Boring is the differentiator in AI customer support
"Our AI agent has 99% uptime" means nothing to a customer. They don't care whether the server was up — they care whether their problem got solved without repeating themselves four times — and most of the metrics companies lead with answer a question nobody actually asked. Figuring out where this market gets won meant doing the same exercise you'd do for any positioning problem: segment the field, target the buyer who actually checks, and position against what the rest of the field is doing instead.
We ran this exercise on our own competitors first, honestly, before turning it on ourselves — it's easier to see the pattern in other companies' marketing than in your own.
Segmentation
Split the field by what a vendor's marketing is actually optimizing for, and two segments fall out cleanly. One segment optimizes for the demo: one happy path, clean data, no edge cases, a smiling case study. The other optimizes for what happens when the order doesn't exist, the API times out, or the customer is lying about the reason for a refund — the scenario that almost never shows up in a demo, because it's the one most agents in the first segment fail.
"I understand how frustrating that must be," followed by nothing actually fixed, is the tell of the first segment. An agent that never says "I don't know" isn't confident, it's untested — the same tell, worn differently. We started keeping a private list of these tells specifically because they're so consistent across vendors: reassuring language with no action attached, uptime and speed metrics with no resolution-quality metric alongside them, and case studies that describe volume handled rather than problems actually resolved.
The second segment is smaller, harder to demo, and — in our experience talking to buyers who've been burned once already — the one that gets the renewal. It's also, not coincidentally, the segment that talks about escalation design and failure rates unprompted, because a vendor that's actually built for the hard cases has something concrete to say about them.
Targeting
The buyer worth targeting isn't the one impressed by the demo. It's the one who already knows, even if they don't say it out loud, that what they actually need isn't a demonstration of intelligence — it's a demonstration of what happens when the agent is wrong, because that's the scenario that keeps them up at night, not the happy path. That buyer responds to a different pitch than "replace your support team": the team stops starting every day at zero, because the agent already resolved the predictable third of the queue overnight, and the humans start their day on the cases that actually need judgment.
Targeting that buyer means leading with exactly the failure-mode conversation the first segment avoids. In practice, that's meant walking prospects through a case where our own agent got something wrong early on, what caught it, and what changed afterward — a conversation that would be a liability in a pitch aimed at the demo-impressed buyer, and a credibility builder with the buyer who's actually going to check.
The same buyer is the one who takes churn seriously as a moment, not a form. "I want to cancel" shouldn't trigger an immediate "sorry to see you go" — it's often the last real moment of attention before someone leaves, and routing it straight to an offboarding form throws that moment away instead of using it. A vendor whose agent treats that moment as worth a real conversation is signaling something to exactly the buyer worth targeting, whether or not either side calls it that.
Positioning
"AI-powered" on a landing page has stopped meaning anything, because it's a claim nobody can check. "Here's exactly what it's allowed to decide on its own, and here's what it always escalates" means something, because it's checkable — and that's the actual axis this market gets positioned on. Not smartest model versus less-smart model. Checkable versus unverifiable.
We tested this directly on our own site: a page that made general capability claims versus a page that specified, in plain language, exactly what the agent decides autonomously and exactly what it always routes to a human. The specific page converted worse with visitors who wanted to feel impressed quickly, and converted better with visitors who stayed on the page longer and asked harder questions in the resulting sales call — which, given who we're actually trying to reach, was the right trade.
The position worth holding is the boring one: policy engines, audit trails, escalation design, and cost-per-resolution, reported honestly instead of rounded up. Boring, provable claims are harder to fake than an impressive demo, which is exactly why they're worth more to the buyer who's actually going to check — and why we've stopped trying to compete on the axis where the checkable-but-boring claim will always lose to a flashier, unverifiable one.
What this looks like from the buyer's side
We interviewed a handful of buyers who'd already gone through a failed AI support deployment before evaluating us, specifically to understand what the demo-impressed version of the sale had actually looked like from the inside, and what changed the second time around.
Every one of them described the same arc: an initial demo that handled a curated set of questions flawlessly, an early rollout that performed reasonably well on the easy majority of tickets, and then a specific incident — usually involving money, occasionally involving a very unhappy customer on social media — that exposed how little anyone actually understood about what the agent did when it was wrong, because nobody had asked that question before signing.
The second time around, every one of them said the evaluation process changed shape entirely: fewer requests for a live demo of the happy path, more requests for a written description of the escalation logic, the audit trail, and a real example of a failure and its fix. None of them asked for this the first time. All of them asked for it the second time, which is roughly the most direct evidence we have that the checkable-versus-unverifiable distinction isn't an abstract framework — it's something buyers learn the hard way, once, and then apply for the rest of their career.
That pattern is also why we've stopped worrying that leading with the boring, checkable claims costs us deals with buyers evaluating for the first time. It costs us some of those deals, honestly — the ones still optimizing for the impressive demo. It wins a disproportionate share of the buyers who've already been through the arc above, and those buyers tend to be better customers besides, because they already understand what they're actually buying.
Common questions
Isn't "boring" a hard thing to market? Doesn't it undersell what the product can actually do? It undersells it to exactly the audience we're not targeting, and that's fine. To a buyer who's evaluated one of these products before, a specific, checkable claim about failure handling reads as far more impressive than a vague capability claim — it's just a different kind of impressive, one that shows up in due diligence instead of in a first demo.
How do you avoid boring positioning just becoming an excuse for a worse product experience? The positioning has to be downstream of something real, or it's just a different kind of marketing trick. We only lead with checkable claims we're actually willing to have checked — published failure-mode examples, real escalation-rate numbers — and the discipline of being willing to publish them is what keeps the positioning honest rather than becoming boring-as-aesthetic instead of boring-as-substance.
Do smaller or newer vendors have a harder time using this positioning, since they have less of a track record to point to? Somewhat, but less than expected. A newer vendor can still show exactly what the agent is allowed to decide autonomously versus what it always escalates, and can still walk through a real failure case from its own limited history — the depth of the track record matters less than the willingness to be specific and checkable about what does exist.
How do you measure whether this positioning is actually working, versus just feeling more honest? Sales-cycle length and quality of the questions asked during evaluation, tracked separately from close rate. We've seen sales cycles get slightly longer with this positioning — buyers ask more questions, which takes more time — and seen the resulting customers churn less, which is the trade we're optimizing for even though it doesn't show up in a simple close-rate number.
Why we still lose some deals, and we're fine with that
It's worth being honest that this positioning doesn't win every deal, and being specific about which ones it loses, because pretending a positioning strategy has no downside is its own kind of the unverifiable-claim problem this whole piece argues against.
We lose deals to vendors whose demo is more immediately impressive, particularly with buyers under time pressure to make a decision quickly and without the bandwidth to ask the harder questions we're trying to invite. That buyer exists, is a real segment, and a pitch built around checkable claims about failure handling genuinely serves them worse in the moment than a pitch built around an impressive-looking demo of the happy path. We've made peace with not winning that buyer, but we don't pretend the loss isn't real.
We also occasionally lose deals to vendors who make the same kind of checkable, boring claims we do, but with slightly better numbers on a specific dimension a buyer cares about most — which is exactly the outcome we'd want this kind of positioning to produce industry-wide, even when we're not the one who wins that particular comparison. A market where vendors compete on checkable failure-rate numbers instead of unverifiable capability claims is a better market for buyers even in the deals we personally lose in it.
The deals we don't lose, and the ones that matter more to how the business actually performs over time, are the ones where a buyer specifically asks the failure-mode question early in the process. We win a large majority of those, not because our failure rate is unusually low compared to competitors — we don't actually know that, since most competitors don't publish comparable numbers — but because we're one of the only vendors willing to answer the question directly and specifically rather than redirecting to a capability claim instead.
That's the actual bet behind this whole approach to positioning: not that boring, checkable claims win every deal, but that they systematically win the deals worth winning most — the buyers who ask hard questions, understand what they're evaluating, and are less likely to churn after a single bad incident because they went in with accurate expectations rather than an inflated one built on an impressive demo.
We revisit this bet every couple of quarters against actual outcomes — win rate specifically among buyers who ask the failure-mode question, churn rate among customers who signed after that kind of evaluation versus those who didn't — and so far the numbers keep supporting the same conclusion the framework predicted going in, which is the main reason we haven't second-guessed the positioning even in quarters where overall growth felt slower than we'd have liked.
In practice
If you're evaluating vendors in this category rather than building one, the single most useful question to ask isn't about the model. It's "show me a case where this got something wrong, and what changed afterward." A vendor that can't answer that concretely hasn't been tested enough to know, and a vendor that answers with a real story is telling you more about how the product will behave under pressure than any benchmark on the page.
The same question works just as well turned inward. We ask it of ourselves on a schedule now, not just when a prospect asks it of us — pick the most recent failure worth talking about, and make sure there's a real answer ready for what changed afterward, before anyone outside the team asks. A vendor that only prepares that answer reactively, in response to a sales question, is one incident away from not having a good one ready.