Too many low-quality inquiries from Facebook and Google ads, and not enough people to screen them. What should you do?
Fix both ends. On the ad side: switch Meta instant forms to the Higher intent type, turn on phone verification and add one or two qualifying questions; where you can, use ads that click to WhatsApp or Messenger so people land in a conversation; add negative keywords in Google; and send back which leads turned out to be qualified so the platforms learn to find more of them. On the handling side: stop having people scroll through forms one by one, let an AI pick up every inquiry in conversation, and sort leads into A, B and C tiers, nurturing the ones that are not ready instead of deleting them. DeepSales, Singoo Cloud’s AI Sales Agent for exporters, comes with its own CRM that connects the two ends.
Why are ad inquiries so often low quality?
Because ad platforms optimize for exactly what you ask for. If the goal is the number of leads, the algorithm finds people who fill in forms easily, not people likely to buy. Meta describes its default instant form type, More volume, as designed to generate a larger number of leads (Meta Business Help Center). Bots and fake entries add to the problem: Meta says its phone verification feature filters out fake leads such as spam bots and bad actors (Meta Business Help Center).
So “poor ad lead quality” is usually three problems: the wrong optimization goal, a form that is too easy to submit, and nobody telling the platform what a good lead looks like.
What can you change on the ad side?
| Change | What it does | Trade-off |
|---|---|---|
| Use the Higher intent instant form | Adds a review screen before submitting and a note that the business may follow up; Meta suggests it when each lead needs follow-up or further qualification | Delivered only to Facebook and Instagram feeds on mobile, and usually brings fewer leads |
| Turn on phone verification | Confirms the number with a one-time code sent over WhatsApp or SMS, filtering bots and fake numbers | Meta notes it may affect lead volume and cost per lead |
| Add qualifying questions | A form can hold up to 15 custom questions, such as order quantity or destination country | Every extra question loses some people; one or two are enough |
| Use ads that click to message | Clicks open a WhatsApp or Messenger conversation where needs can be clarified in chat | Someone (or an AI) has to pick up every conversation quickly |
| Google negative keywords | Exclude irrelevant searches such as jobs, repairs or DIY | Requires regular review of the search terms report |
Sources: Meta Business Help Center and Google Ads Help. Click-to-message is now mainstream: on its Q3 2025 earnings call Meta said click-to-WhatsApp ad revenue grew 60% year over year (Meta, 2025). Meta’s own advice for chat flows is to start with easy qualifying questions and keep to six questions or fewer (Meta, 2023).
How do you teach the ad platforms to find better leads?
Tell them what happened to each lead afterwards. Both platforms have a mechanism for this:
- Meta: send CRM lead stages back through the Conversions API and set the performance goal to maximize the number of qualified leads. Meta’s fit guidelines include at least 200 leads a month, data uploads at least daily, an optimized stage that happens within 28 days of the lead, and a conversion rate for that stage between 1% and 40% (Meta developer documentation). Since April 2026, new campaigns can only use this goal with a Conversions API integration. In a Meta A/B test across 567 advertisers, instant-form lead ads using this setup saw a 21% lower cost per quality lead (Meta, 2025).
- Google: use enhanced conversions for leads or offline conversion import to report qualified or closed leads, and bid toward qualified leads. Google recommends at least 15 conversions in the last 30 days, daily uploads where possible, and a conversion that happens within 7 days of the ad interaction (Google Ads Help). From 15 June 2026 these uploads move to the Data Manager API.
All of this depends on someone judging, quickly and consistently, whether each lead is qualified. If that judgment waits until a rep works through the forms the next day, the data you send back is late and noisy, and the platform learns the wrong lesson. Which brings us to the handling side.
On the handling side: let an AI qualify in conversation and sort leads into three tiers
Every inquiry should be picked up within minutes, by a person or an AI, and placed into a tier with a few questions:
| Tier | Criteria (example) | What happens |
|---|---|---|
| A | Product, quantity, destination or timing mostly clear; the buyer is an importer, distributor or end user | Handed to the right rep immediately |
| B | A genuine need, but details are missing or the purchase is months away | Short-term reactivation first, then long-term nurturing |
| C | Job seekers, competitors fishing for prices, sales pitches, clearly irrelevant | Marked invalid and kept in the CRM for regular spot checks |
Why not delete tier C? Because judgments go wrong. Missing a real buyer costs far more than the minutes a rep spends on a few invalid leads. Spot-checking a batch of C leads regularly catches misjudgments and sharpens the qualifying rules, and the more accurate the tags you send back to the ad platforms, the better they learn. For designing qualifying questions, see how to auto-qualify inbound B2B leads; for catching inquiries that arrive overnight, see after-hours inquiries: capture first, then score intent.
Your current chatbot forgets things in long conversations. What now?
There are two kinds of forgetting. A flow-based bot follows a fixed question tree and stores only a few fields, so once a buyer goes off script it cannot keep up. Bots built on large language models forget too, for different reasons: research from Stanford and others found that performance drops sharply when relevant information sits in the middle of a long context (Liu et al., TACL, 2024), and a Microsoft and Salesforce study of more than 200,000 simulated conversations found that models perform 39% worse on average in multi-turn conversations than in single-turn ones, and rarely recover once they take a wrong turn (Laban et al., ICLR, 2026).
So good memory is not just a bigger context window; it depends on how the system manages conversation state. It should keep the full chat history, record what the buyer has already said — product, quantity, destination, timing — as a buyer profile that updates as the chat goes on, and produce a summary at handoff. When you evaluate a vendor, test these six things live:
- The buyer gives a quantity in message 3. Twenty messages later, does it still remember?
- The buyer changes the request midway (a different model or quantity). Does it follow the new one?
- The buyer comes back the next day. Does it pick up where it left off instead of starting over?
- The buyer switches language. Is the earlier information still there?
- A rep jumps in mid-conversation. When the AI resumes, does it contradict what the rep said?
- At handoff, is the summary the rep receives complete?
How does DeepSales handle this?
DeepSales is Singoo Cloud’s done-for-you AI Sales Agent for exporters. Our team works with your sales staff to define the qualifying questions, tier criteria and handoff rules, then keeps tuning them after launch.
| Stage | What DeepSales does | Plan |
|---|---|---|
| Lead arrives | Picked up 24/7 across six channels including website chat, WhatsApp, Messenger, Instagram and LINE; every lead lands automatically in the built-in Singoo Cloud CRM with its channel and ad source labelled | All plans |
| Qualifying in conversation | Replies in the buyer’s language, recognizes intent, asks the qualifying questions you define, and tags quality and stage (the AI or a rep can tag) | All plans |
| Tier A | Triggers a handoff to the right rep’s sub-account, with a Feishu card carrying the buyer profile, summary and trigger reason; automatic translation is available when the rep takes over | All plans |
| Tier B | Short-term reactivation before the 24-hour window closes, then long-term nurturing on a schedule set per tag | All plans |
| Tier C | Marked invalid and kept in the CRM for spot checks, never deleted | All plans |
| Back to the ad platforms | Sends the profile of high-quality leads back to ad platforms such as Meta, TikTok and Google, so your ads reach more buyers like them | — |
| Long conversations | Multi-turn, long-conversation memory that never re-asks what the buyer already said, with a summary at handoff | All plans |
The limits are clear: ad accounts, forms and conversion goals are set up in your own ad managers, and negotiation and final decisions always stay with people. To see how ad inquiries get qualified and tiered, book a demo.