Comprint founder Pankaj Sancheti outlines a practical approach to evaluating AI investments

India PR Distribution
New Delhi [India], September 15: Comprint founder Pankaj Sancheti has outlined a practical approach to evaluating artificial intelligence (AI) investments, drawing on more than three decades in technology. His advice centres on four decisions: which task to improve, how to test the system, what it will cost to get the work right and who will take responsibility after deployment.
A quotation assistant illustrates why these decisions matter. An AI-generated document may look ready to send, but the price must be current, the configuration compatible, availability checked and commercial terms authorised. Producing it faster is not enough if the business cannot rely on its contents.
“A quick quotation containing an expensive mistake has failed the business,” Sancheti said.
His career has spanned personal computers, the internet, enterprise servers and the cloud. Each shift, he notes, required customers to make purchasing decisions before they fully understood what they were buying. With AI bringing that challenge into multiple departments, he believes technology providers have a responsibility to help customers become better buyers.
Start with the task
Sancheti recommends beginning with a specific piece of work rather than a broad instruction to “implement AI across the organisation”.
Reducing quotation preparation time is a defined task. So is helping a service engineer find the right troubleshooting history or giving a collections team a reliable view of overdue invoices.
The next step is to establish how that work is done today. How long does it take? Where do mistakes occur? What improvement would justify the investment?
These answers give the buyer a baseline for evaluating a proposal. Without them, Sancheti cautions, a business can spend months comparing technology without establishing what success means.
Test beyond the demo
Sancheti advises buyers to test systems with the difficult inputs their employees actually encounter: scanned invoices with missing fields, catalogues that list the same component under different names and customer requests that contradict earlier emails.
The evaluation should examine what happens when the system cannot resolve a problem. Does it recognise uncertainty? Does it ask for help? Can an authorised employee correct the output and, where necessary, address errors in the source information through an approved process?
For Sancheti, these questions reveal the distance between a convincing demonstration and a dependable system. They also help establish whether the vendor understands the business well enough to support it.
His infrastructure experience informs this approach. Applications depend on servers, storage, networks, power and people. AI adds further dependencies, including the quality of its information, its access permissions and the way its output enters a business process.
The purchase evaluation should therefore follow the work through to completion. Otherwise, a company may save time preparing a document only to spend it checking and correcting the result.
Count the full cost
Sancheti applies the same scrutiny to cost. In infrastructure, the purchase invoice has never represented the entire cost of ownership. Maintenance, downtime, upgrades and eventual replacement also matter.
For AI, he recommends accounting for integration, data preparation, human review and ongoing changes to the system. The useful measure is the cost of completing work to an acceptable standard, not simply the cost of generating an answer.
A low cost per query, he argues, offers limited insight if employees must spend substantial time checking every response.
He also cautions against treating cloud or on-premises deployment as a universal choice. The decision should follow the workload’s volume, variability, response requirements, data controls and available support.
Buyers should ask vendors to explain their assumptions, including how costs and performance would change if usage doubled or fell well below expectations. A credible proposal should address both possibilities.
Make ownership clear
Before deployment, Sancheti recommends identifying a business owner who can assess whether the system improves the work and a technical owner responsible for keeping it dependable.
Employees should know where to report an error. Responsibility for correcting the underlying information should be clear, and someone must have the authority to restrict or stop the system when necessary.
A project champion may secure approval for a purchase. Sustained ownership helps keep it useful after priorities change or people move on.
“The customer has to live with the decision long after the sale,” Sancheti said. “Someone must use the system, maintain it, pay for it, and explain why it was bought.”
He also sees an opportunity for established businesses to organise knowledge accumulated over years of operation: why a particular configuration works, which replacement part is suitable, how a recurring fault was resolved and what a customer was promised.
That knowledge may be scattered across documents, messages and experienced employees. AI provides another reason to organise it, but the people who understand the work must help verify what is captured. Making an incorrect instruction easier to retrieve can spread the mistake faster.
For a first investment, Sancheti recommends choosing one clearly defined process, recording its current performance and running a controlled trial. Businesses should measure time saved alongside errors, correction effort and total cost.
Expansion should follow the evidence. When a trial exposes a weak business case, the finding should be used to change course rather than justify further spending.
“Technology deserves enthusiasm. A purchase deserves scrutiny,” he said.
After more than three decades in technology, his question for AI vendors is direct: “When your system becomes part of our daily work, what exactly are you taking responsibility for?”
The answer, Sancheti believes, should carry as much weight as the demonstration.
About Comprint
Established in Mumbai in 1994, Comprint provides enterprise IT products and services, including servers, storage, workstations and spare parts. Its services include rentals and leasing, maintenance, cloud solutions, and IT refurbishment and recycling. (Comprint)
(ADVERTORIAL DISCLAIMER: The above press release has been provided by India PR Distribution. ANI will not be responsible in any way for the content of the same.)