The History Inside the InboxNarrow moat
Text (TXT) — moat facet
Six years of a customer's transcripts is what trains the automation - and the reason Text says its 74% resolution rate has further to run.
The least visible thing Text holds is the archive.
A customer that has run its support through LiveChat for six years has six years of transcripts inside it: what people asked, which answers ended the conversation, which ones did not. That corpus is what the automation is trained on, and Text says so in the sharpest way available — its resolution rate of 74% against an industry average of 59% is reported as including customers who have not yet had the chance to train the models on their own data1. The implication is that the figure goes up once they do.
This is the part of the moat that genuinely strengthens with age, and it is the reason the multi-product statistics move the way they do. Accounts using more than one product reached 38,8% of recurring revenue and accounts above 500 dollars a month passed half of it2: customers who have committed their history are the ones who buy the next thing.
The limit is that a transcript archive is portable in principle and increasingly portable in practice. It is text. A rival that offers to import it, and that can train on it just as well, converts Text's deepest asset into a migration task.
The other limit is the one the numbers already show. Whatever the archive is worth to the customers who have one, 4% of LiveChat's customers left every month of the year3 — meaning a large share of the base never accumulates enough history for any of this to bind them.
The test is the resolution rate on trained accounts specifically. Text has not disclosed it separately, and until it does the 74% is a company-wide average carrying an argument it cannot yet support.
The archive is worth more every year because its purpose changed: it is now training material, and Text says its 74% resolution rate includes customers who have not yet trained on their own data.
The archive of past conversations costs Text little to keep and trains the automation that answers new ones. If that advantage is real, margin should hold as more conversations are handled without people.
Source: Text Group Q1 2026/27 results release ↗- ReportedThat corpus is what the automation is trained on, and Text says so in the sharpest way available - its resolution rate of 74% against an industry average of 59% is reported as including customers who have not yet had the chance to train the models on their own data.Text Group Management Board report for 2025/26, the chief executive's letter (the statement that on the financial results the past fiscal year was not a successful one, with dollar revenue similar to the prior year but lower after conversion to złoty and costs incurred in both currencies; an artificial-intelligence resolution rate of 74% against an industry average of 59%, a figure said to include customers who have not yet trained the models on their own data; the intention to measure success in dollars earned rather than dollars saved on customer support, building an ecosystem that becomes a growth engine for customers rather than a cost centre; and the assessment that Text operates in a market with strong competitors, many with enormous resources they can and sometimes must allocate to customer acquisition, that Text must be smarter, more creative, harder-working and more efficient, and that success is not guaranteed) — FY2025/26 · publ. June 2026 · source ↗
- ReportedAccounts using more than one product reached 38,8% of recurring revenue and accounts above 500 dollars a month passed half of it: customers who have committed their history are the ones who buy the next thing.Text Group Management Board report for 2025/26, clients and customer acquisition (customers in around 150 countries across effectively all sectors, with the USA, Great Britain, Australia, Canada and Indonesia the most important markets and Poland about 1,5% of revenue in twelfth place; customers using more than one product at 38,8% of MRR, up seven percentage points, and accounts above USD 500 a month passing 50% of MRR; and the year's largest new contracts including an increase to seven figures with an American online retail company, a postal service in the British Isles, an Asian industrial automation company, a Kyoto corporation, a Texas dental group, a Scandinavian food company and universities in the United States, New Zealand and Singapore) — FY2025/26 · publ. June 2026 · source ↗
- ReportedWhatever the archive is worth to the customers who have one, 4% of LiveChat's customers left every month of the year - meaning a large share of the base never accumulates enough history for any of this to bind them.Text Group Management Board report for 2025/26, margins, costs and cash (gross profit margin on sales of 67,6%, operating margin 38,6% and net margin 35,4%; cloud infrastructure costs rising, the migration completing in July 2025 after more than a year of duplicated cost, and completion not translating into cost reductions because of price increases regardless of provider and a deliberately expanded scope of purchased services; increased consulting, legal and public relations costs; fourth-quarter costs falling on cloud optimisation; a warning that more intensive use of artificial intelligence may cause further cost increases and that rising AI costs should press hardest on competitors offering free or freemium products; operating cash flow of PLN 161,6m and PLN 62,8m of cash; and a dividend policy of allocating the highest possible part of profit to shareholders) — FY2025/26 · publ. June 2026 · source ↗