Claude Opus 5.5 alone vs with Five Dots: a pipeline audit case study
We planted 11 problems in a dummy company's sales pipeline and had Claude Opus 5.5 audit it 20 times, alone and with Five Dots. With Five Dots it found 94% of them, against 65%.
With Five Dots, Claude Opus 5.5 found 94% of the problems planted in a dummy company's sales pipeline, against 65% on its own, and 98% of its findings were right, against 90%. The gap comes from the client: Five Dots asks what the CRM cannot show.
- On what the CRM shows, Opus is strong on its own. The difference is in the problems only the client knows: 100% with Five Dots, 24% alone.
- Opus alone asked the client nothing in all 10 audits. With Five Dots it asked two or three rounds of questions every time.
- A blind judge preferred the Five Dots audit in 8 of 10 head-to-head pairs, and the Opus-alone audit in none.
The results at a glance
Both setups ran the same pipeline audit 10 times on 15 November 2026, against the current Five Dots release. The human consultant column is our estimate from Revenue Wizards' own consulting audits, not a result of this test: no consultant audited the test company. It shows where the two AI setups sit against the work they stand in for.
| Measure | Opus 5.5 alone | Opus 5.5 + Five Dots | Human consultant (estimate) |
|---|---|---|---|
| Planted problems found, weighted by severity | 65% | 94% | 97% |
| Problems visible in the CRM data | 84% | 91% | 93% |
| Problems only the client knew | 24% | 100% | 100% |
| Findings that are right (precision) | 90% | 98% | 98% |
| Wrong findings per audit | 0.5 | 0.1 | 0.1 |
| Traps wrongly called a problem, per audit | 0.3 | 0 | 0 |
| Rounds of questions to the client | 0 | 2–3 | 5–7 |
| How specific the fixes are (1–5) | 2.9 | 3.4 | 4.5 |
| Top-3 fixes that match the answer key (of 3) | 1.1 | 1.0 | 2 |
| Time per audit | 1.5 min | 4.7 min | 12+ hours |
| Cost per audit | $0.39 in AI compute | $1.55 in AI compute | about $2,000 in fees |
The test: one company, one pipeline, eleven planted problems
Both setups audited the sales pipeline of the same invented company, with the same CRM, the same client and the same request. The only difference was the Five Dots connector.
The company, Kestrelview Analytics, is fictional, and so is every name in it. It sells product-analytics software to mid-market online retailers: about 70 people, Series A, sales-led, a 14-day free trial and contracts of about $24k a year in the US and UK. Its HubSpot holds 44 deals, three sales reps and their activity history.
The client, a Head of Revenue Operations, was played by a separate AI. She answers what she is asked and volunteers nothing, so some problems only surface if the auditor asks.
An independent judge, also Claude Opus, read every final audit blind, with product names removed, and checked each finding against the answer key and the CRM records. High-severity problems count three times, medium twice and low once. Three traps, things that look wrong but are deliberate, counted against an audit that called them errors.
| Planted problem | Severity | How it can be found |
|---|---|---|
| Open deals with no activity for 60+ days, holding a large share of pipeline value | High | CRM data |
| Open deals with close dates in the past | High | CRM data |
| Open deals with no close date or an amount of 0 | Medium | CRM data |
| Closed-lost deals with no lost reason | Medium | CRM data |
| Renewals and expansions run through the new-business pipeline | High | CRM data |
| Deals stuck in Contract Sent for weeks | Medium | CRM data |
| Won deals that skipped the middle stages | Low | CRM data |
| One rep owns most of the stale deals | Low | CRM data |
| No written stage exit criteria | High | Only by asking the client |
| The forecast is built in the VP Sales' own spreadsheet | High | Only by asking the client |
| No regular pipeline review | Medium | Only by asking the client |
The questions: Opus alone asked none
Opus alone asked no questions in any of its 10 audits: it read HubSpot and went straight to findings. Opus with Five Dots read HubSpot first, then asked about what the data could not explain, two or three rounds per audit. The table follows one typical Five Dots audit.
The last row matters as much as the others. An answer of "I don't know" was kept as a question for the person who would know, not written up as a problem.
| What Five Dots asked | What the client said | What it changed |
|---|---|---|
| 10 of 24 open deals are renewals, expansions or pilots in the new-business pipeline. Is that on purpose? | Pilots are deliberate, paid 30-day pilots. She could not say about renewals. | Pilots not treated as an error; renewals logged as an open question for the VP Sales |
| Is the $140k deal right, against a $24k average contract? | It is real: the company's biggest prospect. | Not called a data error (one of the traps) |
| Is there a regular pipeline review, and who notices when a deal goes quiet? | No review. Deals are looked at when something comes up; the only fixed cadence is a quarterly board update. | New finding: no pipeline review |
| Why do wins jump from the first stage to Closed Won? Are stage criteria written down? | Nothing is written down. Reps move a deal when it feels right. | New finding: no stage criteria |
| Where do the forecast and the board number come from? | The VP Sales builds it every Friday in his own spreadsheet, because nobody trusts HubSpot. | New finding, ranked first among the fixes |
| When were lost deals last reviewed together? | She did not know. | Logged as an open question, not as a finding |
The presentation: a score, owners and a test of done
Both hand over a clear findings table with evidence. Five Dots adds a score, separates what the data proves from what the client said, names an owner and a test of done for each fix, and keeps open questions apart.
In the test company, Jonas is one of the reps and Daniel is the VP Sales. Opus alone wrote its first fix as: "Clean up the pipeline this week. Go through each rep's deals, starting with Jonas. For every stale or overdue deal, either close it out with a lost reason or give it a real next step and close date."
Opus with Five Dots wrote: "Start a weekly pipeline review and build the forecast from HubSpot. Owner: Daniel. Why first: the board number currently comes from a spreadsheet because HubSpot isn't trusted. Done when: the review has run four weeks in a row and the Friday forecast is read from HubSpot."
The first fix is sound but treats the symptom, the stale deals. The second goes after the cause, which Five Dots learned by asking: nobody reviews the pipeline, so the forecast lives in a spreadsheet.
| Part of the result | Opus alone | Opus + Five Dots |
|---|---|---|
| Overall verdict | "A problem, high priority" | Scored red on a fixed scale, with the methodology version and date |
| Findings | One table of 10, data only | Scored findings from the data; what the client said; patterns noticed; what is working |
| Unknowns | Mixed in with the findings ("worth checking") | Listed apart, each with who should answer it |
| Fixes | 5 in order, with what to do | 4 in order, each with owner, why it comes first, what changes and when it counts as done |
| Where it lives | A notes page in the client's Notion | The module page, the audit overview, and the client's memory, journal and decision log in their Notion |
How to read this
One area, one company. This covers the sales pipeline only, one of the ten areas in the Five Dots audit, for one invented company. A real audit covers more and depends on the real CRM.
The client is simulated. Real clients say more, less and differently. The simulated client keeps hidden facts back until asked, which is what makes asking matter.
The judge is also an AI. It reads each audit blind, without product names, and was checked first on a known-good and a known-bad audit.
No client data leaves the client's own setup. In the Five Dots runs, the connector received only the names of the connected tools and the module requested, never a deal, name or number. Each run checks this.
Our recommendation
Use Five Dots for audit work that follows a known method, and bring in human expertise where the method runs out.
| Situation | Use | Why |
|---|---|---|
| Standard audits: a pipeline health check, a quarterly review, a new area of the GTM audit | Five Dots | It finds nearly what a consultant would, in minutes instead of days, for a fraction of the cost, and keeps the result in your own workspace. |
| Complex work, with RevOps experience in the team | Five Dots, led by your RevOps lead | An experienced operator takes the findings and open questions further: weighs the trade-offs, runs the stakeholder conversations and turns the roadmap into a change plan. |
| Niche topics and knowledge gaps: commission design, pricing, complex CRM architecture, an unfamiliar market | Human expert, with Five Dots as the starting point | A specialist brings what neither AI setup has: patterns from many similar companies and judgement on the edge cases. |
| Validation of a big decision before it goes to the board or changes how people are paid | Human expert | An outside check from someone accountable for the call. |
Sources & further reading
Put the method to work.
Run the same audit on your own pipeline, in the AI assistant you already use.