The quotes were not lost on price. They were lost on silence.

The company builds CNC machine tools and sells them across India, direct to about four hundred active accounts and through fourteen dealers. Roughly a fifth of revenue comes from spares, service contracts and consumables on an installed base of around twelve hundred units. Revenue had been flat for six quarters while the market grew, and the sales head's explanation was coverage: not enough accounts, not enough feet.

That is almost always the first explanation, and it is almost always the wrong place to start. Adding accounts is expensive, slow to prove and hard to attribute. Before agreeing to it we asked for one export: every quote issued in the last twenty-four months, with its revisions and its final status.

What the export showed

Thirty-one percent of open quote value had no recorded event of any kind for more than twelve days. Not a call, not an email logged, not a revision. Of the quotes that were eventually marked lost, sixty-two percent carried the loss reason price, which is the reason people select when the field is mandatory and nothing else was recorded.

The median time from first quote to booked order was 9.2 days on non-tender business. The best-performing rep, measured on the same basis, ran at 4.1 days. He had been with the company nineteen years and was two years from retirement.

That last figure is the one the promoter reacted to. Not the leak, which he had suspected. The fact that the only working method in the building lived inside one person who was leaving.

The number we agreed

We took two weeks of discovery with read access to SAP, the CRM and the dealer portal, and we spent most of the first week not on models but on a disagreement. Finance and sales did not mean the same thing by an open quote. Finance excluded anything more than ninety days old. Sales did not. Both were reporting honestly and both numbers were being presented in the same board pack.

The definition we settled, and that the CFO signed, was this. Quote cycle: median days from first quote issue to order booking, tender business excluded, revisions of the same requirement counted as one quote, quotes with no event for sixty days marked dead and excluded from the median rather than left open to inflate it.

What we built

Resolution first, because nothing else was possible without it. The same customer existed in SAP as a legal entity, in the CRM as a spelling, in the dealer portal as an abbreviation, and in the service app as a plant address with no company name at all. 4,912 golden records emerged from three systems, with 611 merges. The sales operations team labelled the ambiguous pairs themselves, which took nine hours of their time in total and gave us a golden dataset that scored 97.1 on resolution accuracy.

Then one play. Stall detection: a quote with no recorded event past its threshold surfaces on the sales head's queue with the account's full history attached, the last three interactions, the open service tickets and the previous two quotes and their outcomes. If he approves, a follow-up task is written into the CRM under the owning rep's name.

Nothing was written into any system without his approval. That was not our caution, it was his condition, and it turned out to matter more than we expected: the approvals themselves became evidence. Every time he rejected a surfaced quote, the system learned something about which stalls were real.

What the memory formed

By month four the system had accumulated enough cases to propose patterns rather than just flag events. Three of them survived review and were promoted with written definitions.

• Quotes to first-time buyers in the auto component segment stall at day nine, not day twelve, and recover at half the rate. The threshold for that segment is now different from the rest.
• A revision that reduces price by more than eleven percent without a change in specification precedes a loss more often than a win. It is now flagged at the point of revision, not after.
• Accounts with an open service ticket older than thirty days convert at fourteen points below the base rate. Service and sales now see each other's queues on those accounts.

None of those three were known to anyone in the company as a written fact, though the retiring rep recognised two of them immediately when they were shown to him. The third surprised him. That is roughly the ratio we expect: memory mostly makes explicit what your best people already do intuitively, and occasionally finds something nobody knew.

What it cost and what it returned

Ten weeks from signature to the first play in production. Median quote cycle moved from 9.2 days to 4.6 over six months, measured on the agreed definition and reported by their own finance team rather than by us. The second play, installed base win-back, cost thirty-four percent of the first because the memory it needed already existed.

What did not work

The first version of the stall threshold was a single number across all segments, and it was wrong. It surfaced too many quotes in the spares business, where a nine-day gap is normal, and too few in projects. The sales head stopped trusting the queue in week three and told us so directly, which was the most useful thing that happened in the engagement. Segment-specific thresholds fixed it, and the episode is the reason segment thresholds are now part of our standard build rather than something we add when someone complains.

We also over-invested early in a dealer coverage play that the client did not have the appetite to act on. It was technically fine and it changed nothing, because no one in the room owned the dealer number. We should have established that before building it, and we now do.

Where it stands

The rep retired on schedule. The company did not lose what he knew, because the parts of it that could be evidenced had been written into definitions and patterns that outlast him. His replacement started against a median of 4.6 days rather than 9.2, which is a different job from the one he would have inherited eighteen months earlier.