Portfolio Company Benchmarking Without Bad Data

Executive summary

Portfolio company benchmarking compares the operating performance of companies in a portfolio, or against outside peers, so that owners and managers can see where performance differs and ask why. Done well, it finds the companies that have a solved problem another company still has. Done badly, it ranks companies by who defines a metric most generously.

The usual failure is not the arithmetic. It is the data. Company A computes days sales outstanding on last quarter's sales. Company B uses the trailing year. Company C counts unbilled revenue as receivables. Each reports "DSO" in the monthly pack, and the portfolio team puts the three on one slide. The slide is wrong before anyone reads it.

This guide sets out a method to benchmark without that problem:

1. Write one definition per metric, including what is in, what is out, and the period basis.
2. Normalize each company to the definition from its own ledger, rather than accepting its reported figure.
3. Label the quality of each data point, so a weak figure is visible as weak.
4. Choose peer groups by method, with stated criteria, rather than by convenience.
5. Read gaps as questions. A gap tells you where to look. It does not tell you who is better.
6. Record what you decide and what happens, so the benchmark leads to action and the action is checked later.

It uses no external statistics. You will build comparison from your own companies' data and from named public sources you choose and cite yourself.

When to use this

Use this guide when:

• A sponsor or operating partner compares KPIs across portfolio companies and the numbers look inconsistent.
• You are building a portfolio dashboard and need definitions that hold across companies.
• A company is told it is "behind the portfolio" on a metric and wants to test whether the comparison is fair.
• You are setting targets and want something better than a number from a blog.
• You are preparing information for LPs or a board and need the comparison to survive questions (see LP and investor portfolio reporting).

Do not use benchmarking to replace the review of an individual company. A benchmark identifies where to look. The company's own operating review decides what to do.

Diagnostic questions

1. Is there a written definition for each KPI that the portfolio compares, including inclusions, exclusions and period basis?
2. Do all companies use the same definition, or do they report what their own system produces?
3. Are figures taken from the companies' ledgers or from a figure the company typed into a template?
4. Do companies close their books on the same schedule, so that monthly figures cover the same period?
5. Do you know which companies use different accounting policies that affect the metric (revenue recognition, inventory costing, capitalization)?
6. Are intercompany items, one-time items and acquisitions treated the same way across companies?
7. When a company is acquired mid-year or integrates an add-on, how are its figures restated?
8. Are peer groups defined by stated criteria (size, model, geography), or by who was convenient to include?
9. Is each data point labelled by quality, for example reconciled to the ledger, reported but unreconciled, or estimated?
10. When a benchmark shows a large gap, who asks the company about it, and what record is kept of the answer?
11. Do companies feel safe reporting weak numbers, or do they smooth them?
12. Has any benchmark in the last year led to a decision that was later checked?

If questions 1 to 3 are answered no, begin with definitions and data collection. If 8 and 9 are no, your comparisons may be built on weak peer groups and unlabelled data. If 10 to 12 are no, you are producing slides, not results.

The framework: from reported number to usable benchmark

Think of benchmarking as a pipeline with five stages. A weak stage contaminates everything after it.

Stage 1: The metric dictionary

For each metric, write the following. Keep it to one page per metric.

• Name and purpose: what decision the metric informs.
• Formula: numerator and denominator in precise terms.
• Inclusions and exclusions: for example, "receivables exclude unbilled revenue, include disputed invoices."
• Period basis: monthly, trailing quarter, trailing twelve months, and the day count.
• Source: which ledger accounts or system reports feed each element.
• Known variants: other ways companies compute it, and how to convert.
• Owner: the person who maintains the definition.

Change a definition rarely, and log every change with the date, so that history stays comparable.

Stage 2: Collect from source, not from summaries

Ask for the underlying data (trial balance, aging report, inventory listing) and compute the metric centrally. When a company supplies only its own KPI, you inherit its definition. Where central computation is not possible, require a reconciliation from the reported figure to the standard one.

Stage 3: Normalize

Normalization means restating each company's figure to the standard definition. Common adjustments:

Some accounting policy differences cannot be fully removed. Say so on the benchmark rather than claiming precision you lack, and rely on the company's auditors or advisers for treatment questions.

Stage 4: Peer groups

A benchmark against the wrong peers misleads. Define peer groups using stated criteria, recorded before looking at the results:

• Business model (distribution, manufacturing, services, software, project-based) and customer type.
• Scale, as a revenue or headcount band wide enough to contain several companies.
• Capital intensity or working capital profile, and geography or regulation where it affects the metric.

Three kinds of reference exist, in descending order of reliability for your purposes:

1. The company's own history, for example the best trailing period. It shares the company's definition, customers and systems.
2. Other portfolio companies on the same normalized definition, with similar models.
3. Named public sources, such as filings of listed companies in a similar business. Cite the source, the period and the definition. Read their definitions, since public figures may be computed differently from yours.

Avoid unsourced industry averages. A figure you cannot trace to a definition and period is not a benchmark.

Stage 5: Read the gap

A gap between two companies has four possible explanations. Test them in order.

1. Definition difference that normalization did not catch.
2. Business difference, such as customer mix or contract terms.
3. Timing, such as a large shipment or a one-time item.
4. Real performance difference, which is the finding you are looking for.

Only the fourth justifies action on the company. The first three justify better data or better peer groups. This ordering protects you from a common error, which is to push a company to close a gap that does not exist.

Step-by-step process

1. Choose the metrics. Start with five to eight that drive decisions and that every company can produce. Owner: operating partner and CFO. Output: a metric list. See financial KPIs for executives and KPI design: choosing 12 metrics for selection.
2. Write the metric dictionary. One page per metric. Owner: portfolio finance lead. Output: a signed-off dictionary.
3. Define the data request. List the source files per metric, the period and the due date. Owner: portfolio finance lead. Output: a standard data request.
4. Collect source data. Each company supplies ledger-level extracts. Owner: company controllers. Output: files by company and period.
5. Compute centrally and reconcile. Compute each metric on the standard definition. Where the company reports a different figure, write the bridge. Owner: portfolio finance. Output: reported-to-normalized bridge per metric per company.
6. Label data quality. Mark each point as reconciled to the ledger, reported but not reconciled, or estimated. Owner: portfolio finance. Output: a quality flag.
7. Define peer groups. Write criteria, then assign companies. Owner: operating partner. Output: peer group list with criteria.
8. Build the comparison. Show each company against its own history first, then against peers, with quality flags visible. Owner: portfolio finance. Output: benchmark view.
9. Review with each company. Walk the gaps with the management team and test the four explanations in order. Owner: operating partner and company CFO. Output: a list of gaps classified as definition, business, timing or performance.
10. Decide and record. For the performance gaps worth acting on, log the decision, owner, expected result and check date. Owner: operating partner. Output: logged decisions, as described in operational decision systems.
11. Recheck at the check date. Compare the new figure with the expectation, and feed the result back into the next benchmark. Owner: decision owner. Output: closed decision entries.

Templates

These are designed to be copied into a spreadsheet.

Metric dictionary entry

Reported-to-normalized bridge

Data quality labels

Metrics and KPIs

These measure the quality of your benchmarking, not the companies.

For the company metrics themselves, set targets using three anchors in order: the company's own best trailing period, any contractual floor or ceiling that sets the limit, and named public peers with a similar model, on a definition you have verified. Do not apply a figure from an unsourced average.

Common mistakes

1. Benchmarking reported numbers. Each company's definition goes into the comparison, and the ranking reflects the definitions.
2. Comparing across business models. A distributor and a services business have different working capital profiles. A cross-model comparison shows model differences.
3. Choosing peers after seeing results. It builds in the answer you wanted. Write the criteria first.
4. Mixing period bases. Quarterly, trailing-year and annualized-month figures for the same metric can differ by several days or points.
5. Ignoring data quality. Mixing reconciled and estimated figures in one ranking gives the estimates false authority.
6. Treating a gap as a verdict. Test definition, business difference and timing before assuming a performance gap.
7. Averages that hide the spread. An average of three companies can be a number no company resembles. Show the individual points.
8. Making companies defend their numbers. If it feels like an audit, companies will smooth their figures.
9. Producing slides, not decisions. A benchmark that does not lead to a logged decision produces nothing to check.
10. Moving definitions without a log. A silent definition change breaks trend lines.

Worked example

Illustrative. The numbers below are made up to show the method. They are not benchmarks.

A sponsor wants to compare days sales outstanding (DSO) across three portfolio distributors. The reported figures in the monthly packs are:

On the reported figures, Alder and Cedar tie at 40.5 and Birch is worst at 42.1. The standard definition in the metric dictionary is: billed receivables only (excluding unbilled revenue), divided by last-quarter credit sales, multiplied by 90.

Source data from each ledger:

Normalization:

• Alder: 5.4 ÷ 12.0 × 90 = 40.5 days. No adjustment.
• Birch: 3.0 ÷ 7.2 × 90 = 37.5 days. Reported 42.1 equals 3.0 ÷ 26.0 × 365. Birch is growing, so the trailing-year basis overstates how long its customers take to pay. The bridge is -4.6 days from the period basis.
• Cedar: billed AR is 8.1 − 1.5 = $6.6M. 6.6 ÷ 18.0 × 90 = 33.0 days. The reported 40.5 equals 8.1 ÷ 18.0 × 90. The bridge is -7.5 days from removing unbilled revenue.

Reported and normalized figures:

Verification: 8.1 − 1.5 = 6.6; 6.6 ÷ 18.0 = 0.3667; × 90 = 33.0. 3.0 ÷ 7.2 = 0.4167; × 90 = 37.5. 3.0 ÷ 26.0 = 0.1154; × 365 = 42.1. 40.5 − 33.0 = 7.5. 42.1 − 37.5 = 4.6.

What the normalization changes:

• Alder moved from tied-best to worst on the standard definition. Its gap to Cedar is 7.5 days, which is large enough to examine.
• Birch's apparent weakness was mostly a period-basis artifact.
• Cedar's figure was flattered by its unbilled revenue inclusion in the reported number, but even excluding it, Cedar still leads. The unbilled balance of $1.5M is itself worth a question about how fast Cedar bills.

Gap review for Alder versus Cedar, testing the four explanations in order:

1. Definition: now the same.
2. Business: Cedar's customers are mostly on net-30 terms and Alder's mix includes customers on net-45. This is a legitimate part of the gap and is recorded.
3. Timing: Alder had one large late-quarter shipment, which is noted but not removed from the benchmark.
4. Performance: after the first three, a remaining gap is checked using best possible DSO for each company (current, not-yet-due AR ÷ credit sales × 90). See how to reduce DSO for the formula.

The sponsor's operating partner logs one decision: Alder will compute best possible DSO and average days delinquent, with a check date at the next quarterly review. Birch and Cedar are not asked to change anything. The benchmark found where to look.

AI prompt assistance

AI can help standardize data and draft bridges. It cannot know your ledger. Supply the data and verify the output.

1. Draft a metric dictionary entry.

I will give you the name of a KPI and how three of our companies currently compute it. Write one dictionary entry with the formula, inclusions, exclusions, period basis and known variants. List the differences between the three companies' methods, and propose conversion steps to a single standard. Do not choose figures. Mark anything you are unsure of.

Check: the finance lead confirms the standard, and the controllers confirm their own company's method.

2. Build a reported-to-normalized bridge.

Here is the reported KPI, the ledger inputs, and our standard definition for one company. Calculate the figure on the standard definition, show every step, and list each adjustment between the reported and normalized figures with its size.

Check: recompute the normalized figure by hand and tie the inputs to the trial balance.

3. Suggest explanations for a gap.

Company A's normalized metric is X and Company B's is Y on the same definition. List possible explanations under four headings: definition, business difference, timing, performance. For each, name the data we would need to confirm it. Do not assert any explanation as fact.

Check: treat the list as hypotheses and test them with company data.

Implementation guide

Start with three to five metrics. Expect the first cycle to be mostly about discovering that companies compute things differently, which is useful in itself. Where the portfolio has shared finance resources that could produce the standard figures centrally, see shared services for PE portfolios.

Where accounting policies, audit requirements or lender and investor documents define how figures must be computed or presented, follow those agreements and involve your accountants, auditors or counsel. This guide covers management benchmarking, not external reporting rules.

Downloadable assets

The metric dictionary entry, bridge, data quality labels, peer group record and gap review log above are designed to be copied into a spreadsheet as they are. The DSO calculator can compute the standard-basis figure used in the worked example for each company.

Related resources

• Portfolio operations hub: the parent collection for portfolio operating topics.
• Shared services for PE portfolios: producing standard figures centrally.
• LP and investor portfolio reporting: presenting portfolio comparisons to investors.
• Financial KPIs for executives: candidate metrics and how to read them.
• KPI design: choosing 12 metrics: selecting the measures that matter.
• Operating review system: where benchmark findings become logged decisions.

Frequently asked questions

What is portfolio company benchmarking?

Portfolio company benchmarking compares operating metrics across the companies in a portfolio, or against outside peers, to find where performance differs and ask why. It is only reliable when every company is measured on the same written definition, computed from ledger data, and compared with peers chosen by stated criteria.

Why are portfolio KPIs hard to compare?

Companies compute the same KPI in different ways: different period bases, different inclusions such as unbilled revenue, different accounting policies. Two companies can report the same number for different realities. Comparing reported figures ranks the definitions, not the performance. Normalizing each company to one definition from its own ledger removes most of that distortion.

How do you choose peer groups for benchmarking?

Write the criteria before looking at results: business model, customer type, scale, working capital profile, and geography or regulation where relevant. Assign companies against those criteria. Prefer the company's own history first, then similar portfolio companies on the same definition, then named public sources whose definitions you have verified.

What do you do when a benchmark shows a large gap?

Treat it as a question, and test four explanations in order: definition difference, business difference, timing, and real performance difference. Only the last justifies pushing the company to change. Record the conclusion, and if action is warranted, log the decision with an owner, an expected result and a check date.

Can you benchmark without external data?

Yes. The most reliable benchmarks are the company's own best trailing performance and other portfolio companies on a common definition. External references help only when you can name the source, the period and the definition. An unsourced industry average, with no definition or period, should not be used to set a target.

How this works in a System of Performance

A benchmark only helps if the definitions, the adjustments and the decisions that followed stay connected to the numbers. Today the bridge sits in one spreadsheet, the explanation in an email, and the follow-up in nobody's list. A Performis routine can take on a recurring job like this, with the scope and fit agreed with you: it works from the records your team supplies, prepares the work for your named reviewer, and keeps each approval on record. If you want to see how your own metrics line up on one definition, bring it to us as a launch partner.