Stage 07 in the Paitho pipeline is labeled HUMAN. Not AI, not hybrid. Human.
Contact enrichment — finding the right person at a qualified company and pulling their verified contact information — is the one stage we will not automate, and we have been asked about this decision more than any other in the product.
Here is the reasoning. It is not what most people expect.
The obvious objection
Enrichment is a mechanical task. Find the company. Identify the role. Pull the email. API providers can do this in under a second. Why require a human to click through?
The objection is right about the mechanics. It is wrong about what the human is actually doing.
When an operator clicks through Stage 07, they are not retrieving data. They are making a strategic decision: Is this company worth spending enrichment credits on?
That decision has several components that the model cannot make on their behalf.
What the human is deciding
First: does the research hold up on inspection?
By the time a lead reaches Stage 07, it has passed through six prior stages of AI processing. It has been discovered, audited, researched, qualified, signal-extracted, and briefed. The model is confident it is a good lead. The model is sometimes wrong.
An operator who clicks through Stage 07 and reads the lead summary before approving enrichment has a second look at the underlying research. They catch things. A qualified company whose key signal turned out to be 14 months old. A contact role that sounds right but would not actually buy this product. A company that passes the ICP filter numerically but the operator knows has a specific relationship with a competitor that makes the outreach pointless.
The human review step at Stage 07 is not primarily a credit-gate. It is a final research check.
Second: is the outreach timing right?
Models do not know about a conversation that happened last week. The operator does. If an operator spoke to this company at a conference three months ago and the conversation ended with "maybe next year," a cold email right now is not outreach. It is a different conversation, and probably not one that should be automated.
The operator knows the account context. The model does not.
Third: what budget is this against?
Enrichment credits cost money. On managed credits at the $100 bundle, one enrichment costs roughly $0.014. That sounds small. At scale — 500 leads reaching Stage 07 — it is $7, which is real but manageable.
The real budget question is: if enrichment is automatic, the operator loses a forcing function to think about which leads are actually worth pursuing. The credit is tiny. The discipline of deciding is not.
The cost argument people make and why it does not hold
The most common pushback from operators: "If I have to click through every lead, I lose the time advantage."
The time math does not support this objection. Stage 07 in the current interface presents the operator with the lead summary, the qualifying signals, and the proposed contact role. The decision is binary: approve enrichment or skip. An operator spending eight seconds on this decision is working at the pace of 450 leads per hour. At realistic batch sizes — 50–100 leads per session — Stage 07 takes six to twelve minutes.
The pipeline queues leads efficiently. The operator reviews Stage 07 in a batch after qualifying leads are surfaced. It is not a constant interrupt.
The eight seconds an operator spends deciding whether to enrich a lead is not overhead. It is the moment they confirm they know what is in their pipeline.
Six to twelve minutes to make the enrichment decision on 100 leads is not a time cost. It is a review cost. The distinction matters. Time spent reviewing is time spent knowing what is in your pipeline. That knowledge has value.
What happens when enrichment is automatic (a short case study)
We built an auto-enrichment mode in early v0.8. It ran enrichment automatically on any lead that passed Stage 04 Qualification above a threshold score.
We ran it for six weeks with four beta operators.
Three problems emerged.
The first: credit spend was higher than expected because qualification thresholds that looked calibrated in testing turned out to be looser in production. Leads were passing Stage 04 that operators, on inspection, would have skipped. Auto-enrichment was spending credits on leads that would have been rejected in human review.
The second: operators reported feeling less confident about the pipeline. When they did not see Stage 07, they did not know what was in Stage 08. The pipeline felt like a black box. One operator described it as "emails going out and I'm not sure who approved them." They had not approved them. That was the problem.
The third: a small number of enrichment errors — wrong contact, wrong company — made it to Stage 08 without being caught. In the human-gated flow, these are caught at Stage 07 by the same operator who sees "VP of Engineering, Company X" and knows that Company X just promoted someone else to that role last month. In the auto-enrichment flow, the error continued.
We shipped a fix, not a feature. Auto-enrichment was removed in v0.9. It has not been requested back.
The principle behind the decision
Principle 3 — Humans review. Every time — is written about Stage 09, the email draft review. But the design intent runs through the whole pipeline.
Paitho is not a system that produces outputs you trust because you cannot see what is happening. It is a system that produces outputs you trust because you reviewed the key decision points.
Stage 07 is a key decision point. The question it forces — is this company worth pursuing right now — is one an operator can answer in eight seconds. The model cannot answer it at all.
There is a version of Paitho that automates more. We have discussed it. Our consistent conclusion is that automating Stage 07 would save operators six minutes per hundred leads and cost them the confidence that comes from knowing what is in the pipeline. That trade is not worth it.
What we are willing to automate at Stage 07
The data retrieval, once the operator approves enrichment, is automatic. The operator clicks approve; the enrichment API runs; the verified contact data populates Stage 08 without additional interaction. The human decision triggers the automation. The automation does not replace the decision.
We are also building a "skip reason" system for Stage 07 rejections — the same rejection taxonomy we built for Stage 09 email review. If an operator skips an enrichment, they tag a reason: wrong timing, wrong contact role, account context (existing relationship or competitor lock-in), or signal stale. Those tags feed back into the qualification model the same way Stage 09 rejections feed the email draft model.
That system is in beta. The concept is that every human decision in the pipeline is a data point, not just a gate.
Receipts
Figures from Paitho v0.9 beta operator data, Q1 2026. Illustrative.
- Stage 07 average review time per lead: 8 seconds
- Stage 07 rejection rate (operator skips enrichment on a passing lead): 17%
- Top skip reason: "Account context / existing relationship" — 31% of skips
- Second skip reason: "Signal stale on inspection" — 26% of skips
- Enrichment errors reaching Stage 08 in human-gated flow: ~1-3%
- Enrichment errors reaching Stage 08 in auto-enrichment mode (v0.8 test): ~4-6%
- Auto-enrichment mode credit overrun vs. projected: +38% in 6-week test
- Operators who requested auto-enrichment after its removal: 0
Closing
Principle 3 — Humans review. Every time — means the human is in the loop at the stages where human judgment changes the outcome. Stage 07 is one of them.
The eight seconds an operator spends deciding whether to enrich a lead is not overhead. It is the moment they confirm they know what is in their pipeline. We have not found a way to automate that knowledge. We have stopped looking.
Related:
- Why We Rejected ~4-5K Drafts Last Quarter
- BYOK Economics: What 1,000 Leads Actually Cost Across 6 Model Providers
- Contact Enrichment — Docs
— Jane Doe , Founder & CEO
Principle 3 — Humans review. Every time.