Turn your closed-lost list into pipeline

What you’ll do

You'll pull every closed-lost opportunity plus its call transcripts and email threads into a working table, with your CRM as the single source of truth. An AI extraction pass reads what no human will re-read and outputs structured fields: loss-reason category and rationale, the specific product gap, the champion and stakeholders, and any timing promises made on calls. You'll write all of it back to the CRM immediately and trigger the pass on every future closed-lost event. A second pass combines that recorded context with live signals — fundraising, announcements, contract expirations, champion job changes, and your own releases that close a recorded gap — into a likelihood score, a timing bucket, and a drafted email referencing the old blocker and the fresh signal. High scores flow into sequences; everything else becomes a rep task. A self-refreshing query resurfaces unactioned deals on every run, and the aggregate layer turns the whole corpus into an exec report on why you lose.

The steps

1. 01
   Pull closed-lost deals and their raw context into a working table

   Week 1

   Start with a working table — a spreadsheet-style orchestration layer, not your CRM itself — and pull in every closed-lost opportunity along with the raw interaction data behind it: call transcripts from your call-recording tool and the email threads on the deal. The opportunity fields alone are nearly useless; the loss-reason picklist is the least reliable field in most CRMs. The context you actually need lives in what was said on calls and written in threads. Your CRM stays the single source of truth throughout — the working table reads from it and, in step 3, writes back to it.

   • Get transcripts and emails syncing into the CRM first, then pull from the CRM as the one source. Pulling from three systems that disagree is how these projects die in week two.
   • More first-party context is always better: if you have product-usage data from a trial or pilot, pull that too. A deal lost after heavy trial usage is a different re-engagement story than one lost at first call.
   • Scope the first pass to the last 12-18 months. Older deals still matter for the aggregate report in step 7, but the re-engagement motion works best where a human at the account still remembers you.
2. 02
   Run an AI extraction pass over everything a human would never re-read

   Week 1-2

   This is the unlock. An AI research agent reads the full transcript and thread history for each deal — days of human reading — and outputs structured fields: a bucketed loss-reason category plus a free-text rationale, the specific product gap if one was named ('missing the integration with their billing system,' not 'product fit'), the champion and other stakeholders on the deal, and any timing signals buried in the record. Timing signals are the gold: 'ping me after our Series B,' 'no budget until Q3,' 'our current contract runs through June.' Reps forget these within weeks. Reorgs and territory changes erase them entirely — the new rep inherits an account with no memory that a re-engagement date was ever promised.

   • Bucket loss reasons into a fixed category list (price, product gap, timing/budget, lost to competitor, went dark, bad fit) and let the free-text rationale carry the nuance. Fixed buckets are what make the aggregate report in step 7 possible.
   • Extract the specific gap verbatim where possible. 'Needed SSO' is actionable; 'feature gaps' is not.
   • This pass needs no web access — it's pure text analysis over data you already have, which also makes it the cheapest step in the whole system to run.
3. 03
   Write the results back to the CRM — and trigger on the closed-lost event

   Week 2

   Push every extracted field straight back onto the opportunity record. The orchestration table is where you analyze; the CRM is where the results live, because it's where sellers, marketers, and every downstream tool already look. Then change when this runs: not as a quarterly batch, but triggered the moment a deal is marked closed-lost. Enrichment at the event keeps the record accurate while the context is fresh and the transcripts are complete — and it means the loss-reason field on every future deal is filled in by a system that actually read the calls, not by a rep racing to close the record and move on.

   • Keep the enrichment table and the re-engagement plays in separate tables. The enrichment table has one job — keep the CRM perfect. Blending it with campaign logic makes both unmaintainable.
   • Map the extracted fields to dedicated CRM properties (loss category, gap, timing signal, champion) rather than dumping a JSON blob into a notes field. Structured fields are filterable; notes fields are where data goes to be forgotten.
   • Backfill history once, then let the trigger carry it forward. The one-time backfill is also the input for step 7.
4. 04
   Score for re-engagement: recorded context plus live external signals

   Week 2-3

   A second AI pass takes the full extraction output for each deal and layers live research on top: has the company raised a round, made a major announcement, or hit the contract-expiration date mentioned on a call? Has the champion changed jobs — or landed somewhere new where they could buy again? Critically, it also checks your own side: have you shipped a release that closes the exact gap recorded on this deal? The output per account is a re-engagement likelihood score (high, medium, low), a timing bucket (immediate, within 30 days, 3-6 months, hold), and a drafted email that references both the old blocker and the fresh signal — 'when we last spoke, budget was the blocker; saw you just closed your round.'

   • Feed the scoring pass the entire structured extraction, not one summary field. The draft quality depends on the model seeing the loss reason, the gap, the stakeholders, and the timing signal together.
   • The 'we shipped the thing you were missing' signal is the highest-converting re-engagement premise available to you, because it answers the exact objection on record. Wire your release notes into this pass.
   • A champion who left is not a dead end — it's two signals: a new stakeholder to find at the original account, and a warm contact at a brand-new account.
5. 05
   Turn scores into conditional plays — automation only above the threshold

   Week 3

   Do not send to everyone the model scored. High-score accounts with an immediate timing bucket can flow into an automated sequence — the drafted email, reviewed once for voice, referencing the real history. Everything else becomes a task or an alert for the account owner: a Slack notification or CRM task that says 'this account just raised; last objection was budget; here's a draft' and lets a human decide. The score is a routing mechanism, not a permission slip. Closed-lost accounts already know you — a tone-deaf automated re-approach costs more here than in cold outbound, because there's a relationship to damage.

   • Add everyone touched to a CRM campaign the moment they're actioned. That membership drives the self-refreshing query in step 6 and is your only honest way to measure the motion later.
   • Suppress hard-disqualified accounts before anything sends: went-bankrupt, acquired-by-competitor, bad-fit-confirmed. The extractor's loss category gives you the field to filter on.
   • For high-value deals, route the draft to the rep even when the score is high. The automation's job is to make the rep's next touch effortless, not to replace it on the accounts that matter most.

   ↳ Reference: Gifting & Direct Mail — a dormant closed-lost account is a natural gifting target
6. 06
   Make it self-refreshing: the query that never lets a deal fall through

   Ongoing

   The single most valuable ops trick in this motion is the shape of the source query: pull closed-lost opportunities that are not already in the re-engagement campaign. Deals that scored low or landed in the hold bucket aren't discarded — they simply resurface on every scheduled run, get re-scored against whatever signals have changed, and keep resurfacing until a signal fires and they graduate into the campaign. No manual re-running, no spreadsheet of 'check back in Q3,' no lost timing when the rep who made the promise has moved teams. The system remembers so nobody has to.

   • Segment the re-engagement tables by loss reason and give each segment its own cadence. The product-gap segment should re-run monthly — you ship constantly, and every release potentially flips a recorded blocker.
   • Timing- and budget-loss segments can run quarterly; lost-to-competitor works well timed to the competitor's typical contract cycle, when switching costs are lowest.
   • Schedule it and stop touching it. The difference between a re-engagement project and a re-engagement system is whether it runs when everyone forgets about it.
7. 07
   Run the aggregate layer: the closed-lost report your product team will actually read

   Quarterly

   The same extraction that powers re-engagement, run across your entire closed-lost corpus, produces something arguably more valuable than any individual send: an executive report on why you lose. Revenue impact by loss category — 'this much pipeline lost to product gaps, this much to pricing.' Product gaps ranked by the revenue they blocked, so the roadmap conversation becomes 'build this, unlock that' instead of 'sales keeps asking for things.' Competitive intelligence from what prospects actually said about the alternatives they chose. Pricing perception in the buyers' own words. This is the layer where one motion starts serving three teams: sales gets the re-engagement engine, product gets a ranked gap list, and PMM gets positioning intel straight from lost buyers.

   • One of the sharpest GTM teams we track ran ~500 closed-lost deals through exactly this kind of extraction and turned the output into an exec report — and shipped a revenue-blocking CRM integration that the analysis surfaced. That shipped feature is the most concrete, verified win this motion has produced.
   • An honest caveat on expectations: the teams publishing this architecture share the system in detail but no public reply or meeting conversion numbers for the re-engagement sends themselves. Treat the roadmap intelligence as the proven return, and the pipeline lift as the plausible one you'll measure for yourself via campaign membership.
   • Re-run the aggregate quarterly and diff it. Loss categories shifting quarter over quarter — pricing complaints falling, a new competitor rising — is a leading indicator most teams have no other way to see.

What goes wrong

The failure modes that catch most founders.

• You trust reps to remember timing promises

  'Ping me after our Series B' is a real, dated buying signal — and six months of pipeline noise erases it from every human memory, especially across reorgs and territory changes. If the promise isn't extracted into a structured field and watched by a scheduled run, it never happened. Automate the memory; let humans handle the conversation.

• You blend infrastructure and execution in one table

  The table that keeps your CRM enriched and the tables that run re-engagement plays have different jobs, different cadences, and different failure modes. Merge them and every campaign edit risks corrupting your system of record's inputs. One table keeps the CRM perfect; separate tables — ideally one per loss-reason segment — run the plays.

• You batch-enrich months after deals close

  Enrichment as a quarterly cleanup project means every record is stale by the time it's structured, transcripts have drifted out of sync, and the rep who could sanity-check the extraction has moved on. Trigger the pass on the closed-lost event itself. The marginal cost is identical; the data quality is not even comparable.

• You auto-send to everyone the model scored

  A re-engagement email that misreads the history does real damage precisely because there is a history. Threshold hard: automated sequences only above a high-score bar, human-reviewed tasks for everything else, and hard suppression for accounts that closed-lost for disqualifying reasons. The score routes work to the right lane — it doesn't authorize sending.

• You treat the orchestration layer as data storage

  The working table is where you analyze and orchestrate — it is not where results live. Everything the extraction produces gets written back to the CRM, immediately, because that's where sellers, marketers, and every downstream tool already look. A brilliant analysis stranded in a side table is institutional knowledge with a half-life of one quarter.

• You assume transcripts and emails are already in the CRM

  This is the prerequisite most teams miss, and it's the real week-one work. The entire motion runs on call transcripts and email threads being reliably synced to the opportunity record — and at most companies they aren't. Audit the sync before building anything downstream: if the calls aren't captured, there is nothing for the extraction to read.

Want the technical depth?

The chapters with the full reference detail.

• → Gifting & Direct Mail reference— Gifting a dormant closed-lost account is a natural re-engagement play
• → ABM gifting and direct mail playbook— The warmth-manufacturing motion this can feed into
• → Multi-thread deals playbook— The stakeholder map you extract here is the input
• → Orchestration and attribution— Triggers, campaign membership, and honest measurement