How to protect free AI resolutions in a helpdesk
Helpdesk AI free tier abuse means paying model costs for a business that never becomes a customer. The arithmetic, the basic fixes and where each one breaks.
Start with one ticket your AI agent tries to answer, then keep adding lines. This is a scenario and every price in it is an assumption, so swap in your own as you go.
One AI attempt, model and search $0.12
One resolution, at 60% resolved $0.12 / 0.60 = $0.20
Free plan, 100 resolutions a month 100 x $0.20 = $20
One store on 6 free workspaces 6 x $20 = $120 a month
A year of that 12 x $120 = $1,440
The $299 plan it would have bought 12 x $299 = $3,588
One store's cost in a year $1,440 + $3,588 = $5,028Look at line four. Six free workspaces make 600 resolutions a month, and 600 is the workload your $299 plan was priced for. Your model bill is what it would be if the store were paying. Only the $299 is missing.
That's why helpdesk AI free tier abuse hurts more than someone farming free image credits. The free output is worth real money to a business. If a contractor would charge that store $4 a ticket, those 600 resolved tickets save it $2,400 a month, so it has every reason to keep going and very little reason to buy.
That store, month by month:
Say it's a store you were counting on. After your demo in September they loaded their return policy into a free workspace, and the owner said they'd pay once it proved itself. On day 5 the counter hit 100, your upgrade email went out, and you added $299 to next month's forecast. Then the workspace went quiet.
Six weeks later you're reading free transcripts at 10:40 p.m. to check answer quality, and one stops you. A shopper is asking about a refund on refurbished headphones, and your agent is answering with a 30-day return policy you remember from the demo. The workspace is called Support 4, opened from a different email. There's a Support 2, 3, 5 and 6, and on the 1st of the month every one of them refilled. An illustrative first month for that store:
Now say nine stores are doing this, and you have 40 paying customers at $299, which is $11,960 a month.
- 9 x $299 = $2,691 a month you were counting on and won't get, 22.5% on top of what you bill today
- 9 x $120 = $1,080 a month in model costs with no revenue against it
- 9 x $5,028 = $45,252 a year, about 2.8 months of runway if you spend $16,000 a month
Then comes the morning that costs you customers. Right after a big sale, support peaks at many stores at once. Say you can take 40 ticket attempts a minute, paying customers need 36, and those nine stores add 8. The line grows by 4 a minute. Start at 9:30 a.m. and by 11:30 there are 480 tickets waiting, a 12-minute wait.
A shopper is now sitting in the chat widget of someone paying you $299, waiting 12 minutes for an agent you sold as instant. That customer's message asks if your AI is down. You credit three customers a month each, another $897, and you know that 36 would have fit under 40.
So you go after the free plan. Cut it from 100 resolutions to 25 and the store needs 24 workspaces instead of 6 (600 / 25), while the owner of a tiny shop who was going to upgrade next spring gets a quarter of what you promised. That's the decision in front of you: keep paying for businesses that won't buy, or take free AI away from the shops that would have.
Why one business can keep claiming free resolutions
Your free limit is attached to a workspace, and a workspace costs a store almost nothing to open. The help articles that make your agent useful load again in minutes. Tickets keep arriving for as long as the store exists, so the need never runs out.
Nothing in your product flags it, either. Those workspaces are some of the healthiest free accounts you have, with real tickets and high resolution rates, and your onboarding emails congratulate them.
The obvious fix is a tighter limit per workspace, which is how you ended up with 24 workspaces. A limit counts usage inside one workspace, and this usage is spread across several. Something still has to decide whether a new workspace gets a new allowance.
The basic fixes, and where they break
Most teams try these first, and most are worth having. Small shops often run support from a Gmail inbox. Gmail ignores dots and plus-tagged mail reaches the same inbox, so one mailbox can be written many ways.
| Basic fix | What it catches | What it misses | Who it wrongly blocks | What it costs to keep working |
|---|---|---|---|---|
| Email verification and unique emails | Typos, dead addresses, the same address twice | support@, help@ and returns@, all real | Owners whose code lands in spam | Sending and deliverability work |
| Disposable-domain blocks | Known throwaway services | A store's own mailboxes | People using privacy forwarding | A list someone updates forever |
| Gmail dots and plus tags | One Gmail inbox written several ways | A second real mailbox | Owners who use tags to sort their mail | Upkeep as mail services change |
| CAPTCHA | Simple scripts | An owner signing up by hand | People using screen readers or privacy tools | Fees and lost signups |
| Card requirement | Casual second workspaces | A second company card | Small shops who want proof first | Fewer signups |
| One per IP address | Repeat signups on one connection | Staff at home or on a phone | Offices and coworking spaces | Exceptions and tickets |
| VPN blocks | Known VPN ranges | Everything off your list | Remote support teams on company VPNs | Another list to maintain |
Look down the third column. Most of these were built to stop scripts and throwaway inboxes, and the store in this scenario uses neither. It has real staff with real inboxes, plus a company card. Stack the fixes anyway and a real shop has to confirm an email, solve a puzzle and enter a card before it has seen one resolved ticket, and CAPTCHA has limits of its own.
Four more live in your product.
Rate limits catch one workspace flooding you. A store on six workspaces sends about 20 resolved tickets a day (600 / 30), so it never trips one. A real free shop gets slowed on its one busy day, and someone has to keep tuning the numbers.
Usage caps are already in place: that's your 100 resolutions. They work perfectly inside a workspace, which is why the store opened another, and they cut off real shops in a busy month too.
A shorter trial in place of a monthly refill does fix something: those six workspaces stop refilling on the 1st. But a new workspace brings a new trial, and a shop with three tickets a day sees only 42 tickets in 14 days, which may not convince it of anything.
Manual review is how you found Support 4, and it does work, one transcript at a time. At 2,000 free workspaces and 2 minutes each, that's about 67 hours a round. It also means reading a business's customer conversations, so your terms and privacy policy have to allow it. And an agency running support for six real brands looks just like one store with six workspaces.
What to set up this week
Price your own resolution. Divide last month's AI spend by attempts, then divide by your resolution rate. This cost guide covers the rest.
Give free AI its own budget and its own line. Paying customers go first and free tickets wait. A hard cap on free spend keeps a bad month away from the people paying you.
Hold the costly work until the email is confirmed. Don't import help articles or run the agent on an unverified signup.
Write down what each business gets. Say plainly what one free AI allowance means, and put agencies and multi-brand companies in the policy. Turning on free AI and each monthly refill are separate claims, so decide each one.
Decide what a no looks like. Turn the AI off, keep every ticket flowing to the human inbox, and explain it kindly, with a paid plan and a way to reach you.
The fourth step is where your own data runs out. Your records describe one workspace at a time, and whether a new claim on free AI looks like a repeat is much harder to keep getting right on every refill.
Deciding every free AI claim without building it
You don't have to build your own abuse prevention system for this. Before you switch on free AI or refill it on the 1st, one API call to AbuseZero returns a clear allow or deny, under rules you control. Your email rules stop at the address, while AbuseZero checks mailbox aliases, disposable domains and mail routing. A card requirement only proves a card works, and AbuseZero detects repeated payment fingerprints across users, even when new accounts and emails hide the connection. One-per-IP limits give up at the next connection, and AbuseZero sees beyond an IP address with VPN, hosting, network reputation, Tor and proxy intelligence. It recognizes repeat devices across new accounts, even when email addresses keep changing. A single check looks at one thing. AbuseZero's multi-layer AI risk analysis examines 140,000+ signals, detects linked claims and flags abuse that single checks miss. There's a proprietary algorithm too, with AI that checks the claim again. Creating a fresh account isn't enough to erase previous activity.
Your helpdesk still delivers the resolutions and enforces its caps. Eligibility can be once per user or once per calendar month, so the first turn-on and the refill can each have the right limit. Observe mode shows you what would be denied before anyone is blocked. The quickstart walks you through your first benefit.
Frequently asked questions
What counts as helpdesk AI free tier abuse?
One business running several free workspaces to collect more AI resolutions than you meant to give it. One shop on one free workspace, using all of it, is the plan working.
Should a free plan count attempts or resolutions?
Resolutions feel fairer, but you pay for attempts. At a 60% resolution rate, 100 free resolutions is about 167 attempts, so budget for those.
Should I turn off AI on the free plan?
Keep that as a last resort, because free AI is what shows a store what it would be paying for. AbuseZero evaluates each claim when it's made, so the size of your free plan can stay a product decision.
Protect your next free resolution
Start with Builder
Pay as you go. Evaluate your first claim today.
See pricing
Builder, Pro, Scale and Enterprise.
Talk to us about Enterprise
High volumes and custom terms.