Zoho Says Cheap AI Pilots Get Expensive at Scale

Chandrashekar Lalapet Srinivas Prasanna, Zoho Canada's managing director, says the snag is rarely the model. It is "fragmented systems, lack of integration, and governance maturity." A sales assistant that works in a sandbox still has to hit the CRM, write into ticketing, honour access rules and keep retention. Layer AI onto a split stack and you get data leakage, shadow tools and confident errors in live decisions.

The other leak is price. AI built on rented models and someone else's servers inherits that vendor's capacity and that vendor's bill. "If the answer is 'it depends,' expect pricing challenges," he said. Zoho's in-house Nathu La servers, built with Intel, use 12 to 18 percent less power and cost 20 to 30 percent less to own and run. Match a small model to the job. A cheap trial is not the invoice at 100 seats.

In 2023 Zoho opened data centres in Toronto and Montreal. It sells region-locked processing and residency guarantees, and says its general models are not trained on consumer data and do not keep customer information. Ask for the data-flow diagram: stored, processed, cached, backed up. If any piece sits outside the claimed region, residency is not guaranteed.

Canadian shops will want "trusted, sovereign, workflow-embedded AI." The pilot is not the sticker. Inference is the inflation once the whole company is on it.