# Reconstructing a Competitor Org Chart From Public Profiles

*You can produce a defensible org chart of one target company with every reporting line tagged by confidence, and know when it is complete enough to hand up.*

- Canonical URL: https://www.refolk.ai/guides/reconstruct-competitor-org-chart
- Pillar: Market and talent intelligence
- Format: Playbook
- Published: 2026-08-04
- Last reviewed: 2026-08-04
- Reading time: 15 min

Rebuilding a competitor's reporting structure from public data is a job strategy, research, and talent-intelligence teams get asked to do and rarely get taught how to do. This is the tool-agnostic method for one target company: harvest the people, draft the lines, grade every line by evidence, and know the point at which the chart is honest enough to hand to a decision-maker. It is written to be run start to finish on a single target, including the awkward cases where title hierarchy lies.

The vendor pages that rank for this work sell you a finished chart or an executive-search retainer. Neither shows you how to infer a reporting line and defend it. That is the gap this fills.

## What actually establishes a reporting line

A reporting line is established by triangulating multiple public signals, never by trusting a title alone. In a documented competitor reconstruction, analysts read metadata from thousands of profiles, reviewed active job postings, and scrutinized press announcements about executive hires. No single one of those was treated as proof.

The signals are not equal. Rank them by how hard they are to fake:

| Signal | What it proves | When it lies |
|---|---|---|
| Press "X will report to Y" language | An explicit, stated reporting fact | Announcement predates a quiet reorg |
| Direct-report co-mention or team page | Two people place themselves on one team | Team page is stale or aspirational |
| Title-band adjacency | Two people sit one layer apart | Bands overlap across a matrix |
| Raw title seniority | Self-declared rank only | Title is inflated; person has no reports |

The strongest explicit signal is line-style language borrowed from formal charts: a solid line usually represents the primary reporting relationship between an employee and their department head, while a dotted line signifies a secondary, often less formal relationship. Title seniority ladders are the weakest signal because titles are self-reported and inflated. A "Director" is a claim, not a fact about who reports to whom.

> **Rule:** No edge above "inferred" without a second signal
>
> Draw a line from a title band if you must, but it stays inferred until a second independent public signal corroborates it. One page, however official, is never enough to call an edge confirmed.

## Grade the source and the claim on separate axes

Grade the source and the information as two independent things, or a confident-sounding page will drag an unconfirmed line onto your chart. The intelligence community's Admiralty Code, part of NATO AJP-2.1 and in use in defence and security since the 1940s, is the reusable model, and you do not need clearance to borrow it.

The Admiralty Code uses a 6 by 6 alphanumeric grid. Source reliability is rated A to F based on the source's track record and verifiability. Information credibility is rated 1 to 6 based on the consistency and confirmation of the specific claim, independent of the source. The two axes are deliberately separate: a usually reliable source can still carry a claim that nothing else confirms.

The move from inferred to confirmed is corroboration. An edge only reaches credibility "1" when a second independent signal confirms it. A press release from the company is a strong source, so it might rate B, but if only that one page states the line, the claim is still credibility 2 or 3. It reaches B1, usually reliable and confirmed, only when a team page or a report's own profile agrees.

For org work, a lightweight two-axis tag is enough:

**Edge grade tag**

```
edge: <person> -> <manager>
source_reliability: A | B | C | D | E | F
information_credibility: 1 | 2 | 3 | 4 | 5 | 6
signals: <signal 1> :: <signal 2>
verified_as_of: <date>
```

*Append this to every reporting line in your chart data. Nothing ships above C3 as "confirmed".*

> A trusted sourcer's single assertion is still only possibly true until a second signal confirms it.

## The denominator problem, and why coverage lies

The chart is only as honest as the number you measure it against. Because only 10 to 20 percent of LinkedIn profiles reflect real-world, up-to-date employment, coverage measured against raw profiles flatters completeness. Measured against disclosed headcount, the same coverage often collapses.

The two diverge because inactive and mis-attached profiles inflate the base. One firm with 30 real employees showed 350 on its company page; a one-person company showed over 1,500. At the same time, employee-count accuracy typically runs 70 to 90 percent depending on company size and industry, with tech companies higher. So the count band can be roughly right while the individual profiles are badly stale.

**10-20% - Share of LinkedIn profiles that reflect current employment**

Everything else is stale or mis-attached, which is why your roster is never your denominator.

Pin your coverage to a headcount you can defend: the company's own disclosure, a self-reported size band such as 201-500 or 1,001-5,000, or a filtered third-party estimate. Then state the coverage ratio plainly, for example "reconstructed 118 of an estimated 260 engineers, 45 percent coverage." One honest gap I have to flag: no public source establishes a specific percentage of headcount you must place before spans become trustworthy. That threshold is not established publicly. Treat any such number as your own working assumption, and disclose it.

## Read the shape before you trust the lines

Before grading a single edge, know what a plausible org looks like so anomalies jump out. Two shapes matter: the ratio between layers, and the span within a layer.

Refolk's index gives a clean picture of how an engineering leadership pyramid widens as you descend. These are population counts across a whole market, not one company, but the ratios are the sanity check you apply to your reconstructed tree.

| Layer | Population | Ratio to layer above |
|---|---|---|
| VP Engineering | 1,778 | - |
| Director of Engineering | 17,261 | 9.7x VP count |
| Engineering Manager | 41,977 | 2.43x Director count |

Populations are from Refolk's index and are published nowhere else; the ratios are derived by dividing each layer by the one above. The point is the shape: real pyramids fan out sharply at the top and less so lower down. If your reconstructed chart has more Directors than Engineering Managers, you have mis-sorted a title band.

The trap is transferring these ratios across markets. In Refolk's index, the same title tells a very different story in a thin market.

| Market | VP Engineering population | US-to-market multiple |
|---|---|---|
| United States | 1,778 | 1.0x |
| United Kingdom | 80 | 22.2x |

Populations from Refolk's index; the multiple is derived. US VP Engineering profiles outnumber UK ones by 22.2 times. Applying US layer ratios to a UK subsidiary will read the noise of a small population as a flatter org. Benchmark within the market you are mapping.

#### Where profile volume drops as you go down one function

| Stage | Figure | Note |
| --- | --- | --- |
| Leadership layer captured | 100% | Executives are the best-covered public layer |
| Named managers with a report | narrows | Requires a direct-management artifact |
| Individual contributors placed | narrows | The stalest, thinnest data |
| Edges corroborated to credibility 1 | narrowest | Only two-signal lines survive here |

*Fewer profiles at every step below leadership is normal; a plateau or reversal usually means mis-sorted title bands.*

For spans, published benchmarks give you the range a real manager sits in:

| Band | Reported span |
|---|---|
| Manager/Senior Manager | 4.9 avg |
| Director/Senior Director | 4.6 avg |
| Knowledge-work mid-manager | 7-10 |
| Frontline standardized | 15-25 |
| Senior executive | 3-7 |

Manager and Senior Manager spans now average 4.9 direct reports, based on an analysis of over 257,000 managers; Directors and Senior Directors average 4.6. On depth, McKinsey holds that even the largest organizations should not exceed six layers, and truly agile orgs often show only three. If your tree runs deeper than six layers, you have almost certainly split one real band into two.

## Run the reconstruction

Here is the full procedure, in order, with who does it and how long each stage takes. It is built for a single analyst working one target company and one function. Executive-search practitioners front-load primary interviews; the public-data method treats interviews as optional corroboration and leads with harvesting.

#### Reconstruct one competitor org chart

1. **Scope and fix the denominator** - Pick one target company and one function. Pull a headcount baseline from company disclosure or a third-party estimate; do not use the raw profile count. Done: a target number the chart is measured against. (0.5 day)
2. **Harvest the leadership layer** - Capture named executives from the company site, press releases, and profiles. This layer is the most reliable public data but covers only the top. Done: top two layers named with titles. (0.5-1 day)
3. **Enumerate the function's population** - List every profile in the target function and deduplicate against the denominator. Done: a clean roster with a recorded coverage ratio. (1-2 days)
4. **Draft candidate edges from titles** - Sort by title band into provisional layers and connect them. Every edge is inferred only. Done: a first-pass tree with nothing graded above inference. (0.5 day)
5. **Corroborate edges with evidence** - For each edge seek a second independent signal, then apply the two-axis grade. Done: every edge tagged with a reliability letter and credibility number. (2-3 days)
6. **Handle matrix and flat exceptions** - Flag dual-report and dotted-line cases; separate managers from senior ICs using direct-management evidence only. Done: ambiguous nodes marked, not force-fitted. (0.5-1 day)
7. **Sanity-check against benchmarks** - Compute spans and layer count; compare to the benchmark ranges. Flag spans of 1-2 and depth above six. Done: an anomaly list. (0.5 day)
8. **Set freshness and re-verify** - Timestamp every edge; monitor WARN feeds and press for reorgs. Done: a chart with a verified-as-of date and a re-check trigger. (ongoing)

Step 2 leans on a real strength of public data: leadership sources are extremely official and reliable and updated at least once annually, but they concentrate on the senior level only. Do not mistake good top-of-house coverage for good overall coverage.

The hard stage is step 5, corroboration at scale, because you are chasing a second signal for every person one at a time. This is where a search tool that takes a plain-English relationship query earns its place.

I ran this search: `Engineering Managers at Stripe in the US who have posted a job opening for their own team.` - [see the full result list](https://www.refolk.ai/s/a01tm9rb82).

*Returns managers with a hiring-manager job post, which is direct-management evidence you can use to promote an edge above "inferred".*

That kind of query is exactly what [Refolk](/) does well: it takes a relationship you are trying to prove and returns the people who evidence it, across the public GitHub graph, LinkedIn, and the open web, instead of leaving you to page through profiles by hand.

## The matrix and flat-org exceptions

Matrixed and flat structures are where confident charts go wrong, so handle them as explicit exceptions rather than forcing them into a clean tree. A matrix organization is by definition one where people report formally to more than one manager, which defeats single-parent inference outright.

Line weight will not save you. The weight of a line is meant to represent the level of power and influence of the different managers, not the reporting fact, so a heavier line is a judgement, not evidence. When authority is disputed, true power remains with the solid-line manager who handles objective-setting and performance evaluation. The rule follows directly: if you cannot see who owns performance review, you cannot grade the primary edge, so leave the node dual-tagged.

Flat orgs break layer-count inference from the other direction. In a highly digitally enabled world spans are often around 1 to 30, and agile organizations often show only three layers. A three-layer target is not necessarily missing data; it may genuinely be flat. The judgement call is whether a wide, shallow shape is real or an artifact of a thin roster.

#### Is this node's shape real or an artifact?

Horizontal axis runs from Narrow span (1-2 reports) to Wide span (15+ reports). Vertical axis runs from Weak corroboration to Strong corroboration.

| Quadrant | What it means |
| --- | --- |
| Narrow + weak | Almost certainly missing people; keep enumerating before you draw |
| Wide + weak | Possible flat org but unproven; tag as inferred and seek team pages |
| Narrow + strong | A real small team or a senior IC; confirm no hidden reports exist |
| Wide + strong | A genuinely flat, high-span layer; record it as confirmed |

*Cross span width against corroboration to decide whether to trust a flat or narrow node.*

To separate a real manager from a senior individual contributor sharing the same title band, you need direct-management evidence, never the title: a named report who lists that person as manager, a "team of N" statement, or a hiring-manager job post. Absent that, a "Director" is a senior IC and any line beneath them is invented.

## How this goes wrong

Most bad org charts fail in a handful of predictable ways. Each has a false positive and a specific check.

- **Title inflation faked as seniority.** A "Director" with no reports is a senior IC; the false positive is drawing an edge under them. Check: require one direct-management artifact before grading above inferred.
- **Matrix collapsed into one parent.** Forcing a dual-reporting node under a single manager invents a false line. Check: flag dotted-line and dual cases; the solid line wins only where performance-review authority is evidenced.
- **Stale edges read as current.** A profile not updated on departure keeps a departed person on the chart. Check: timestamp edges and cross-check WARN feeds and press.
- **Inflated denominator.** Using a raw profile count, which can run 10 times reality, makes coverage look complete when it is not. Check: anchor to disclosure or a filtered estimate.
- **Narrow-span artifact.** A manager showing 1 to 2 reports usually means you missed people, not a real narrow span. Check: compare against the 4.6 to 10 benchmarks and treat sub-3 spans as incomplete.
- **Over-deep tree.** More than six layers usually reflects title-band mis-sorting. Check: apply the six-layer ceiling and collapse duplicate bands.
- **Single-source confirmation.** Grading an edge confirmed off one page violates the corroboration rule. Check: require two independent signals for credibility 1.
- **Country transfer bias.** Applying US span and title norms to a thin market over-reads a 22-times-smaller population as a flat org. Check: benchmark within-market.

> **Watch out:** A span of one or two is a data problem, not a finding
>
> Because real knowledge-work spans sit between 4.6 and 10, a reconstructed manager with one or two reports almost always means you missed people. Do not report a "narrow span" as a structural insight until you have exhausted the roster.

The single most common upgrade error is confirming an edge from one confident source. Two independent signals is the bar, and it is not negotiable if a decision-maker will act on the chart.

## Keep the chart current, and know when it is done

Official pages are a trailing indicator; filings are a leading one. LinkedIn's own pressroom showed the same headcount figure more than two months after a real cut, so it did not yet reflect the reduction. Profiles lag too: coaches advise updating only once a quarter, and many people never update on a job change.

The fastest structured public signal for departures is the WARN filing. The WARN Act requires employers with 100 or more employees to give at least 60 calendar days' advance written notice of a plant closing or mass layoff affecting 50 or more workers at a single site. WARN notices are public records, and aggregators refresh them daily; one aggregated database holds over 82,000 filings covering more than 8.87 million workers. Because the notice is legally mandated ahead of the event, it surfaces reorg signals before marketing pages catch up.

> **Note:** Watch filings, not just profiles
>
> WARN filings post with 60 days of statutory lead, so a mass layoff is public before the affected people update their profiles. Monitor the feed for your target and re-verify any edge it touches immediately.

Adopt a simple freshness rule: re-verify any reporting edge older than one quarter, and re-verify immediately after any press-reported reorg or layoff. Stamp every edge with a verified-as-of date so a reader knows exactly how fresh each line is.

Before you hand the chart to a decision-maker, run this check.

#### Ready-to-hand-up check

- [ ] Coverage is stated against a disclosed or estimated headcount, not the raw profile count.
- [ ] Every edge carries a source-reliability letter and a credibility number.
- [ ] No edge is marked confirmed on a single source.
- [ ] Every dual-report and dotted-line case is tagged, not forced under one parent.
- [ ] No manager shows a span of 1-2 without an explicit missing-data note.
- [ ] The tree is no deeper than six layers, or the extra depth is justified.
- [ ] Span and layer benchmarks were applied within the target's own market.
- [ ] Every edge has a verified-as-of date and a re-check trigger.

When those all pass, the chart is defensible: not complete, but honest about what it knows and how well it knows it. That is the deliverable. A chart that admits its gaps and grades its lines is worth more to a decision-maker than a tidy tree that hides its guesses, because the reader can see exactly which lines to trust and which to treat as leads worth a phone call.

## Frequently asked questions

### How do I build a competitor org chart from public data alone?

Fix a headcount denominator first, harvest the named leadership layer, then enumerate everyone in one target function. Draft reporting edges from title bands, but treat each as inferred until a second independent signal confirms it. Grade source reliability and information credibility separately, flag matrix and flat exceptions rather than forcing them into a single parent, and sanity-check spans and layer counts against published benchmarks before handing the chart up.

### How can I tell if someone is a real manager or just a senior individual contributor with an inflated title?

Titles are self-reported and inflated, so seniority alone proves nothing. Require at least one direct-management artifact before drawing an edge under a person: a named report who lists them as manager, a 'team of N' statement, or a job post where they are the hiring manager. A Director with no reports and no such artifact is a senior individual contributor, and any line beneath them is a false positive.

### What confidence scale should I use to grade reporting lines?

Borrow the Admiralty Code used in defence intelligence since the 1940s. It rates source reliability on a letter scale from A to F and information credibility on a number scale from 1 to 6, independently. An edge only reaches credibility 1, meaning confirmed, when a second independent signal corroborates it. A single trusted page still leaves the claim unconfirmed, so it caps at credibility 2 or 3.

### How current is public profile data for org mapping?

Not very. Only 10 to 20 percent of LinkedIn profiles reflect up-to-date employment, and career coaches advise updating only once a quarter, with many users never updating on a job change. Even official pressroom figures can lag a real headcount cut by more than two months. WARN filings are the fastest structured signal, posting with 60 days' statutory lead, so watch them for departures and reorgs.

### How do matrix organizations break org-chart inference?

Matrix structures put people under more than one manager by design, so single-parent inference invents a false line. Line weight is ambiguous too, since it represents power rather than the reporting fact. When you cannot see who holds performance-review authority, leave the node dual-tagged. True authority sits with the solid-line manager who sets objectives and runs evaluations, but you must have evidence of that authority, not just a heavier line.

---

*From the Refolk guide library. I revise these guides rather than replacing them, so the current version is always at https://www.refolk.ai/guides/reconstruct-competitor-org-chart*