Are vendor-arranged reference calls worth taking at all?

Yes, but not for the purpose they are offered for. The reference has been selected for enthusiasm and often briefed beforehand, so asking whether they like the product returns nothing. What a hand-picked happy customer will still tell you honestly is operational detail: how long onboarding really took, who inside their organisation actually logs in a year later, which integration they had to build themselves, and what they wish they had negotiated. None of that requires the reference to criticise the vendor, which is why the answers stay reliable.

What is the single most useful reference question?

Who logs in every week now, and how does that compare to the first month? Almost every accessibility platform is used enthusiastically for six weeks and then narrows to one person, or to nobody, with reports arriving by email that nobody reads. A reference who can name three roles that still open it a year in is describing a tool that stuck. A reference who pauses, or who says the reports get forwarded to the development team, is describing a subscription rather than a programme — and that is the most common failure in this category by a wide margin.

How do I find a reference the vendor did not pick?

Three routes, all legitimate. Search professional networks for people listing the platform in a current or former role — a former user is the most candid reference available and is not bound by any relationship. Ask in accessibility practitioner communities, where tool experience is discussed openly and where the failure modes of each platform are well known. And ask the vendor for a customer with your specific shape — your stack, your estate size, your sector — because a request they cannot satisfy is itself informative about how much of their base looks like you.

What should I ask about support?

Not whether support is good. Ask for the last real ticket: what it was, how long it took, and whether the answer came from someone who understood accessibility or from a generalist reading a script. This category has a specific support failure worth probing — a finding you believe is a false positive, or a scan that cannot get past your login. Ask the reference what happened the first time they disputed a result. The vendors that handle that well have accessibility specialists on support; the ones that do not will have told the customer to file a feature request.

Should I ask about price on a reference call?

Not the number, which they may not be permitted to share and which is not comparable to your quote anyway. Ask about the shape instead: what has happened to their bill since they signed, what was excluded from the original quote that they later had to buy, and what their renewal looked like. Authenticated crawling, single sign-on, extra seats and additional properties are the four things most commonly discovered to be upsells after signature, and a reference will describe that experience freely because it is a fact about their invoice rather than a criticism of the product.

Accessibility Vendor Reference Calls: What to Actually Ask in 2026

Q: How many references do I actually need?

Two per finalist is usually enough if they are the right two: one that resembles you structurally, and one that has been a customer for at least two years. The long-tenured customer is the valuable one, because everything that goes wrong with an accessibility platform goes wrong slowly — the backlog that was never triaged, the score that stopped moving, the team that quietly stopped reading the report. A six-month reference can only tell you about onboarding, which is the part of the relationship that every vendor is good at.

Why Evaluative Questions Fail and Operational Ones Do Not

A reference is a favour the customer is doing for a vendor they like. Ask them to evaluate the product and you are asking them to either praise it, which costs nothing and tells you nothing, or to criticise a company that has been good to them in front of a stranger, which almost nobody will do on a thirty-minute call.

Ask them to describe what happened instead, and the incentive disappears. How long onboarding took is a fact. Which integration they built themselves is a fact. What they added to the contract at renewal is a fact. A satisfied customer will narrate all of it cheerfully, and the facts are what you needed — the opinion was never going to survive contact with your own environment anyway.

The Questions That Return Information

Who logs in weekly now, and who did in month one? The narrowing from "the whole team" to "one person" is the standard decay curve in this category, and where it stopped tells you what the tool is actually for in a real organisation.
How long from contract to first useful report? Not first scan — first report that made someone fix something.
What did you have to build yourselves?Almost every buyer builds one thing: a ticket bridge, a report transform, a deduplication step, a spreadsheet. Which one it was tells you where the product ends.
What happened the first time you disputed a finding? The false-positive conversation is where accessibility support is either specialist or scripted, and there is no middle.
What was not in the original quote?Authenticated crawling, SSO, extra seats and extra properties are the four usual answers.
Did your score ever go up for a reason that was not a fix? A reference who laughs here has been through an engine upgrade or a scope change, and will tell you how the vendor communicated it.
What do you use that you did not expect to, and what did you stop using? The abandoned features are the honest feature comparison.
If you were buying again tomorrow, what would you put in the contract? The most productive question on the call, and the least threatening one to answer.

Ask for the Right Two People

Reference quality is set before the call, by who you ask for. Request two specific shapes and push back if you are offered something else.

The first is structural resemblance: a customer with your stack, your estate size and your sector. A single-brand retailer's experience of a platform is not informative about a forty-property estate on six stacks, and a public-sector reference bought a different product from the one an e-commerce team buys even when the SKU is identical. If the vendor cannot produce a customer who looks like you, that is a real finding about their base and worth noting.

The second is tenure: someone past their second renewal. Everything that goes wrong with an accessibility platform goes wrong slowly. The backlog that never got triaged, the score that flattened at seventy and stayed there, the alert channel everybody muted, the annual price step tied to page count — none of it is visible at six months, and six-month references are the ones vendors have the most of.

The Reference the Vendor Did Not Choose

One unarranged conversation is worth several arranged ones, and there are three ordinary ways to get one. Former users are the most candid people in the market and are easy to find on professional networks, since the tool is usually listed in the role description. Practitioner communities discuss accessibility tooling openly and the characteristic weaknesses of each platform are common knowledge inside them. And a competitor's customer — someone who evaluated your finalist and chose otherwise — will explain the specific thing that ruled it out, which is the fastest route to the question you have not thought to ask.

Keep it proportionate. This is a subscription, not an acquisition, and two good calls per finalist is the point of diminishing returns. The failure to avoid is not insufficient diligence; it is spending a month on reference calls and none on running the tool against your own hardest pages.

Three Answers That Should Change Your Shortlist

"The reports go to the dev team." Nobody on the customer side reads the product. You are being referred to a subscription that renews on inertia, and the same thing will happen to you unless one named person owns the backlog.
"We never got it working behind our login." The authenticated product is where the failures that produce complaints live. A reference who gave up on it is describing a tool that measures your marketing site.
"Our bill went up and nothing changed."A pricing unit tied to page count or seats, meeting a growing site. Model your own three-year number against the same unit before you sign.

What a Reference Cannot Tell You

Detection quality on your site. Their site is not your site, their stack is not your stack, and the automated engines underneath most mainstream scanners overlap heavily anyway — so "it finds a lot" is not transferable evidence. The only way to know what a platform sees on your pages is to point it at your pages, which is what a trial is for. Keep the reference call for the things a trial cannot show you: what month nine looks like, what the renewal does, and what the customer wishes they had put in the contract.

Questions Buyers Ask About References

The vendor will only offer a reference under NDA or with their rep on the call. Is that normal?

A rep on the call is common and it is also the condition that removes most of the value, because the reference will not say anything they would not say to the vendor directly. Ask politely for fifteen minutes without the rep; the request is routine in enterprise procurement and a flat refusal is informative. If you cannot get it, restrict yourself to the purely factual questions — timelines, who logs in, what they built, what was not in the quote — which survive an audience because none of them is a criticism.

How much should a reference call move the decision?

It should rarely pick the winner and it should often remove a candidate. The useful output is disqualifying information — a support model that cannot handle a disputed finding, a pricing unit that punishes how you grow, an authenticated crawl that nobody has got working. Selection should come from your own trial on your own pages, because that is the only evidence about your site that exists.

Is a published case study a substitute for a call?

No, but it is a good source of questions. A case study is written by the vendor's marketing team and approved by the customer's, so read it for the specifics it does mention — the stack, the estate size, the integration, the timeline — and take those into the call as things to verify. The most productive use of a case study is the question 'your case study says eight weeks; what did week nine to week fifty-two look like?'

Should I ask the reference about their legal exposure or complaints?

Ask carefully and expect a limited answer, because active or settled matters are frequently confidential and the reference has no reason to discuss theirs with a stranger. A better version of the question is procedural rather than personal: when they needed to produce dated evidence of what was found and fixed, could the platform produce it, and in what form. That gets you the capability information without asking anyone to disclose their own exposure.

What if the vendor has no customers like us at all?

Treat it as a real data point rather than a disqualification. Being the first customer of your shape means the roadmap has not been shaped by your problems, support has not seen your failure modes, and the pricing model was not designed for your growth pattern — all manageable, and worth trading for a concession. Ask for the discount, the contract terms and the onboarding attention that reflect the risk you are carrying for them.

How do I make sure the call is not just thirty minutes of pleasantries?

Send three of the questions above in advance and say you have fifteen minutes of specifics. References respond well to preparation because it respects their time, and a reference who has thought about 'who logs in weekly now' before the call gives a far better answer than one improvising. Save 'if you were buying again, what would you put in the contract' for the end — it is the question that produces the most generous answer, and it produces it after rapport exists.

The Evidence a Reference Cannot Give You

No reference can tell you what a scanner finds on your pages. Take your own neutral reading first, so that when a finalist's trial produces a number you already know roughly what your site looks like — and you can tell a real detection difference apart from a scope difference.

Then use the calls for the four things a trial never shows: what month nine looks like, what the renewal does to the bill, what the customer had to build themselves, and what happened the first time they disputed a finding.