Sales said some leads weren't getting touched. Operations suspected the dialer. Partners were hearing from customers who said no one had called. I started by asking what had actually happened to those leads—before assuming who or what was responsible.
Al Romero · October 2026 · Fictional trial extract: May 4–17, 2026 · 206 leads / 379 attempts
OfferNest · No calls or texts
18 / 52
More than one in three OfferNest leads received no outreach.
LeadBridge · Long waits
20 / 45
Nearly half waited almost 3–6 hours for the first attempt.
Failed texts · No calls
12
Leads kept getting texts that failed, with no recorded calls.
A question I want answered
Why were Engineering's checks green?
Technical health
Systems can be running, jobs can complete, and APIs can respond successfully.
Customer and partner health
But are leads actually being worked? And would anyone know if an important partner's leads stopped progressing?
What exactly were the green checks measuring, what makes them go red? And why didn't they flag these leads when something appeared to be going wrong?
INITIAL INVESTIGATION What I wanted to understand before blaming the dialer⌄
I first checked what Clever was receiving: contact details, missing information, calling and texting permission, and the possibility that different partners sent leads in different ways. A missing email, for example, doesn't mean a customer can't be called.
Then I compared sources and the individual lead histories. I wanted to know whether an apparent problem was about eligibility, backend handling, the dialer, or follow-up. That distinction matters before changing a process or blaming a team.
The files are complete for this fictional exercise, as Luke later confirmed. They show the outreach activity but not each backend decision, dialer decision, assignment, or actual customer conversation.
02 / What the data shows
Three patterns. Different problems.
Once I compared individual leads and then grouped them by source, three different patterns stood out. Here's what I found, what I think might explain it, and what I'd check next.
What I found
18 of 52
We found no calls or texts for 18 OfferNest leads. Other OfferNest leads were being contacted during the same period.
What I'd check next
Did the backend hold these leads, or did the dialer decide not to contact them? Who was supposed to notice?
Why I took it seriously
Nearly one in three OfferNest leads went without outreach while other leads from the same source were being worked. I wanted to trace what made those groups different.
The exercise logs are complete; the root cause is not yet known.
What I found
12 leads
These leads received 32 texts that failed with a landline error, including 20 later texts after the first failure. None had a recorded call.
What I'd check next
When does the error become available to act on, and can a permitted call be queued in the dialer instead of repeating another failed text?
A gap Luke confirmed
The fictional system does not check phone type and does not automatically switch channels after failed texts. We can propose a targeted recovery path.
The recovery should put a call-eligible lead into the existing dialer's calling workflow—not assume reps work from CRM task lists. The dialer still applies its calling rules.
What I found
20 of 45
Twenty LeadBridge leads waited 170–361 minutes before the first recorded call or text; the other 25 were attempted in 2–12 minutes.
What I'd check next
At what point were the slow leads delayed: partner delivery, backend handling, dialer scheduling, or another step?
Use the healthy comparison
We know when Clever received each lead, and both timestamps are UTC. The faster group gives us a useful comparison without inventing a 15-minute SLA.
25: 2–12 min20: 170–361 min
One chart worth seeing
When OfferNest leads went without follow-up
At first I wondered whether the source had stopped sending leads. It hadn't. New OfferNest leads kept arriving, and some received outreach, but others didn't. The pattern becomes much clearer when plotted by day.
With recorded outreachNo outreach
Each column shows OfferNest leads received that day (UTC). This chart shows a pattern beginning May 11, not proof of an outage or a specific change in the systems.
OPEN AN INTERACTIVE EXAMPLE Follow three leads through their recorded histories⌄
These sample IDs show why the three findings need different investigations. Choose a lead to see its actual recorded activity.
No outreach recorded
L-20110 · OfferNest
+0 minLead received by Clever
Through the end of the extractNo outreach recorded
Observed timestamps shown as minutes after receipt. IDs are from the fictional trial; no personal contact details displayed.
Failed texts, no call
L-20160 · HomeFlow
+0 minLead received by Clever
+12 minText · failed (landline)
+19 minText · failed (landline)
+26 minText · failed (landline)
Observed timestamps shown as minutes after receipt. IDs are from the fictional trial; no personal contact details displayed.
Long wait before first attempt
L-20187 · LeadBridge
+0 minLead received by Clever
+271 minCall · no answer
+283 minText · delivered
Observed timestamps shown as minutes after receipt. IDs are from the fictional trial; no personal contact details displayed.
WHAT I BUILT TO INVESTIGATE Open my investigation tool and supporting views⌄
I built a local investigation tool so I could compare groups, ask follow-up questions, and inspect individual leads without repeatedly rebuilding spreadsheets. It's an investigation aid—not what I'd ask Clever to deploy.
Results view · Click to enlarge
Questions worth asking · Click to enlarge
What I learned about “New”
All 206 leads show “New,” including leads with outreach. Luke explained that this is a legitimate milestone in the sales journey—not an outreach status. I withdrew that concern but kept the need to track early follow-up separately.
What we still don't know—yet
We know which leads received no outreach, which texts failed, and which leads waited hours. What we don't know is why. I still want to know who owns each lead, what each partner expects, and which customers were actually reached.
The finding that changed my priorities
Then I noticed the two sources with problems were also the two largest.
When I compared where the leads were coming from, OfferNest and LeadBridge accounted for almost half of the entire sample. That was the point where I stopped thinking only about missed calls and started thinking about the partners supplying Clever with customers.
OfferNest52 leads
No calls or texts for 18 leads
LeadBridge45 leads
20 leads waited 170–361 minutes
AgentMatch Pro29 leads
HomeFlow27 leads
SellFast Direct27 leads
ListingLoop26 leads
47.1%
97 of 206 leads came from OfferNest and LeadBridge. Their actual ongoing volume and revenue are unknown, but the concentration raises a much bigger business question: what if a partner loses confidence in us?
Counts are from the complete fictional Clever receipt file, not independent records of everything partners tried to send.
What I took away from the investigation
These aren't three versions of the same problem.
OfferNest has leads that never received outreach. Failed texts have no automatic alternative. LeadBridge has a group waiting hours while other leads are contacted within minutes. The causes still need to be traced—not guessed.
But the bigger concern is that these failures can go unnoticed while technical checks look healthy. And when two of the largest sources in the sample are affected, we're talking about protecting partner relationships as well as individual customers.
That changed my priorities: protect customers and partners now, fix the problems at their source, and make the next breakdown visible before someone outside Clever has to report it.
03 / My decisions
Protect the biggest risks. Don't rebuild what we haven't diagnosed.
The investigation changed how I'd rank the work. OfferNest and LeadBridge deserve parallel attention because of the partner exposure. Failed-text recovery is a targeted gap we can start addressing without waiting for an entire dialer rebuild.
01
OfferNest — Find out why 18 leads got no outreach
Trace each lead from arrival through backend rules and dialer decisions. Compare the affected leads with OfferNest leads that were contacted.
Do now: Recover eligible customers, contain any new misses, and fix the confirmed breakdown.
02
LeadBridge — Find out why some leads waited hours
We know when Clever received these leads. Identify where the extra wait was introduced and what response time this partner expects.
Do in parallel: Check incoming leads and fix the cause of unnecessary waits.
03
Failed texts — Give customers a working alternative
Luke confirmed there is no phone-type check or automatic alternative after a failed SMS. Stop repeating texts once the failure can be acted on.
Do early: Route eligible, call-permitted leads into the existing dialer’s calling queue; confirm that a rep makes the follow-up.
Catch a source-wide problem before the partner calls us
I'd compare how many leads Clever received from each source with how many got the expected follow-up. If one source suddenly has a cluster of missed or overdue leads, Operations and Partnerships should hear about it—not just individual reps.
I'd coordinate any partner communication once we've confirmed the scope. Matching what a partner says it sent against what Clever received is another worthwhile safeguard, but we'd need the partner's delivery records for that.
What I'm deliberately not building
No full dialer rebuild. No new CRM. No broad duplicate-merging policy based on matching contact details. No arbitrary retry requirement. First fix what we can demonstrate is going wrong.
ONE MORE STAKEHOLDER COMPLAINT What about the customer who says Clever contacted them twice?⌄
I wouldn't ignore that complaint, but I also wouldn't assume our similar-looking lead records are duplicate inquiries. Some records share a name and phone number but show different property ZIP codes. I'd look at the specific customer's full contact history and the partner's inquiry details before changing calling rules or combining leads.
What I would avoid: a broad duplicate-merging rule or a blanket limit on calls without evidence that either caused the reported experience.
VIEW SEPARATE RISK CHECK Consent questions worth reviewing promptly⌄
Ten SMS events are associated with lead records showing no or missing SMS-consent snapshots. That doesn't establish whether a send was unauthorized. I'd verify the permissions in effect when the texts were sent and escalate any confirmed violations independently of the partner recovery work.
Business impact
What could these missed opportunities be costing us?
We found 30 distinct leads worth recovery review: 18 with no outreach and 12 with failed texts and no calls. We don't know how many sales were actually lost.
I wanted to know how many of these were actually lost opportunities and what they might be costing Clever. But the data doesn't tell me which customers would have converted, so I can't put a reliable dollar amount on the problem.
OPTIONAL SCENARIO · NOT A LOSS ESTIMATE Explore Luke's $500-per-converted-lead assumption⌄
This is a user-selected assumption—not the observed number of lost conversions.
Illustrative value
$1,500
Additional conversions × $500, using Luke's rough exercise assumption. Partner relationship risk is separate and not quantified.
04 / What I'd do
Stop the damage now. Fix the problems. Keep them from happening again.
I wouldn't give the business a 90-day timeline to respond to a problem that's affecting customers today. Ninety days is for proving the fixes are lasting—not for getting started.
First 24–48 hours
Protect customers and partners
Find and recover affected leads, give each an owner, check every new lead from the impacted sources, and raise the partner risk internally.
First seven days
Locate and contain the failures
Trace backend and dialer decisions, find the delay point for LeadBridge, and put manual failed-text call recovery in place.
First 30 days
Fix and verify
Deploy targeted corrections once confirmed, automate eligible recovery, and alert people when the next expected action doesn't happen.
By day 90
Prove it stays fixed
Improve repeatable safeguards and show better timely outreach, actual customer contact, and partner-level performance.
The lasting solution
Every lead needs a next step—and someone who owns it.
I'd connect the leads Clever receives with what should happen to each one, then check whether it happened. If leads are being missed—or failures start clustering within one partner's leads—the right team should know who needs attention and what to do next.
1. Lead receivedPartner and receipt recorded
2. Plan the follow-upEligible? Owner? When?
3. Check it happenedContact, exclusion, or failed path
4. Recover or escalateSomeone acts; partner pattern flagged
What I'd lead versus what Engineering would build
Start with people and a reliable process. Add code where it earns its keep.
I'd work with Sales and Operations to recover affected customers, define who owns each lead and its next action, set sensible partner-specific expectations, and prototype the checks and recovery rules. Engineering would trace and correct the production backend/dialer logic, add missing timestamps, and automate the alerts and permitted recovery paths. I'd write the expected behavior and test it with the teams before calling it fixed.
What I can start with Operations
Identify and work through at-risk leads; run a daily received-versus-followed-up check; agree on ownership, escalation and the first success measures.
What needs Engineering
Confirm the root causes, fix production rules, connect failed-text recovery to the dialer, capture the missing handoffs, and automate reliable alerts.
A practical check I would put in place
What came in—and what happened next?
I'd start with a daily source-by-source comparison while Engineering fixes the problems, then automate it inside Clever's existing workflow. If expected follow-up is missing or late, the report should point to the affected leads and an accountable team.
We can check this with the supplied files
OfferNest · Received vs. followed up
52 received
34 with an attempt
18 with no attempt
Luke confirmed these exercise logs are complete. The missing calls and texts are real in this extract; the reason remains unknown.
Additional safeguard · Needs new data
What partners sent vs. what Clever received
I'd eventually compare partners' sent-lead counts or delivery receipts against Clever's intake records. A lead that never reached Clever won't appear in this dataset at all.
This requires partner-side delivery records; it isn't a finding we can establish from these two files.
Small product demonstration
What happens after a text fails?
Here's how I'd handle a landline-labeled failure for a lead that's still eligible to call. Use Next step to see the proposed recovery flow. It isn't connected to Clever's systems.
Illustrative recovery path
Lead L-20160 · HomeFlow
In the exercise, this lead received three texts marked as failed because of a landline error, and no call. The lead's call-consent snapshot is true.
Before calling, the team still needs to confirm current eligibility and applicable contact rules.
1 of 4
Proposed response
1. Confirm an actionable failure
The system identifies that the text failed because the number cannot receive SMS.
Observed in dataText failure labeled as landline
Problem detected
What I would measure
Three questions the operating team should be able to answer.
Did we follow up on time?
Share of eligible leads attempted within a source's agreed timeframe.
Did we actually reach them?
Confirmed customer conversations, not just attempts or delivered texts.
Are our partners being served?
Problems and recovery by source, with downstream conversion checked over time.
VIEW ADDITIONAL CONTROLS How I'd know a lead or partner needs intervention⌄
Lead-level: No expected next action, no owner, an overdue first attempt, or a failed channel without a permitted fallback should generate an owned item with a due time and escalation.
Partner-level: A concentrated increase in open or overdue items should alert Operations and Partnerships—not wait for the partner to call. I'd also work toward reconciling partner-sent leads with Clever-received leads once source-side delivery records are available.
Engineering's green checks: Keep technical health checks, but add measures for actual customer progression. The metric should point to affected leads and an accountable next step.
05 / Optional: Behind the decisions
A peek at how I worked through it.
I thought it might be useful to share some of the thinking behind these recommendations. I went down a few paths, challenged what the data seemed to say, and changed direction when I learned more. These are a few moments that shaped where I landed.
01Before blaming the dialer, I wanted to know what “normal” looked like.I started with the leads, their contact information, and who was actually eligible for what.⌄
Where I started
Three stakeholders gave three explanations that sounded related. I didn't want to decide Sales or Engineering was wrong before checking whether the leads were even eligible to be contacted in the same way.
What changed
Looking source by source and lead by lead, I found concentrated patterns that didn't fit one simple explanation.
What I decided
Compare affected leads with leads that were handled normally. Trace the two decision points Luke described—the backend and dialer—before assigning blame or recommending a rebuild.
02I used multiple AI agents—and challenged the audit when it didn’t hold up.I wanted reproducible evidence, not a group of models agreeing with one another.⌄
Where I started
I had ChatGPT analyze the CSVs, then ran an independent analysis with DeepSeek. The findings were similar in places, but there were differences in methods and interpretation. I didn't want to decide which one sounded more convincing.
What changed
I brought in another AI agent to audit the analyses—not to force consensus, but to check the numbers against the source. I challenged its first assessment because it hadn't reproduced the work with all the necessary original inputs. The follow-up found some hard-coded 'PASS' claims where real checks should have been, along with interpretation issues.
What I decided
I asked for the original files, hashes, execution evidence, and reproducible calculations. Once I had a better foundation, I stopped trying to audit the auditors and moved back to the product decisions. A correct calculation can still support a misleading conclusion.
03I was suspicious of every lead still being marked “New.” Then I learned what it meant.Luke's clarification changed the diagnosis, not the underlying need for visibility.⌄
Where I started
As someone who has managed leads, I questioned why leads with calls and texts were still marked New.
What changed
Luke explained that New is a real milestone before a realtor match. It isn't intended to indicate outreach progress. I had interpreted it too broadly.
What I decided
Drop the criticism of that field. Keep its original purpose and propose a separate, clearer view of whether leads are being worked and what needs to happen next.
04I built a better investigation tool—then decided to stop building.More functionality was no longer changing the next business decision.⌄
Where I started
Spreadsheets were getting in the way of asking follow-up questions. I directed AI-assisted development of a searchable tool to compare lead groups and inspect histories.
What changed
The tool got more capable with each iteration. That was useful until I started asking myself whether one more feature would change the decision I needed to make for Clever. It wasn't the product they needed.
What I decided
Freeze the feature work. Keep the tool to support my findings and the live discussion. Recommend the smallest operational changes that would actually help Clever recover and protect leads.
The investigation tool helped me test ideas. The proposed customer and partner recovery process is a different product decision.
05I stopped counting individual misses and started thinking about partner relationships.The two biggest sources in this snapshot changed what I would prioritize.⌄
Where I started
I was looking at 18 leads with no calls or texts and another group with failed texts. I wanted to understand the financial impact, but I didn't know how many would have converted.
What changed
I noticed OfferNest and LeadBridge were the two largest sources in the sample. Luke also emphasized how damaging it could be to lose a partner supplying many customers.
What I decided
Investigate both sources in parallel and give partner concentration a central place in the recommendation. I also removed the speculative revenue figure rather than make the economics look more precise than they were.
06I asked myself whether I'd accept a 90-day fix if I were Luke.That changed the plan from a project timeline to an immediate business response.⌄
Where I started
The assignment called for a first-month and third-month plan, so I started organizing the work around those milestones.
What changed
I looked at the scale of the partner exposure and thought: would I be comfortable being told to wait three months while customers were falling through the cracks?
What I decided
Separate immediate containment from permanent fixes. Start recovering customers and checking new leads in 24–48 hours. Use day 90 to prove that the process is more reliable, not to finish the initial response.
I wasn't trying to defend my first answer. I was trying to arrive at a better one.What this exercise reinforced for me
This is a retrospective account of my decisions, edited for readability. Analytical work and tool development were AI-assisted; no production improvements are claimed.