AI can rank your sales leads. Check the bottom of the list.

Thryv's new platform combines online listings, websites, social posts, paid marketing, revenue attribution and AI Lead Insights. The lead feature pulls together inquiries, estimates which ones show stronger buying intent and sends a summary and recommended action into the owner's existing customer system. MarketBeat reports that Thryv offers free and $99-per-month self-service plans, with separate paid “Boosts” for services such as ads, reviews and website work.

Most owners do not need the whole product diagram to understand the appeal. They need to know who to call before the van leaves the neighborhood. If the score puts a broken furnace above a vague price shopper, it may save a round of reading and return a few minutes to the day.

But ranking is not the same as answering. A score changes attention. The high-scored customer hears back first. The low-scored customer waits, tries another company or disappears. That makes the order of the list a business decision, even when the interface presents it as a helpful hint.

Thryv says businesses using AI Lead Insights generated 40 percent more revenue per client and converted leads at 1.5 times the rate of non-scored leads. Its press-release footnote says the result came from an internal analysis of roughly 6,000 potential customers active in Thryv's Marketing Center and Business Center systems between January 1 and April 22, 2026. Leads from paid and organic sources were traced to revenue recorded in the business system, then AI-scored leads were compared with non-scored leads in the same dataset.

That is more useful than a testimonial with no denominator. It is still a company-run observational comparison. The release does not say businesses were randomly assigned to use scoring, publish the score bands, show whether response speed differed before the score appeared or separate the effect of scoring from the rest of Thryv's marketing and software package.

Read the result as a reason to test the product, not as proof that adding a score will raise any shop's revenue by 40 percent. A company that connects cleaner customer data, follows up faster and buys more marketing may outperform for several reasons at once. The score can be helpful without receiving credit for the whole difference.

High-intent language is often easy to spot: a clear problem, a location, a budget, a date and a direct request to book. Real customers are not always that tidy. Someone using voice-to-text from a noisy basement may send a garbled message. A homeowner may ask one small question before revealing the expensive repair. A good repeat customer may call from a new number with none of the history attached.

Those misses matter because the system can create its own evidence. If a shop replies quickly to the top of the list and slowly to the bottom, high-scored leads get more chances to book. Low-scored leads leave before anyone learns whether the score was wrong. A month later, the conversion report can make the original ranking look smarter than it was.

This is why ongoing monitoring matters more than a clean launch chart. NIST's AI Risk Management Framework treats measurement and evaluation as part of using an AI system, not a one-time gate before deployment. In plain shop language: keep checking the queue after the numbers start changing behavior.

The cheap test is boring. For the first month, let the system score every lead but do not use the score to delete, suppress or automatically abandon anyone. Keep the shop's ordinary response rule. Save the score beside what actually happened: first-response time, booked or lost, quoted value, final revenue and the reason a person changed the priority.

Once a week, open the bottom of the list first. Find the low-scored leads that booked, the high-scored leads that went nowhere and the messages where missing context caused the mistake. Ten minutes spent there is more useful than admiring the top-five card again.

At day 30, decide what authority the score has earned. Maybe it can order the morning call list. Maybe it can flag urgent jobs but never hide the rest. Maybe one source performs well while website chats do not. The point is to widen its role from observed results, not from the confidence of the number on day one.

Ivy Chen sees a real buying case for a two-person shop. A short first-call list at the end of the day can prevent a good inquiry from sitting under spam and half-filled forms. Her boundary is that low score means later, not gone. Until the shop knows where misses cluster, a named person still owns the bottom lane.

Priya Rao wants the reported lift kept next to the method. The internal comparison is useful evidence, but it does not isolate scoring from response speed, channel mix or the rest of the product. In a pilot she would track booked revenue per lead, median response time, low-score wins and high-score dead ends. If the owner cannot see who fell below the cutoff, the owner cannot tell whether the queue improved.

Ivy is trying to make the list usable tonight. Priya is making sure tonight's shortcuts do not become next quarter's blind spot. Both are compatible with using the tool. Neither is compatible with treating “low intent” as a fact about a person.

Ask what the score can see. Does it use the customer's words, call transcript, source, location, prior history and response behavior? Which of those fields are usually missing in your business? A precise model fed half a customer is still guessing.

Ask whether every lead stays visible and exportable. Then ask whether the product explains why a score changed, lets staff correct bad inputs and learns from booked, lost and misranked outcomes without erasing the original record.

Finally, price the whole setup. Include the software tier, marketing add-ons, staff review time and the work required to keep customer records clean. The product has earned its place when good customers hear back sooner and the owner spends less evening time sorting messages. A prettier queue by itself is not the win.