Competitor switching intent: how to catch buyers mid-migration

Competitor switching intent is a buyer saying in public, in their own words, that they are leaving a product you compete with.

The first real test run of LeadSurface returned 847 results. Six mattered.

I read all 847 by hand, because at that point there was nothing else to do with them. Most were people discussing a category. Some were vendors marketing to each other, and a handful were three-year-old threads that still rank. The six worth a human reply had one thing in common I hadn't designed for. Every one of them was someone in the middle of leaving a specific product, and saying so by name.

That's my own number, so treat it as an anecdote and not a measurement. It changed what I built next.

What is competitor switching intent?

Competitor switching intent is an observable, public statement that a buyer is moving away from a named vendor. It sits between a complaint and an RFP. The buyer has decided the current tool is no longer sufficient and has started asking what to use instead. Neither the vendor being left nor the one that should replace it sees it.

What separates it from ordinary category intent is that the incumbent is named. A surge score tells you an account is researching observability tooling. The switching post names which product failed, what it failed at, and who is irritated enough to type all of that into a public box. It's a different class of buying signal, because the competitive frame is already set and you didn't set it.

Which phrases predict a switch?

Six phrase families do most of the work in practice: direction words naming the vendor being left, shortlist-building questions, reactions to a pricing or license change, stuck-with-it complaints carrying no churn keyword at all, renewal-date mentions, and general grumbling. Urgency runs highest on the trigger reactions and lowest on the grumbling.

Before the table, here's the part most posts on this topic skip. There is no published corpus study of B2B software switching language. The only labeled churn-intent datasets in the public record are consumer telecom tweets from 2015 and German chatbot conversations from 2018. Neither is software, and both predate LLMs. So anyone claiming that "migrating away from X" converts at a specific rate is repeating folklore, including anyone at LeadSurface. What follows is practitioner observation, labeled as such.

Phrase pattern	What it tells you	How urgent	How a keyword rule misreads it
"migrating off X", "moving away from X"	The decision is made. The replacement may not be chosen yet.	High. The shortlist is being built this week.	Matches "migrating to X" from a happy new customer unless the rule parses direction.
"alternatives to X", "what are people using instead of X"	Shortlist actively forming, requirements already written.	High. This is the last moment your name can be added.	The same words dominate SEO roundups and competitor ads.
"X changed their pricing / license / terms"	A trigger fired across the whole user base at once. The thread itself is a lead list.	Highest, and shortest. The thread stops being read in days.	A brand-plus-pricing rule surfaces the vendor's own announcement post first.
"we're stuck on X because Y"	Dissatisfaction with no search started. The earliest observable point.	Low urgency, longest window, best access.	No churn keyword appears at all, so no alert fires.
"our contract with X is up in March"	A renewal date, volunteered.	Scheduled, not immediate.	Reads as calendar chatter. Nothing in it looks like intent.
"anyone else having trouble with X?"	A complaint. Frequently not a purchase.	Usually none.	Highest volume, lowest yield, and the row that inflates every mention dashboard.

The urgency column is judgment, not measurement. Every published signal decay table I checked sources its numbers to the vendor's own case studies, so I won't add another.

The fourth column is where the peer-reviewed evidence lives. Hadi Amiri and Hal Daume III built a 50-term keyword classifier for switching intent and tested it against three telecom brands in an AAAI-15 paper. Precision came in at 36.5, 39.0, and 35.4%. Learned models on the same data averaged 75% F1. Roughly two of every three flagged posts are noise, and the named failure modes are the ones in that column: prepositions reversing the label, negation, competitor poaching promos matching the same terms, and this:

churny contents can be expressed in subtle ways as in "debating if I should stay with BrandName" or "in 2 months, bye BrandName" that contain no obvious churny keywords but clearly express churny contents with respect to the brand.

It's a consumer telecom study on 2015 tweets, and it's the only real test of the method this category still sells.

Why does timing decide who wins a switching deal?

Because the buyer writes the requirements before they contact anyone, and the switching post is the moment those requirements are still soft. Wait for the RFP and you're answering a document shaped by whoever was in the room earlier.

Craig Elias named the moment in 2010 and called it the Window of Dissatisfaction, in SHiFT!:

The Window of Dissatisfaction begins after the decision-maker experiences a Trigger Event and decides that what he or she has is no longer sufficient, but before he or she has started doing anything about it.

His deck pairs that with a close-rate ladder: under 1% in status quo, 10% to 20% once buyers are searching, 60% to 90% inside the window. The ladder is worth tracing back. It comes from an InnerSell survey of 230-odd sales executives in 2003, asking salespeople to estimate their own close rates, with no control group and no observed deal outcomes. Treat it as the origin story of trigger-based selling, not as evidence.

The measured timing data is less flattering and more useful. 6sense's 2025 B2B Buyer Experience Report, surveying roughly 4,000 buyers at a $200,000 to $300,000 median purchase size, found cycle length fell from 11.3 to 10.1 months, first seller contact moved from 69% to 61% of the journey (7.8 months into the 2024 cycle against 6.2 into the 2025 one, so about seven weeks earlier), and buyers evaluate 5.1 vendors at roughly two months each. The companion analysis found more than 80% of buying groups had their requirements fully or mostly defined before speaking to any seller.

Here's the part that cuts against my own pitch. The same research found buyers choose from their day-one shortlist 95% of the time, and that 85% of winning vendors already had prior buyer experience, holding at 85% for replace-existing purchases specifically. A cold, perfectly timed message from a vendor nobody has heard of loses most of the time. So the honest promise is not "get there before the shortlist exists." It's to be one of the few names already in the room when the trigger lands, and to be the first to say something useful after it does.

Which vendor events produce real migrations?

License changes, pricing model changes, and acquisitions. Those three fill public forums with switching language on a clock you can measure in days. Outages, the event everyone expects to move buyers, mostly do not. The difference is that a license change alters the terms of the relationship permanently, and an outage is a bad week.

HashiCorp moved Terraform to the Business Source License on August 10, 2023. The Linux Foundation announced OpenTofu 41 days later, on September 20, 2023, with formal pledges spanning 140+ organizations and 600+ individuals and a minimum of 18 full-time developers committed for at least five years. Forty-one days from a vendor's licensing decision to an organized, funded exit.

Redis relicensed in March 2024. The Linux Foundation announced Valkey on March 28, 2024. Same play, faster clock.

WordPress.org banned WP Engine on September 25, 2024. Automattic built a public tracker for the exodus, and WP Tavern reported 18,280 websites gone as of November 12, 2024, with Pressable the top destination. Automattic owns Pressable. It's the most precisely counted migration on record, and it was counted by the party that gained from it.

Pricing changes work at larger scale still. Azul, which sells an alternative to Oracle's Java, had Dimensional Research survey 663 Java professionals after the per-employee license change, and reported in July 2024 that 86% of the Oracle Java SE users among them were moving all or some applications off Oracle. It does not say how large that subset was. On acquisitions, CloudBolt surveyed 302 director-level-and-above IT decision-makers at North American enterprises in January 2026 and found 86% actively reducing their VMware footprint, and 56% who had changed strategy two or more times since the Broadcom acquisition. Both vendors benefit from those numbers.

Then there's the counterweight. CrowdStrike's July 19, 2024 update took down an estimated 8.5 million Windows devices by Microsoft's count, the largest IT outage in history. CEO George Kurtz reported the following quarter:

Q3 gross retention was over 97%, down less than half a percentage point.

Complaint volume is not churn volume. After two years of loud VMware anger, Channel Insider's reporting of that same CloudBolt study put the share of respondents fully migrated at 4%. Anger identifies the account. It doesn't schedule the purchase.

Why does evidence beat a keyword alert?

Because the phrases are rarer than the tooling implies, and precision on the ones that fire is bad. Both problems are measurable, and together they explain the gap between 847 results and six. The fix is not a better keyword list. It's reading the post, keeping the link, and remembering what you already acted on.

We counted every comment on Hacker News for July 2026 through the Algolia search API, querying each phrase as an exact phrase (advancedSyntax=true, the phrase quoted, the window set with created_at): 31,299 stories and 314,410 comments, summed from day-scoped queries because a whole-month query returns a non-exhaustive estimate. Across five common intent phrases, the totals were "alternative to" 297, "switched from" 82, "any recommendations" 22, "migrating from" 16, and "looking for a tool" exactly once. All five combined came to 418 matches, 0.13% of the month's comments. That's the honest scale of the haystack, and it's our measurement, not a vendor's.

Stack the AAAI precision rate against that volume. The tools promising that a saved search will fix it are also fighting the platforms. GitHub's search API caps a query at five Boolean operators, which isn't enough to express "leaving vendor X, not joining vendor X, not the vendor's own announcement."

The other half is memory. A switching thread gets cross-posted, quoted, and answered in four places, and without a record of what you already replied to, someone on your team messages a person you already talked to.

That's what LeadSurface does. It watches Reddit, Hacker News, GitHub, Stack Exchange, and Discourse forums, classifies each post rather than pattern-matching it, scores switching signals above the rest, deduplicates by URL per account, and hands you the ones worth a human reply with the source thread attached. A person decides and sends. Nothing goes out automatically.

To test the idea without buying anything, search your closest competitor's name alongside "off", "from", and "instead of" in the communities where your buyers troubleshoot. Filter to the last 30 days.

The buyer already told you what broke, which vendor broke it, and when. The only open question is whether anyone was reading.