A similar material still needs somebody to approve the change

The proposed alternative feature addressed a familiar marketplace problem. A buyer found material that appeared to fit the project, but the lot became unavailable before the enquiry could proceed. The product team wanted to keep the conversation moving by offering a replacement with similar visible characteristics. Dimensions, appearance and description gave the discovery model useful signals. The difficulty lay in the word replacement. It could make a candidate found through similarity sound like a material already accepted for the project, even though the relevant specification decision had not been made.

We reviewed 120 suggested alternatives with the records available. Sixty-eight required qualified specification review before a conclusion could be reached. Thirty-four contained clear stated differences, and eighteen lacked necessary supporting records. These groups described the state of the review; none represented automatic approval. The sponsor proposed presenting the sixty-eight as equivalent and dealing with objections later. I opposed that step. The fact that an option had reached the review stage was the reason to retain the decision, not a reason to treat the decision as already settled.

Discovery finds a possibility, not an approval

A useful search result can bring an unfamiliar option into view. It can reduce the effort of finding a lot with relevant characteristics and make available records easier to inspect. That is a meaningful product benefit without a claim of equivalence. The buyer's project may depend on requirements that a visible description does not capture, and the appropriate professional judgement may sit outside the marketplace. I wanted the feature to help the buyer reach that judgement with better information, rather than borrow its authority to make the discovery experience look more complete.

The commercial argument deserved attention. Every additional review step can delay a transaction, and some enquiries will end without a purchase. A platform that asks users to repeat avoidable work can lose business for no useful reason. My objection was not to reducing friction. It was to removing the substantive approval while leaving the appearance that it had occurred. We should simplify the collection and presentation of evidence, clarify who needs to decide and avoid unnecessary repetition. Those improvements make a real decision easier; they do not replace it with a similarity score whose meaning has quietly expanded.

The distinction also protects the usefulness of the matching system. If every suggestion is presented as an assured replacement, one unsuitable result can undermine trust in the whole route. If suggestions are clearly candidates for review, users can understand why some are declined without treating the discovery process as a broken promise. That does not excuse careless matching or incomplete records. It sets the product a more precise task: find relevant possibilities, explain the basis and differences, and support the next responsible decision. The task can be evaluated without assuming that every successful search must end in an accepted substitution.

Missing records cannot be replaced by a strong score

The eighteen suggestions with missing records were particularly important. A model could rank them highly because the visible fields looked similar, yet the absence of supporting evidence remained. A high score does not fill that gap. The appropriate response depends on the actual information needed and the people qualified to assess it. The product should make the missing material clear and support a legitimate request for it, rather than infer a favourable answer from the attributes that happen to be present. An incomplete account can look impressively precise when its uncertainty is kept outside the displayed score.

The thirty-four suggestions with clear stated differences needed equally careful handling. A difference did not necessarily establish that the material was unsuitable for every use. It did establish that the product should not silently treat it as the same offer. The buyer might consider a different application or decide that the proposed change was unacceptable. That choice belonged with the relevant project review. I wanted the comparison to preserve the difference, including where it made the alternative less attractive, rather than hide it in pursuit of a higher number of enquiries continuing through the flow.

We spent ₹49 lakh changing the proposition to an explicit alternative for buyer-led review. Available records and stated differences travelled with the suggestion. No option entered an automatic replacement path. Engineering implemented the product behaviour, while quality and legal colleagues reviewed the marketplace's part. Buyers and their qualified advisers retained specification decisions. The programme did not certify any particular material or show that all 120 enquiries were recovered. Its delivered result was a more accurate decision path, with the remaining judgement assigned instead of assumed.

Commercial friction can reveal a real decision

The revised route was less convenient to describe as a seamless transaction. It included questions that could remain unanswered and alternatives that could be rejected. Those outcomes were commercially uncomfortable, but they were part of the decision the buyer actually faced. Hiding them would not make the original material available or establish the suitability of another. It would move the uncertainty to a later point, when the buyer might already have relied on the platform's language. Product leadership needs to recognise when friction is unnecessary administration and when it marks a decision whose consequences deserve attention.

There was still work to do around the review. We could improve the organisation of records, distinguish a request for evidence from a completed approval and help users return to an unfinished comparison. We could learn which information was repeatedly missing and work with sellers to improve their listings. Those changes would make the product more useful without expanding the authority it claimed. The right measure would examine whether the feature helped users make an informed choice, including a decision not to proceed, rather than count every interruption as evidence that a valuable approval step should be removed.

I would now define the boundary between a recommendation and a commitment at the start of a matching programme. Who is the recommendation for, what does it imply and what must happen before someone may rely on it? Those questions should be agreed before a team optimises the rate at which suggested options are accepted. An acceptance measure can reward the wrong behaviour if the product has made approval appear automatic. The design must establish the decision's meaning first, then evaluate how well the system supports that decision within the actual service and professional responsibilities.

For a board, the wider issue is whether a product feature changes who appears to approve a consequential choice. Directors can ask management to identify the point at which discovery becomes a promise and to show the evidence supporting that transition. They need not select materials or inspect individual specifications themselves. They need to understand the boundary of the company's proposition and the cost of respecting it. Our alternative feature became more credible when similarity remained an invitation to examine another offer, with the approval still visible as work the product could support but could not perform on someone else's behalf.