Category: Scorecards

We have now scored four AI customer support platforms on what they publish. Somewhere in the middle of doing it, a problem showed up that none of the comparison articles mention, and it makes almost all of them wrong.

Every one of these vendors bills you twice. Seats, then AI. The seat number is easy and everybody prints it. The AI number is the one that decides your year, and here is what the four of them charge:

Vendor	AI price	Unit	Included
Freshdesk	$49 per 100	session	First 500, every plan
Help Scout	$0.75	resolution	None, but you can cap it
Intercom (Fin)	$0.99	resolved outcome	None
Zendesk	Not published	automated resolution	Varies by plan

Four vendors. Four different units. Three of them look like the same kind of number and are not.

Freshdesk looks cheapest and that is not a fact

$49 per 100 sessions works out at $0.49 each, roughly half of Intercom’s $0.99. If you are building a spreadsheet, that is where you stop and Freshdesk wins.

The spreadsheet is wrong, because a session and a resolution are not the same event. A resolution bills when the AI actually finished the job. A session bills when a conversation happened. If your customer asks something the bot cannot handle and gets passed to a human, Intercom has nothing to charge you for. Whether Freshdesk does depends entirely on what it counts as a session, and its pricing page does not say.

That is not us being pedantic. It is the whole bill. On a support desk with mediocre deflection, the gap between paying for attempts and paying for successes is most of the invoice, and it runs in the direction the vendor chose when it picked the unit.

To be fair to Freshdesk, it is the only one of the four that includes anything: the first 500 sessions come free on every plan, including the $19 one. Nobody else gives you a single free resolution.

Zendesk is not in this comparison at all

Its seat prices are published, at $19, $55 and $115 per agent per month. The per resolution rate is nowhere on its site. Enterprise with Copilot says Talk to Sales.

Third party blogs will tell you Zendesk lands somewhere around $1.50 to $2.00 per resolution. Maybe it does. We do not score numbers a vendor has not published, because the point of publishing is that you can hold someone to it, and a figure sourced from somebody else’s blog post is not a commitment anyone made to you.

What we can say is that Zendesk’s consumption charges for App Builder, Action Builder and Voice bill above plan allowances at rates it also does not disclose. So there are at least four unpublished meters running on the largest platform in the category.

Only one vendor lets you put a ceiling on it

This is the finding we did not expect and the one worth taking to a procurement conversation.

Help Scout lets you set a monthly cap on AI Answers resolutions. You pick a number, the meter stops there. It is one line on a pricing page and it is the only published spend control any of the four offer.

Zendesk sits at the other end. Its AI resolution overage has been automatically charged and uncapped since January 2026. If a marketing campaign triples your ticket volume in a week, nothing in the product stops the bill following it.

Nobody writes comparison posts about spend caps because they are boring. They are also the difference between a surprise and a budget.

What to actually ask

Four questions. Put them in writing, because a rep answering out loud is not a published rate.

1. Define the unit. What starts one and what ends one? Get it in a sentence you could hold them to.
2. What happens when the AI fails? If it cannot answer and hands off to a human, does that bill? This single question separates the four vendors more than the headline prices do.
3. Is there a cap, and can I set it? If the answer is no, ask what happens on a bad month.
4. What is included? Freshdesk gives 500 sessions on every plan. If a vendor includes nothing, that is a real difference in a small account.

Where the four sit overall

Pricing is one criterion of six. Across the whole index, weighing model training, data export, subprocessor notice, uptime history and attestations as well:

• Help Scout, 72.8 on 6 of 6 assessed
• Freshdesk, 71.4 on 5 of 6 assessed
• Intercom (Fin), 70.5 on 6 of 6 assessed
• Zendesk, 61.2 on 6 of 6 assessed

Read the coverage figures next to those numbers. Freshdesk rests on five criteria, not six, because we could not verify its status page from anything a reader can open. Its page renders entirely in the browser and serves nothing to somebody without JavaScript, so we left that criterion unscored instead of guessing at it. A composite built on five is a smaller claim than one built on six, and printing both is the only honest way to put them in the same list.

The pricing detail behind every figure above, with a source link on each one, is on the AI customer support ranking. Scoring rules and weights are on the methodology page.

If any of these vendors publishes a rate we said was missing, send us the URL. It goes up as a new dated score with the old one still beside it.

Search for whether AI support tools train on your customer conversations and look at who answers. Crescendo. Freshworks. Fin. CustomGPT. Help Scout. Every result on the first page sells a support tool, and every one of them concludes that the responsible approach happens to be theirs.

So we did the boring version. We opened the AI policy page for three platforms, read what each company had actually committed to in writing, and scored it. Same criterion, same weight, same date. Here is what they say.

	Help Scout	Zendesk	Intercom (Fin)
Model provider trains on your data	No	No	No
Vendor trains its own models on your data	Limited samples only	Yes, with a control	Yes, anonymised
Default position	Off	Not stated	On
Opt out available	Yes	Yes	Yes
Score, out of 100	80	65	60

There are two questions here and most people only ask one

When a vendor tells you it does not train on your data, it is usually answering about OpenAI. That is the easy half. Every one of these three has a written commitment that its model provider will not use your inputs for training, and all three run generative features on zero retention endpoints. If your worry is that a customer’s angry message about a late delivery ends up baked into somebody’s foundation model, you can stop worrying. That question is settled across the category.

The harder question is what the support vendor does with your data for its own models. That is where these three go in completely different directions, and it is almost never what the marketing page is talking about.

Help Scout: off by default, and it says so

Help Scout’s AI transparency page contains the sentence we were looking for and did not find anywhere else: no global retraining on your account data. It goes on to say it does not push customer data back to OpenAI to aid training of the global foundation model.

AI Answers is off by default. Somebody at your company has to go into Beacon settings and switch it on before a single customer conversation touches it. You can ask support to remove your data entirely.

It scores 80 rather than higher for one line in that same page: limited samples of customer data may be used to evaluate and improve feature quality, such as tuning adapters for tone. That is a real carve out and it is honest of them to print it. The page also does not give a retention period, pointing you at the privacy policy and the data protection amendment instead.

Zendesk: more machinery, less clarity about the default

Zendesk documents its process in more detail than either of the others. Before any of its own models see your data, identifier fields like username and email are excluded, free text is de identified using natural language processing, and what is left gets tokenised. There is a customer control over whether it happens at all.

That is genuinely more engineering than Intercom describes, and it earns the extra five points. What Zendesk’s own AI data use page does not say is which way that control points if nobody touches it.

We do not fill in a default a vendor has not written down. It might well be off. If it is, Zendesk should say so on that page, because right now the most cautious reading and the most generous reading are both available to anyone who reads it, and only one of them is true.

Intercom: on by default, and honest about it

Fin may use anonymised customer data to fine tune models by default. You can opt out at any time, and deletion follows within thirty days.

Sixty is not a punishment for the opt out being difficult, because it is not. It is a straightforward setting. The score reflects that the default is on, and defaults are what most accounts run under forever. Nobody audits their AI settings in month fourteen.

Worth saying: Intercom writes this down plainly, in one place, without hedging. That is more than we can say for some of the category, and the 60 sits above the score a vendor would get for saying nothing at all.

What to ask, if you are the one filling in the security review

Four questions, in this order. The first one usually gets a yes and tells you nothing.

1. Does your model provider train on our data? Expect no. Every vendor in this category can answer this cleanly now.
2. Do you train your own models on our data? This is the question. Ask it separately and in writing, because the answer to the first one is often offered as though it answered this one too.
3. If yes, is it on or off when we sign? Get the default in writing. An opt out you were never told about is not really an opt out.
4. What is the deletion window after we turn it off? Intercom publishes thirty days. The other two do not publish a number.

What this does not tell you

These are scores for what each company has published, not for how carefully any of them behaves. We cannot audit a training pipeline from outside. Nobody writing a comparison of these products can, including the vendors writing about each other.

What we can check is whether a company was willing to put its position in public where a buyer can hold it to the words. Help Scout was clearest. Zendesk was most detailed and least clear about the part that matters most. Intercom was direct about a default that costs it points.

This is one criterion out of six. On the full index, weighing pricing, export, subprocessors, uptime history and attestations as well, the order shifts and Zendesk drops further back. The scorecards carry the evidence and a source link behind every number:

• Help Scout, 72.8 overall
• Intercom (Fin), 70.5 overall
• Zendesk, 61.2 overall
• The full AI customer support ranking

Scores here are append only. When any of these policies changes, the new reading lands as a new dated row and the one above stays on the page next to it. If a vendor thinks we have read them wrong and can point at a URL, that correction gets published the same way.

Update, 19 August 2026: Freshdesk has since been scored on this criterion and came in at 50, below all three above. Freddy AI trains Collaborative Models on de-identified data drawn from multiple customers, so one company’s conversations improve a model serving others, and opting out means emailing support rather than changing a setting. The Freshdesk scorecard is here. We also compared what all four charge for AI, and found they do not use the same billing unit: four tools, four units.

Every page that ranks for this comparison is written by somebody selling against both of them. Go and look. The helpdesk startups, the AI agent startups, the migration tools. All of them have a horse in the race, and all of them arrive at a conclusion that happens to leave room for their own product.

So here is a different cut at it. We took the two platforms and scored them on one thing only: what each company publishes where a buyer can check it. Not which is better. Not which we would pick. Just what is written down, on their own sites, without a login or a sales call.

Zendesk came out at 61.2. Intercom, now renamed Fin, came out at 70.5. Both were assessed against all six criteria, so the composites are directly comparable.

Criterion	Weight	Intercom	Zendesk
Pricing transparency	1.5	85	45
Training on your data	1.5	60	65
Data export	1.25	55	35
DPA and subprocessors	1.0	90	90
Status page history	1.0	40	55
Security attestation	1.0	95	90
Composite		70.5	61.2

The gap is almost entirely one number

Both of these products bill you twice. You pay for seats, and then you pay again every time the AI resolves something. In this category the second number is the one that decides your annual bill, because seats are predictable and resolution volume is not.

Intercom publishes it. Fin costs $0.99 per resolved outcome, printed on the pricing page next to the seat prices, the same figure whether you are on Essential at $29, Advanced at $85 or Expert at $132 per seat per month.

Zendesk does not publish it. The seat prices are there, $19 for Support Team, $55 for Suite Team, $115 for Suite Professional. The per resolution rate is nowhere on the site. Neither are the consumption rates for App Builder, Action Builder or Voice, though the pricing FAQ confirms all three bill above plan allowances. Suite Enterprise with Copilot shows Talk to Sales where a price should be.

You can price Zendesk’s seats down to the dollar. You cannot price the year. That single omission is worth 40 points on the heaviest weighted criterion, and it accounts for most of the nine point gap between these two companies.

Neither of them makes it easy to leave

This is the finding that surprised us, and it goes against both of them.

Zendesk ships with data export switched off. To turn it on, the account owner has to contact Zendesk Customer Support and ask. Team plan customers cannot have it at all and are pointed at the API. Once it is enabled the formats are fine, and JSON in particular keeps full comment history, but the first step in getting your own data back is a support ticket. That is 35.

Intercom does better without being good. The self serve export under Reports gives you reporting metadata, not conversations. Browser downloads stop at 10,000 rows. To get the actual message content you need the Conversations API in ninety day slices, or an S3 export. That is 55.

So one of them makes you ask permission and the other makes you write code. If portability matters to you, budget for it either way.

Training on your data: close scores, different shapes

Zendesk edges this one at 65 against 60, and the reasoning is worth spelling out because the headline numbers hide what is actually going on.

Both companies are strong on the third party question. Neither lets OpenAI or any other model provider train on your data, and both run generative features on zero retention endpoints. If your worry is that your customers’ messages end up in somebody else’s foundation model, both have written the answer down.

The difference is what each does with its own models. Intercom says plainly that Fin may fine tune on anonymised customer data by default, that you can opt out whenever you like, and that deletion follows within thirty days. Clear, and the default is on.

Zendesk documents more machinery. Identifier fields like username and email are excluded, free text is de identified with natural language processing, and what remains is tokenised before any training happens. There is a customer control over whether it happens at all. What Zendesk’s own AI data use page does not say is which way that control points if nobody touches it, and we do not fill in a default that a vendor has not written down. Zendesk gets the extra five points for the sanitisation detail and loses some for the silence.

Where Zendesk clearly wins

Two places.

Its subprocessor policy is the better of the two. Thirty days notice before a new subprocessor starts processing data, against Intercom’s twenty, with an email sign up and a written thirty day objection window. Both score 90, for different reasons, and if you are the person who has to answer a vendor questionnaire, Zendesk’s version gives you more to work with.

Its status page keeps ninety days of incident history across a long list of components. Intercom’s, which now lives at finstatus.com, tells you about today and not much else. Ninety days is not twelve months, and Zendesk’s status API will only return active incidents so you cannot pull the archive programmatically, but 55 against 40 is a real difference to anyone who has been asked to justify a renewal.

On attestations they are both excellent and Zendesk has the longer list, including FedRAMP LI-SaaS and CSA STAR AI. Intercom still scores higher, at 95 against 90, because Zendesk’s SOC 2 Type II report only comes out under NDA and its HIPAA coverage needs a paid add on. Breadth is not the same as access.

One thing that will change this

Intercom renamed itself Fin in May 2026. On 15 June, Salesforce signed a definitive agreement to buy it for around $3.6 billion, with the deal expected to close in Salesforce’s fourth fiscal quarter of 2027.

Nothing in the score above is wrong because of that. Every figure was checked against a live page this week. But an acquisition is exactly the moment a disclosure record can move, quietly, in ways nobody announces: a subprocessor list absorbed into a parent company’s, a DPA replaced, a pricing page rebuilt. We will reassess when the terms settle, and the score you see today will still be on the page next to the new one.

What this does not tell you

It does not tell you which product is better. We have not measured resolution quality, how either one handles a messy inbox, or whether your team will get on with the interface. Those things matter enormously and we cannot check them from outside.

What we can check is whether a company is willing to write its terms down in public. A low score is not an accusation. Zendesk’s 45 on pricing means the rate is not published, not that the rate is bad, and for all we know it comes in under Intercom’s for your volume. You just cannot find out without a call, and that is the finding.

Both scorecards carry the evidence and the source link behind every number:

• Intercom (Fin) scorecard, 70.5
• Zendesk scorecard, 61.2
• The full AI customer support ranking

Update, 19 August 2026: Help Scout has since been scored on the same six criteria and came in at 72.8, ahead of both. It publishes its AI resolution rate and lets you cap the monthly spend, which neither of these two do. We also read all three training policies side by side: which AI support tools train on your data. The Help Scout scorecard is here.

Ask a support platform whether it trains on your conversations and the answer you get is usually written by the company you are asking. Ask what it really costs and you get a demo booking. That is the normal condition of this category, and it is why KiteIndex exists. We score what a vendor publishes, we date the score, and we keep the old one.

Intercom is the first entry in the index. Assessed on 17 August 2026 against six published criteria, it scores 70.5, on 6 of 6 assessed. Every number below comes from a page you can open yourself, without a login or a sales call, and every one is linked on the scorecard.

Criterion	Weight	Score
Pricing transparency	1.5	85
Training on your data	1.5	60
Data export	1.25	55
DPA and subprocessors	1.0	90
Status page history	1.0	40
Security attestation	1.0	95

Does Intercom train on your data?

Partly, by default, and you can turn it off.

Fin may use anonymised customer data to fine tune models unless you opt out. Opting out is available at any time and the data is deleted within thirty days. Separately, the third party model providers behind Fin are contractually restricted from training on your data and run on zero retention terms once an output is generated.

The subprocessor posture is genuinely strong and the opt out is real rather than theatrical. It scores 60 rather than higher because the default is on, and the default is what the majority of accounts actually run under. A criterion that scored the best available configuration instead of the shipped one would flatter every vendor in the category equally, which would make it useless.

What Intercom actually charges

All three standard tiers publish a list price: Essential at $29, Advanced at $85 and Expert at $132 per seat per month on annual billing. Fin, the AI agent, is $0.99 per resolved outcome, published alongside the seat prices rather than quoted on a call.

That is worth saying plainly, because most of this category does not do it. Publishing the AI unit price at all puts Intercom in the minority.

Two things keep it off 100. The standalone Fin offering, for teams keeping an existing helpdesk, shows Contact sales instead of a price. And the minimum monthly outcome commitment is illustrated by example, as in 50 outcomes, rather than given as a figure you could budget against. A buyer can estimate their bill. They cannot calculate it.

Can you get your data out?

Some of it, self serve. The rest needs code.

The dataset export under Reports returns reporting metadata such as conversation IDs, timestamps, attributes and metrics, but not conversation content. Browser downloads cap at 10,000 rows and anything larger arrives by email. To export the actual conversations you need the Conversations API, which works in ninety day windows per request, or an Amazon S3 export. Bulk export through the interface handles 200 to 300 conversations at a time as PDF or text.

So you can leave. You cannot leave in one self serve step with your content intact, and that is what the 55 records.

The weakest number, and what would move it

Status page history scores 40, the lowest of the six. The page at intercomstatus.com now redirects to finstatus.com, which publishes current incidents, issues and planned maintenance. We could not find published historical uptime percentages, per component uptime figures, or a durable incident archive with postmortems.

That is a dashboard, not a record. A green light today tells you nothing at renewal, which is exactly when the question gets asked. A published uptime series would move this score further than anything else on the list.

The strongest number

Security attestation, at 95. SOC 2 Type II, ISO 27001, ISO 27701, ISO 27018 and ISO 42001, plus HIPAA and GDPR/CCPA coverage, with the documents self serve from a public trust centre instead of released after a sales call. ISO 42001 in particular is an AI management system certification that most of this category has not obtained.

The Data Processing Agreement scores 90 for similar reasons. It reads in full without signing in, it links a public subprocessor page, and it commits to at least twenty days notice before a subprocessor changes. The only deduction is that change notice is opt in by emailing the vendor, rather than a feed a reviewer could monitor.

A low score is not an accusation

A 40 on status page history means we could not find published historical uptime. It does not mean Intercom is unreliable, and nothing here should be read that way. This index measures disclosure, which is what a vendor puts in public where a buyer can check it. Product quality is a different claim, and only one of the two can be verified from outside.

Scores are append only. When Intercom is reassessed, the new score becomes a new dated row and this one stays visible beside it. If something here is wrong and you can point at a URL, the correction is published the same way, as a new dated score, with the old one still on the page. Most comparison sites cannot offer that, because they overwrote the previous answer.

Update, 19 August 2026: Intercom renamed itself Fin in May 2026, and Salesforce has since signed a definitive agreement to acquire it for around $3.6 billion. We have also now scored Zendesk on the same six criteria. The two side by side is here.

See the full Intercom scorecard, with the evidence and source link behind every score.