If your helpdesk or AI vendor is shutting down, will not sign a BAA, or has been acquired, move to a platform that can run inside the helpdesk you keep, sign your compliance paperwork before any customer data flows, and go live one intent at a time while the old tool still answers everything else. Speed depends on the new vendor: Fin says it sets up in under an hour, Lorikeet publishes a time to first live ticket of as little as hours and an average under 29 days, and a simple phone number port must be processed in one business day under FCC rules.

The safest timeline is not the fastest one. It is the one where the old system is switched off last, after the new one has handled real tickets on every intent you care about.

Key takeaways

• Export first, sign second. Zendesk notes that account data exports are not enabled by default and that AI agent tickets cannot be exported, so request your exports on day one.

• No PHI before a BAA. HIPAA requires a business associate contract, and at termination the old vendor must return or destroy protected health information if feasible.

• Do not cancel phone service before porting. The FCC tells consumers not to terminate service with the existing company before starting service with the new one.

• Cut over by intent slice, not by date. Move one topic, check it against real tickets, then move the next.

• An acquisition is a trigger to review, not a reason to panic. Neither of this year's two big acquisition announcements describes a shutdown, and Salesforce says Fin will continue to serve its customers.

How we chose and checked these steps and tools

The eight steps below cover what breaks most often when a support team changes vendors under time pressure: data you cannot get back, automations that only exist as configuration, phone numbers, compliance contracts and the contract exit itself. Every vendor fact was checked on that vendor's own public page on 22 October 2026, and every regulatory statement links to the regulator or the legal text. Where a vendor does not publish something, we say "Not published" rather than guess.

Lorikeet publishes this guide and is one of the tools listed. We have tried to describe every product, including our own, using only what each vendor publishes, and we link to the source for each claim.

The migration checklist at a glance

1. Name the trigger: shutdown, acquisition, compliance gap or renewal

Write down which of the four triggers you are facing, because each one sets a different deadline. A shutdown gives you a hard date. A compliance gap, such as a vendor that will not sign a Business Associate Agreement, means you cannot route protected health information through it at all, so the deadline is now for those flows. A renewal gives you a negotiating window. An acquisition usually gives you time.

Two acquisitions closed this year. Zendesk completed its acquisition of Forethought on 26 March 2026 and says Forethought AI agents work within Zendesk and on any platform. Salesforce completed its acquisition of Fin, formerly Intercom, on 10 September 2026 and says Fin will continue to serve its customers as part of Salesforce's AI Labs. Neither announcement describes a shutdown, so if your vendor was acquired, ask for the product roadmap and contract terms in writing before you decide to move.

Then find three dates in your contract: the end of the term, the notice period for non-renewal and any data retrieval window after termination. Your migration plan has to finish inside the shortest of those, with a margin for testing.

2. Export tickets, conversations and AI agent logs before anything else

Start exports on day one, because they take longer than teams expect and some data cannot be exported at all. Zendesk's help centre says account data exports are not enabled by default; the account owner must contact Zendesk Customer Support to switch them on. The same page says the export tools are not available on Team plans, though every plan can use the REST API, and notes that AI agent tickets cannot be exported. Intercom's guide to exporting conversation data lists several methods, from bulk download through the API to JSON exports to Amazon S3, and notes that internal notes are excluded from transcript exports.

Build a manifest of what you need and in what format:

• Closed and open tickets with full message history, tags and custom fields.

• AI agent conversation logs and outcomes, if the vendor stores them separately.

• Customer and organisation records, and the IDs that link them to your CRM.

• Satisfaction scores and QA results you rely on for reporting.

Keep the raw export somewhere you control. Historical tickets are also your best test set for the new vendor in step 7.

3. Capture macros, flows and SOPs as plain-language policy

Automations are the asset most often lost in a switch, because they live as configuration inside the old tool rather than as documents. Export or screenshot every macro, trigger, routing rule and AI flow, then rewrite each one as a short policy a new agent could follow: when it applies, what information it needs, which systems it touches, and when it must hand off to a person.

This step forces useful decisions. Some flows exist only to work around limits in the old tool and can be dropped. Others encode refund thresholds, identity checks or regulated wording that must survive exactly. Mark those as "must not change" so you can test them line by line later.

Written SOPs also make you portable next time. A platform that learns from your documents rather than from vendor-specific flow builders lowers the cost of the next switch. Lorikeet's how it works page, for example, says it learns from your help docs, SOPs and brand guidelines and acts through your APIs, webhooks and MCP servers, which means the policy you write here is the input, not a throwaway.

4. Move knowledge and integrations without breaking authentication

Inventory every system the old AI agent reads from or writes to, and who owns each credential. Typical lists include the help centre, order or billing systems, identity verification, the CRM and internal wikis. For each, note whether the connection uses an API key, OAuth app or service account, and whether the outgoing vendor created it. Credentials created by the vendor are revoked when the contract ends, so plan replacements early.

Move knowledge as source documents, not as the old vendor's processed index. Re-import articles into your help centre or document store, then point the new platform at that source. If you keep your helpdesk and only replace the AI layer, most of this step is reconnecting rather than rebuilding, and our guide to an AI layer versus helpdesk built-in AI covers the trade-offs of each model.

Test every integration with a read-only call before any customer conversation touches it. A broken refund API found in testing costs an afternoon; found in production it costs customers.

5. Port phone numbers before you cancel anything

If the outgoing vendor carries voice, your phone numbers are the one asset customers notice instantly. The FCC's number porting guide gives two rules worth building into your plan: do not terminate service with your existing company before starting service with the new one, and simple ports must be processed in one business day. The same guide says the old company cannot refuse to port your number even if you owe a balance or termination fee, and that porting from wireline to wireless may take a few days.

Business ports with many numbers, IVR trees or toll-free lines are rarely "simple ports", so request the port early and keep the old routing live until test calls succeed on the new path. Another option is call forwarding: keep the numbers with your carrier and forward the relevant queues to the new voice agent, then port later if you still want to. Document your IVR menus and after-hours rules before the cutover, because they rarely export cleanly.

6. Sign the BAA or DPA before any customer data flows

Paperwork comes before data, not after. Under HIPAA, the business associate contract rules in 45 CFR 164.504(e) require the contract to say that, at termination, the business associate will return or destroy all protected health information it still holds if feasible, and retain no copies. For EU personal data, GDPR Article 28 requires the processor contract to make the processor delete or return all personal data at the end of the services, at the controller's choice, and says a processor shall not engage another processor without the controller's written authorisation.

So for the new vendor, ask for three documents before testing on real tickets: the BAA or DPA, the current sub-processor list, and the security reports you need for sign-off. Lorikeet's trust page says it holds SOC 2 Type II, ISO 27001:2022, HIPAA and GDPR, publishes them on a public Trust Center, and signs Business Associate Agreements for healthcare customers. Its pricing page lists a DPA or BAA as standard on Start and Scale. Whatever vendor you pick, read the signed version, not the marketing page.

7. Run old and new in parallel and cut over by intent slice

Never switch every conversation on one date. Run the new platform against historical tickets first, then live on a narrow slice, while the old system or your team handles everything else. A practical sequence:

1. Replay a sample of exported tickets per intent and review the answers and actions.

2. Go live on one or two high-volume, low-risk intents, such as order status or address changes.

3. Compare resolution, escalation and satisfaction against the old baseline for a week or two.

4. Add the next intent only when the current one holds; keep regulated or sensitive intents with people until the evidence is strong.

This is how Lorikeet describes its own rollout: replay historical tickets and synthetic scenarios in bulk, then deploy topic by topic so the first topics go live while harder ones are still being trained. Carmoola's customer story shows the parallel approach in practice: it ran a proof of concept before switching from automation that resolved about 30% of inbound questions, resolved 40% from day one, and now resolves 60% of all inbound on WhatsApp, chat and email. For a fuller enterprise plan, see our enterprise migration guide.

8. Exit the old contract cleanly

The migration is done when the old vendor confirms in writing that your data has been returned or deleted, not when the new tool goes live. Use the clauses from step 6: request deletion or return under your DPA, and for PHI, return or destruction under your BAA, with a written confirmation that includes backups and sub-processors.

Before you send notice, check that every export from step 2 opened correctly, every integration credential the old vendor held has been revoked, every number has ported or been forwarded, and every help centre widget or script tag pointing to the old vendor has been removed. Then give notice inside the period your contract requires and keep a copy of the termination confirmation with your compliance records.

If the switch is part of a broader rethink, our list of Forethought alternatives covers how other platforms describe their approach.

Tools that help you switch AI customer support vendors

These platforms can each run alongside a helpdesk you keep, which is what makes a parallel migration possible. Every cell comes from the vendor's own page.

Lorikeet publishes this guide and is one of the tools listed. We have tried to describe every product, including our own, using only what each vendor publishes, and we link to the source for each claim.

Lorikeet bills per resolved ticket, with Start at $2,100 a month and Scale at $5,100 a month paid annually, and escalations to a person are not charged. Its trade-off for an urgent switch is that the published average to first live ticket is under 29 days, so a team with a shutdown date next week should plan to keep the old tool or people on most intents for a while. easykind's story shows the keep-your-helpdesk path: Lorikeet's engineering team built a Salesforce Service Cloud integration so it could keep its existing stack. Fin publishes $0.99 per outcome; Zendesk bills AI agents per automated resolution.

What still needs a human during a switch

Some decisions should not be delegated to either vendor. Compliance sign-off on the BAA, DPA and sub-processors belongs to your legal or privacy owner. Choosing which intents stay with people, such as complaints, vulnerable customers or anything clinical in healthcare, is an operations decision; in healthcare, AI should route clinical questions to clinicians rather than answer them. Someone on your team should also own the parallel-run review, reading a sample of conversations each day, because metrics alone miss tone problems and wrong actions. Finally, the go or no-go call for each intent slice should sit with a named person, not a dashboard.

Questions to ask a new vendor before you switch

• Will you sign our BAA or DPA before we send any real tickets, and can we see the current sub-processor list?

• Can you run inside the helpdesk we keep, or do we have to move tickets to you?

• What is your published time to first live ticket, and what does the first week look like?

• Can we replay our exported historical tickets before going live?

• Can we go live one intent at a time and keep the rest with our team?

• What do we pay for escalated or unresolved conversations?

• If we leave you, how do we export our data, and how do you confirm deletion?

The last question matters most: the best time to plan your next exit is before you sign. If you also weigh building your own agent, see Lorikeet vs building in-house. To test the parallel approach on your own tickets, start a 30-day free trial.