The pipeline was off. Not dramatically — the kind of off that is easy to explain away quarter after quarter. Volume looked fine, spend was within budget, and no single metric had collapsed visibly enough to trigger a formal review. But the conversion numbers kept drifting in the wrong direction and nobody had a clean answer for why.
The standard response to this kind of ambiguous underperformance is a diagnostic process that involves pulling data from multiple systems, scheduling alignment meetings with Sales and SDRs, running cross-functional root-cause workshops, and producing a findings report that takes two to three weeks and arrives after the problem has already compounded further.
I tried something different. I fed the available data into Claude across three structured prompts — campaign plan, performance data with disposition notes, and SDR feedback — and ran an AI-assisted diagnostic in a single focused session. What came back was a structured hypothesis that, when checked against the eventual manual audit two weeks later, was directionally correct on every major finding and had surfaced one insight the manual process had missed entirely.
This is the account of how that session worked, what it found, and the rules I now follow every time I use AI for strategic diagnosis.
What I Fed In — and How
The input quality determines the output quality completely. This is the rule that matters most when using AI for strategic diagnosis — and the one most people underestimate. Before the session, I spent 20 minutes preparing three documents: a sanitised campaign plan summary, an anonymised performance data extract, and a structured summary of SDR disposition notes. No raw exports, no personally identifiable information, nothing that violated data handling policies.
What Went In
Three inputs — prepared before the session, not improvised during it
Input 1 — Campaign Plan Summary
What it contained: Campaign objectives, target segments, channel mix, budget allocation rationale, messaging hierarchy, and the conversion targets set at campaign launch. Pasted as structured text — not a PDF, not a slide deck summary, but the actual strategic decisions written out in plain language so Claude could reason against them rather than just describe them.
Input 2 — Performance & Disposition Data
What it contained: Channel-level lead volume, MQL-to-SQL conversion rates, disqualification rates by reason code, cost per qualified lead by channel, and time-in-stage data from the CRM. All figures anonymised — absolute numbers replaced with percentages and indexed values. Privacy-approved before use. The disposition reasons were the critical layer — they revealed not just that leads were failing but what the SDR team was observing when they did.
Input 3 — SDR Feedback Summary
What it contained: A synthesised summary of informal SDR feedback collected over four weeks — common objections heard on calls, patterns in why leads were being disqualified, observations about lead readiness and company fit, and any notable changes in lead behaviour versus the previous quarter. Not raw call notes — a structured synthesis. This was the layer that most diagnostic processes treat as anecdotal. Claude treated it as signal.
What Was Deliberately Left Out
Personal data, company names, individual lead records. The session worked entirely on aggregated, anonymised patterns — not individual cases. This is both a data ethics requirement and, it turns out, a better diagnostic approach: patterns at the system level reveal strategic failures that individual case reviews rarely surface cleanly.
The Three Prompts — Session Walk-Through
Prompt 01
Strategy vs Reality — find where the plan broke from the data
~ 12 min
The first prompt set the diagnostic frame. I gave Claude the campaign plan and the performance data simultaneously and asked a specific question: not “what is wrong” but “where does the actual performance data contradict the strategic assumptions made in the plan?” This framing matters — it focuses the analysis on the gap between intent and outcome rather than producing a generic performance commentary.
Here is our campaign plan [pasted full text]. Here is the performance data for the same period [pasted anonymised metrics table]. I need you to identify every place where the actual results contradict or undermine a strategic assumption made in the plan. Be specific about which assumption is being contradicted and what the data shows instead. Do not summarise performance — diagnose the mismatches.
Six specific strategy-data mismatches identified. The most significant: the campaign plan assumed mid-market accounts would be the primary converting segment. The data showed enterprise-level leads — defined by company size in the disposition data — were converting at 2.3× the rate of mid-market, despite receiving a fraction of the budget and messaging attention. The strategy had been optimised for the wrong segment for the entire campaign period. This was finding one.
Prompt 02
Channel Layer — trace where budget is working and where it is not
~ 18 min
The second prompt went deeper into the channel mix, asking Claude to reason about the full demand generation architecture — not just individual channel performance but how the channels were meant to work together and where the system was broken. I gave it the disposition data by channel and asked it to hypothesise about structural failures, not just performance variance. This is where the retargeting finding surfaced.
Now look at the channel-level breakdown [pasted disposition data by channel]. For each channel, tell me: what role was it supposed to play in the funnel based on the campaign plan, what is it actually doing based on the data, and what structural failure — not just underperformance — could explain the gap. Pay particular attention to any channel whose leads are being disqualified at a rate significantly above the others and hypothesise why the disqualification pattern looks the way it does.
Four channel-level structural failures identified. The most striking: one channel was producing leads that SDRs consistently described as “already evaluated us, didn’t buy.” The disposition data showed these leads were failing at first touch not because of fit issues but because they had high familiarity with the product and were responding to retargeting — but arriving with no new reason to reconsider. The retargeting creative had not been refreshed in over two quarters. Leads were being re-served the same message they had already rejected. This was finding two — and it explained approximately 23% of the total disqualification volume.
Prompt 03
SDR Signal — what the ground-level observations reveal about strategy
~ 22 min
The third prompt brought in the SDR feedback summary and asked Claude to cross-reference it against the findings from prompts one and two — looking for patterns where the SDR experience on calls revealed something about the strategy that the data alone could not show. This is the prompt that produced the ABM insight — a finding that the data pointed toward but only became fully visible when mapped against what SDRs were experiencing on calls with specific account profiles.
Here is a synthesised summary of SDR feedback from the past four weeks [pasted structured summary]. Cross-reference this against the strategy mismatches and channel failures you identified. Where does the SDR experience on calls confirm, contradict, or add nuance to those findings? Are there patterns in what SDRs are hearing that suggest an opportunity or a failure the data does not capture directly?
Three additional insights. The most strategically significant: SDRs reported that a specific account profile — companies that had recently undergone IT infrastructure changes — were showing unusually high buying intent on calls, asking detailed implementation questions rather than exploratory ones. These accounts were not being targeted by any specific campaign or account list. They were arriving through broad demand capture and being treated identically to all other leads. Claude’s hypothesis: this profile represented a natural ABM opportunity — a defined trigger event, a high-intent signal, and a profile specific enough to build a targeted programme around. The data showed the profile. The SDR feedback showed the quality. The strategy had been completely blind to both.
What Claude Found — The Full Hypothesis
Diagnostic Output
Six root-cause hypotheses from one hour — five confirmed by the manual audit
01
Wrong segment receiving primary budget and messaging weight
Campaign optimised for mid-market throughout. Enterprise-profile leads converting at 2.3× mid-market rate and receiving a fraction of the investment. Strategic assumption about primary buyer had been wrong from campaign launch — and had never been tested against live conversion data.
Confirmed · High Impact
02
Retargeting layer broken — stale creative, recycled rejections
Retargeting creative had not been refreshed in two-plus quarters. Leads previously exposed to the same ads, who had evaluated and not converted, were being re-served the identical message with no new proof point, no new offer, no new angle. SDR disqualification data showed these leads arriving at first touch already resistant — not unconvinced but actively uninterested in a conversation they felt they had already had.
Confirmed · 23% of DQ volume
03
High-intent ABM segment present but invisible — no targeted programme
A specific account profile — companies in a post-infrastructure-change state — was arriving through broad channels with disproportionately high purchase intent, asking implementation-level questions rather than exploratory ones. No ABM list, no targeted outreach, no tailored content. These accounts were being processed identically to all others and falling through a follow-up process not designed for their level of readiness.
Confirmed · Net new finding
04
Messaging hierarchy not matching actual buyer objections on call
Campaign messaging led with integration ease and time-to-value. SDR feedback consistently showed the actual first objection was total cost of ownership and migration risk — neither of which appeared prominently in any campaign asset. The message was answering the wrong question before the buyer had asked the right one.
Confirmed · Medium Impact
05
Speed-to-lead SLA consistently missed on highest-intent channel
The channel producing the highest-intent leads — identified by disposition data showing the lowest disqualification rate among reached leads — also had the longest average time to first SDR touch. The highest-value inbound traffic was being worked last. No SLA differentiation by channel existed in the current process.
Confirmed · Process Fix
06
Content syndication vendor producing geography-mismatched leads
One content syndication partner was generating volume in geographies with no active sales coverage. Leads were being worked, failing on qualification grounds unrelated to fit or intent, and contributing to disqualification rate statistics that made the channel appear weaker than it was for in-coverage geographies. A targeting filter, not a vendor replacement, would have fixed this.
Partially Confirmed · Medium Impact
The Retargeting Finding — Why This One Mattered Most
Deep Dive
Why broken retargeting is one of the most expensive silent failures in demand gen
Retargeting is supposed to be the most efficient part of a demand generation programme. You are reaching people who already know the product, who have already shown enough interest to have engaged with your brand at least once, and who are closer to a buying decision than a cold prospect. The economics should work in your favour.
They stop working the moment the creative becomes stale — and the failure is invisible in most dashboards. Retargeting campaigns typically show decent click-through rates even when the creative is old, because the audience recognises the brand and clicks out of familiarity rather than genuine re-engagement intent. The failure shows up downstream: in SDR disqualification rates, in first-call conversation quality, and in the specific objection pattern of leads who “already looked at this” and are not open to revisiting.
What Claude identified specifically
Retargeting creative had been running on the same creative set for two-plus quarters with no refresh — a timeframe in which the product had released new features, pricing had been restructured, and two new case studies had been published. None of this was in the retargeting assets.
The disposition data showed a distinct language pattern in SDR notes for this lead subset: phrases indicating prior evaluation, familiarity without intent, and specific mentions of having “looked at this before.” Claude flagged this as a retargeting signature — not a cold-prospect disqualification pattern.
The fix was not a budget reallocation. It was a creative refresh with new proof points — specifically the use cases and pricing structure that had changed since the original creative was built — and a sequencing strategy that gave previously-exposed leads a new reason to re-engage rather than a louder version of the same reason they had already rejected.
The estimated impact of fixing this alone: recovering approximately 23% of the disqualification volume that was being generated by this channel, without any increase in spend.
The ABM Opportunity — The Finding the Manual Audit Almost Missed
The Unexpected Finding
Why the ABM signal was there all along — just not being looked for
What the data showed
A specific account profile converting at 3× baseline
Accounts with a recent infrastructure change event in their history were appearing in the pipeline at a low but consistent rate — and converting through to opportunity at approximately 3× the rate of the broader account pool. The conversion rate was high enough to be statistically meaningful across the available sample, but the volume was low enough that the pattern had never been formally flagged in any reporting.
What SDR feedback added
These accounts were arriving ready, not just willing
SDR notes for this account profile were qualitatively different. Where most first calls involved establishing the problem, these accounts were arriving with the problem already defined and asking detailed implementation questions. The SDR team had noticed this pattern informally but had no channel to surface it structurally. Claude connected it to the conversion data and made the pattern explicit.
What the strategy was missing
No trigger-based targeting whatsoever
The campaign had no mechanism for identifying accounts matching this profile proactively. These high-intent accounts were arriving by chance — through broad top-of-funnel channels — and being processed identically to cold inbound leads. The opportunity: build an intent-signal-based ABM list targeting companies matching this trigger event, with tailored outreach calibrated to their actual level of buying readiness.
What the manual audit found
The same pattern — surfaced two weeks later
The two-week manual audit identified this account profile as a priority segment recommendation in its final report. The AI diagnostic had surfaced the same hypothesis in the first session. The difference: the manual audit recommendation was accompanied by supporting evidence that took two weeks to compile. The AI hypothesis was a starting point that could have been validated in days, not weeks, had the diagnostic run earlier.
“The ABM opportunity had been in the data for months. What AI did was not find something new — it connected the dots across three data sources in a single session that no single person had the time or vantage point to do manually.”
AI Diagnostic vs Manual Audit — The Honest Comparison
Head to Head
Where each approach wins and where each falls short
Dimension
Manual Audit
AI Diagnostic (Claude)
Time to initial hypothesis
10–14 days from kickoff to findings report
Under 1 hour from input to structured hypothesis
AI advantage
Cross-source pattern recognition
Dependent on analyst bandwidth — often siloed by source
Simultaneous reasoning across all three inputs — connections that human review misses under time pressure
AI advantage
Stakeholder interviews & context
Captures qualitative nuance, organisational politics, unwritten context
Works only from what you feed it — organisational context you don’t capture in text is invisible to the model
Hypothesis accuracy
High — benefits from full context, follow-up questions, live data access
5 of 6 findings confirmed — directionally accurate, some nuance missing
Comparable output
Meeting cost
6–10 cross-functional meetings across 2 weeks
Zero meetings. One session.
AI advantage
Best use case
High-stakes decisions requiring full confidence before action
Fast directional diagnosis to focus the manual audit — or replace it when time is the constraint
Prompts Used
3
Campaign plan, performance data, SDR feedback — fed sequentially
Findings Generated
6
Root-cause hypotheses across strategy, channels, and process
Audit Confirmation
5/6
Five of six hypotheses confirmed by the subsequent 2-week manual audit
Net New Finding
1
ABM opportunity the manual audit almost missed — identified in session one
The Rules I Follow Now — Every Time
Operating Principles
What makes AI diagnostic sessions produce good output — and what breaks them
1
Prepare the inputs before opening the session
Structured, anonymised, categorised. Feeding raw exports or improvised summaries produces unfocused output. 20 minutes of input preparation saves hours of diagnostic iteration.
2
Ask for diagnosis, not description
“Where does the data contradict the strategy?” produces root-cause analysis. “What does this data show?” produces a summary. The framing of the question determines the depth of the answer.
3
Feed inputs sequentially with a connecting prompt
Each prompt should build on the previous one, not start fresh. The cross-source findings — where retargeting data and SDR notes connect — only surface when Claude is reasoning across all three inputs simultaneously in context.
4
Treat the output as hypothesis, not conclusion
Every finding needs validation before it becomes action. AI diagnostic findings are a prioritised list of things to verify quickly — not a report to present to leadership. The speed advantage is in directing the verification, not skipping it.
5
Data privacy is non-negotiable before the session starts
Anonymised, aggregated, approved. Personal data, individual lead records, and commercially sensitive specifics do not go into a third-party AI tool without explicit approval and data handling compliance. The session in this article used only indexed metrics and synthesised summaries — no raw records.
6
Use AI to direct the manual audit, not replace it
The best outcome of an AI diagnostic session is a prioritised set of hypotheses that makes the manual audit 70% shorter because it knows where to look. Not a replacement for the audit — a guide to doing it in a fraction of the time.
The thing that surprised me most: Claude did not just analyse each input separately and combine the results. It reasoned across all three simultaneously — finding connections between the SDR call experience and the channel data that would only be visible if you held both in mind at once. That cross-source synthesis is the specific capability that makes this more valuable than running three separate analyses and comparing notes afterwards.
This use case is also available as
Same content, different formats
📝 This Article
💼 LinkedIn Post
🎙️ Podcast Episode
💼 LinkedIn Post
🎙️ Podcast Episode
Your Diagnosis
Pipeline off and no clean
answer for why?
answer for why?
Book a 15-minute call. Bring your campaign data summary and whatever SDR feedback you have. I will show you how to structure a three-prompt diagnostic session for your specific situation — and what to look for in the output before treating it as something to act on.
Follow for more on marketing, business & strategy
I write about what it actually takes to build a real point of view — no recycled tips, no engagement-bait contrarianism.
Follow on LinkedIn →
Want AI actually working in your marketing?
Book a free 15-minute call and we'll map one AI workflow you can put to work in your marketing this week — no fluff, real tools.
Book an AI Workflow Call →
Read next: I Replaced a $4,000 Agency With 4 Free Tools and Built a Better Website
Claude wrote the code. ChatGPT wrote the copy. Notion held the structure. Netlify shipped it. Here is exactly how four free tools replaced a…
Read the Essay →