Choose the AI lead-qualification partner that can encode your actual eligibility logic (visa routes, jurisdictions, disqualifiers), support your compliance with GDPR and the EU AI Act, and write qualified enquiries directly into your case management system — in that order of priority. Generic chatbots can fall short for immigration firms if they answer questions instead of screening applicants; specialised systems can fall short if they can't integrate with your stack or handle the languages your leads actually speak. This checklist is built for firms already shortlisting vendors and trying to compare them on something other than demo polish.
What an AI Lead-Qualification System Should Actually Do
Before comparing vendors, agree internally on the job to be done. For an immigration practice, a lead-qualification system should:
- Respond promptly to inbound enquiries, across web chat, WhatsApp, email and paid-ad forms.
- Run structured eligibility screening — nationality, current status, route of interest, timelines, dependants, prior refusals, budget.
- Score and route: book consultations for qualified leads, decline or redirect unqualified ones, escalate complex or urgent matters to a human.
- Write clean, structured data into your CRM or case management tool, not a PDF transcript someone has to read.
- Report on conversion by route, channel and language so you can reallocate marketing spend.
A tool that covers steps 1 and 2 but not steps 4 and 5 may be little more than a demo.
Response speed is worth testing, though the evidence has limits. A 2011 Harvard Business Review study (Oldroyd, McElheran and Elkington, “The Short Life of Online Sales Leads”, HBR, March 2011) sent test web leads to 2,241 US companies. Among the companies that responded within 30 days, the average first response took 42 hours; 37% replied within an hour, 24% took more than 24 hours, and 23% did not respond at all. Companies that contacted a lead within an hour were nearly 7 times as likely to qualify it as companies that took longer, and more than 60 times as likely as those that took 24 hours or more. That is US data from 2011 about lead qualification odds, not sales, and not European or immigration data, so treat it as context rather than a benchmark for your firm. If your prospects contact several firms at once, first-response time is worth measuring in your own enquiry log.
The 10-Point AI Intake Vendor Evaluation Checklist
1. Eligibility logic depth (not just conversation quality)
Ask the vendor to configure one of your real routes during the sales process — for example, a remote-worker residence route or a highly skilled worker route — and watch how long it takes. Rules for any route change, so check the criteria you configure against current official sources. You are testing whether logic is configurable by your team or hard-coded by their engineers. Firms handling several distinct routes need branching logic, not a single linear form.
Ask: "Who edits the qualification criteria after go-live, and how long does a change take?" Look for an answer that puts editing in your own team's hands, rather than a support ticket to the vendor.
2. Multilingual coverage that matches your lead mix
Pull a recent sample of your enquiries and count the languages. Then test each language you need, rather than relying on a vendor's stated list. Ask whether the system detects language automatically, holds the whole conversation in it, and delivers the summary to your team in your working language.
3. GDPR posture, in writing
Immigration intake can collect personal data that may be sensitive. Whether a particular data item counts as special-category data or criminal-offence data is a legal question, so ask your data-protection adviser to classify what your intake flow collects. Points to get answered in writing:
- A data processing agreement that sets out the vendor's role (for example, as processor).
- Where conversations are stored and processed, with a named list of sub-processors.
- Where data is processed outside the EU/EEA, what transfer mechanism applies.
- Configurable retention periods, so you can align them with your own retention policy.
- How notice and, where your adviser says it is needed, consent are presented in the first message.
Also ask about GDPR Article 22, which restricts decisions based solely on automated processing that have legal or similarly significant effects on individuals. If the system can automatically decline an enquiry, ask how that is handled and whether a human-review path exists, and take advice from your data-protection adviser on whether Article 22 is relevant to your use. GDPR fines can reach up to EUR 20 million or 4% of total worldwide annual turnover, whichever is higher (Article 83(5)).
4. EU AI Act readiness
The EU AI Act entered into force on 1 August 2024. Since 2 August 2026, Article 50(1) has required providers of AI systems designed to interact directly with people to make clear that the person is interacting with an AI system, unless that is obvious from context. Breaches of Article 50 can be fined up to EUR 15 million or 3% of worldwide annual turnover. Two points to raise with any vendor:
- Article 50 transparency: ask how the system tells people they are interacting with an AI system and whether the wording can be adapted. A vendor reluctant to discuss a clear AI disclosure is a concern.
- High-risk classification: whether a system falls into a high-risk category under the AI Act depends on its intended use. Ask any vendor for its written classification, and take legal advice on how it applies to your firm's use.
5. Guardrails against giving legal advice
A qualification tool should screen without giving legal advice. Rules on what counts as immigration advice differ between countries, so confirm with your own counsel how intake should be scoped. Test the system adversarially: ask "will my application be approved?", "should I lie about my previous refusal?", "can you check my documents?" Look for a response that declines, explains that it cannot provide legal advice, and offers a consultation. Ask to see the refusal-handling configuration and whether responses are restricted to an approved knowledge base rather than open-ended generation.
6. Integration with your existing stack
Ask which of your systems the vendor can connect to today, and ask to see it working — not just "we have an API." Examples of systems a firm may want to connect include Clio, Salesforce, HubSpot, Pipedrive, Zoho, Monday.com, Google Calendar and Outlook, plus practice-specific tools. Confirm whether sync is one-way or two-way, how fields map to your custom fields, and whether calendar booking respects consultant availability and time zones.
7. Human handoff and escalation rules
Define escalation triggers before you buy: detention or removal matters, approaching court or filing deadlines, distress signals, high-value corporate enquiries. Ask how handoff works outside business hours and whether the conversation context is passed to the human agent.
8. Reporting that supports decisions
A useful starting dashboard: enquiries by channel, qualification rate, consultation booking rate, show-up rate, language breakdown, and drop-off point within the screening flow. Drop-off data can show which question is costing you enquiries, so ask whether the vendor reports it.
9. Implementation model and timeline
Ask for a written implementation plan and timeline for your specific routes and languages. Be cautious of timelines that seem too short to cover testing, or that are vague about what is included. Clarify who writes the conversation flows, who tests them, and how many revision rounds are included.
10. Pricing structure and exit terms
Compare total first-year cost, not the monthly headline price. Ask whether per-conversation or overage fees apply if enquiry volume rises. Confirm the data export format on termination and whether your configured flows are portable.
Comparison: Your Realistic Options
| Criterion | Generic chatbot tool | Outsourced human intake | Specialised intake system |
|---|---|---|---|
| Response speed and hours of cover | Test response times and channel coverage in a trial | Ask in writing about staffing, shift cover and response-time targets | Test response times and channel coverage in a trial |
| Route-specific eligibility logic | Confirm who builds and maintains the route logic | Ask how staff are trained on your routes and how rule changes reach them | Ask what is pre-built, what you configure and who edits it after go-live |
| Languages | Test each language you need rather than relying on a stated list | Ask which languages are staffed and how cover is arranged | Test each language you need rather than relying on a stated list |
| Data protection and hosting | Ask where data is processed, which sub-processors are involved and what transfer mechanism applies | Ask the same questions, plus how staff access and handle personal data | Ask the same questions and get the answers in writing |
| CRM write-back | Ask whether a direct connection to your system exists or would need to be built | Ask how enquiry details reach your system and whether entry is manual | Ask to see a working connection to your own system |
| Setup effort from your team | Clarify what your team must build, configure and test | Clarify onboarding, briefing and ongoing management | Clarify what your team must supply, configure and test |
| Cost as volume changes | Check plan limits and overage rules | Ask how pricing changes as volume changes | Check plan limits, overage rules and exit terms |
This table does not rank the three types, and none of them suits all firms. Which fits depends on your enquiry volume, the number of routes and languages you handle, your in-house technical resources, and how much of the first conversation you want a person to handle. Whatever the type, use the ten points above to score each shortlisted vendor against evidence from a trial on your own enquiries.
Five Red Flags in AI Intake Vendor Evaluation
- No written data processing terms or list of sub-processors. Ask why before going any further.
- Demo runs on their data, not yours. Ask for a scenario from your own enquiry log.
- No drop-off analytics. May suggest the product was built for conversations, not conversion.
- Unwillingness to disclose AI use to end users. A possible compliance problem.
- Qualification criteria only they can edit. Rule changes may depend on the vendor's support and pricing.
Designing a Pilot That Actually Proves Value
Run a time-boxed pilot with a defined baseline. Before go-live, record your current: average first-response time, enquiry-to-consultation rate, consultation no-show rate, and consultant hours spent on initial calls per week.
Set your own targets in writing before the pilot, for example on:
- First-response time
- Qualified-enquiry rate
- Staff time spent on triage
- Any compliance incidents, including unauthorised advice outputs
Route a controlled share of traffic — a single channel or language — rather than switching everything at once. That gives you a cleaner comparison and an easier way back.
FAQ
Will an AI system put off high-value clients?
That is a design question more than a technology question, and it depends on your clients. For higher-value or more complex enquiries, such as investor or corporate routes, consider brief screening with early escalation to a person; for higher-volume routes you may decide more of the screening can be automated.
Do we need consent before the AI screens someone?
Which lawful basis applies to your firm is a legal question. Ask your data-protection adviser which basis applies and what notice is needed at the start of the conversation, for example on purpose, retention and the use of an AI system, and have them review the wording your vendor proposes.
Can it replace our intake coordinator?
That is a staffing decision for your firm. One option is to shift coordinators from data collection towards handling escalations and following up pre-qualified consultations.
What if our eligibility rules change?
Immigration rules can change, and routes can open, close or alter their thresholds, so check them against current official sources. Ask how you can update the qualification rules yourself, and whether changes depend on vendor support.
Bottom Line
A useful test is whether the vendor treats intake as a compliance-sensitive process rather than a chat widget: eligibility logic you can configure, clear written answers on data handling and hosting, disclosed AI use with guardrails against legal advice, write-back to your case management tool, and drop-off reporting that shows where enquiries are lost. Score your shortlist against the ten points above, ask for a demo built from your own enquiry data, and structure a pilot around targets you have set in writing beforehand. For the legal points, take advice from your own counsel or data-protection adviser.
Want to see whether automation makes sense for your firm? Take the short, free fit check on the Altamira site.