AI Chatbot Development for Pearson, 30% Lead Conversion

Pearson India's catalogue is wide. PTE Academic test preparation, school curriculum programmes such as MyPedia and Pearson Classroom, the MePro English programme, MyInsights, professional programmes, and a book list that runs from IIT Foundation Physics for class 7 to Civil Services General Studies. A visitor arriving from a Facebook ad for PTE and a visitor looking for a CLAT workbook land on the same site with very different needs.

The existing route was a contact form or a phone call. That put the load on a sales team that had to answer the same qualifying questions repeatedly, and it lost every visitor who was not ready to fill a form. The brief was to answer the first question, which product is right for me, inside the visit, and to capture a contact only once the visitor had a reason to give one.

Learna opens with a plain statement that it is a bot and what it can help with. A visitor can type freely ("books on test preparation, CAT") and Learna extracts the segment, category and subject in one pass, then returns the matching title with its ISBN, price and a link to buy. Where the visitor prefers to be led, Learna offers tappable options at each step: segment, then category, then subject. When a product line needs a human, Learna asks for an email or phone number and hands the conversation to Pearson's sales team with the product interest already recorded.

The bot was deployed on the Pearson India website and on Facebook Messenger from the same build, so a visitor from a paid social campaign and a visitor from Google search reach the same logic and the same product store.

Scroll sideways to follow the three conversation modes

In the quarter, Learna handled 4,127 product conversations: 3,935 on the website and 192 on Messenger. PTE Academic was the largest line at 1,393 conversations, followed by Books at 1,134 and Pearson Classroom at 567. The bot captured 520 email addresses and 160 phone numbers. From those, Pearson's sales team qualified 118 leads and enrolled 35, a 30% conversion rate, with Rs 3,80,972 in revenue attributed to the bot for the quarter.

Traffic into the bot came mainly from Facebook (2,382 conversations tagged to Facebook campaigns), Google Search (727, plus 254 from remarketing lists and 195 from dynamic search ads), and education portals including Inshorts, CollegeDunia and Shiksha. That the bot sat at the end of every paid channel is what made the revenue attributable.

Across the quarter the session store logged 16,798 sessions. 13,043 were abandoned, meaning the widget was opened and closed without a real exchange. Of the 3,755 that became conversations, Learna answered 2,569 to completion and 1,186 hit a question the knowledge base did not yet hold. That is a 69% completion rate on real conversations. Of 679 users who answered "was this helpful", 366 said yes and 312 said no.

The failed conversations split two ways. Some were questions no product bot should answer. The rest were opportunities: recurring questions that became new FAQs in the knowledge base. Reviewing that list monthly is the maintenance model, and it is why the completion rate is a number that moves rather than a number that is claimed.

Any business with a large catalogue and one recurring question at the top of the funnel has the same problem Pearson had. The pattern that worked here is transferable: put the product data in a store the bot can query, let free text run first, route every failure into a guided menu, capture the contact only where a human is needed, and report the bot on conversion and revenue rather than on message counts. The same build now serves as the reference architecture for the AI chatbot development work ibs Fulcro does today, with large language models replacing the intent model where the catalogue is too wide to script.

Let's talk about what a similar engagement could look like for your brand