Octane AI | Yes | No |
| Refik | No | No |
Both columns are read out of the same study, , and the Refik row out of the pricing study of . Names, badges and review counts move; this table is a reading on one day and it says which day. It is also the only place on this site where any of these names appears, and a test enforces that: the name has to sit in a dated row of a sourced table, or the build fails.
One app at a time
Pages that read one app’s own pricing page.
The table above compares us against the field as a whole, which answers “is this normal in this category”. It does not answer “what would switching from the thing I already pay for actually change”. These pages do, one app at a time, and each one is built the same way: the app’s own pages read on a single day, quoted word for word, with every figure carrying that date. Where a page has an empty cell, it says underneath which side was not read — a gap in our reading is not a claim about their product.
Each row carries the date that app was read, on or later. A row appears here only when the page behind it exists; a name with nothing measured behind it gets no page at all.
| The app | What the page compares |
|---|---|
| Freshdesk | A per-agent seat price against a conversation price, with their AI priced as a separate per-seat line, and which platforms each one connects to. |
| Gladly | A conversation price against a side that could not be read: on the day we looked, the address that should hold their price list answered with the home page instead, so that column is empty and says why. |
| Help Scout | A per-user monthly price against a conversation price, the tiers on each side, and where each one is allowed to take its answers from. |
| Intercom | A help desk priced per seat and an AI agent priced per resolved question, against one assistant priced per conversation. |
| Kustomer | A pricing page that states it does not post prices, against three band prices on a public page — and what can still be compared when one side publishes none. |
| Re:amaze | A per-team-member price with the AI agent metered separately, against a conversation price with the volume already inside the band. |
| Richpanel | Three different meters in one place: monthly order volume, agent seats, and conversations — and why an order and a conversation do not convert into each other. |
| Salesforce | A per-user service platform sold into every trade, against one assistant inside one storefront, with every price cell naming the tier it was taken from. |
| Shopify Inbox | A free app published by Shopify against a paid third-party app billed per conversation, and what the free one is and is not built to do. |
| Siena | A monthly platform fee plus a charge for every automated ticket, against a band with its conversations already inside it — a floor against a ceiling. |
| Yuma | A price charged per fully resolved ticket on top of the help desk a shop already runs, against a price charged per conversation inside the storefront itself. |
| Zendesk | A seat price against a conversation price, the three list tiers on each side, what is billed on top of the subscription, and which platforms each one connects to. |
| Zowie | An enterprise AI agent platform against a per-conversation assistant, and four rows that carry one side only because on the day we read it their price list was not there to read. |
Edges
What this page will not do.
A comparison page is the easiest place on a website to lie, so these are the four things we decided it would not contain.
- No prices on this page. The study behind the table above was read on and its price column has not been re-read since, so printing it here would put a possibly stale number next to someone else’s name. Prices do appear on the per-app pages listed above, and only there: each of those was read from that app’s own pricing page on one day, and the figure carries that date and that page in the row with it.
- No overall verdict. There is no score, no total and no sentence claiming we come out ahead. Five of the twenty-nine rows are a no for us and a yes for most of the field, and they are in the table rather than in a footnote.
- No customer outcomes. No case study, no named store, no measured lift — here or anywhere else on this site — because we have not measured one.
- No links out, and no vendor claim about a vendor’s results. Nothing on this site links to any of the apps named. The per-app pages do quote an app’s own words about its own price and its own positioning, with the date and the page they were read from, because on those two things the vendor is the primary source. What is never used as evidence is a vendor’s claim about how well its product performs — theirs or ours.
This directory is not deployed anywhere and robots.txt disallows every path, so none of this is reachable or indexable yet. If a cell here is wrong, that is a bug in the page and we want to hear about it: a claim can be accurate on the day it was read and still be wrong today.