Building Comparable Company Analysis with AI

Comparable company analysis is the single most-used valuation technique on the buy-side and in sell-side research. The mechanical parts — pulling the universe, computing multiples, screening outliers — are routine. The hard parts are picking the right peers and explaining why a name trades at a discount or premium to them. AI accelerates both.

This lesson assumes you already know how comps work as a valuation method — what an enterprise value bridge is, how to compute LTM vs NTM multiples, why EV/EBITDA differs from P/E. If any of that is shaky, the underlying valuation mechanics are taught in Financial Modeling & Valuation; this lesson is about layering AI on top of a method you already understand, with AI as your associate.

What You'll Learn

• How to use AI to define a defensible peer universe
• How to compute and clean multiples with Advanced Data Analysis
• How to identify outliers and explain trading premiums or discounts
• How to draft the comp commentary section of a research note

Defining the Peer Universe

Picking peers is more judgment than math. Most junior analysts pick whoever the company names in their 10-K under "competition" and call it done. A more rigorous approach: have AI propose the universe, then prune it.

Prompt:

I am building public comparables for [COMPANY], a [BUSINESS
DESCRIPTION IN ONE SENTENCE], headquartered in [COUNTRY],
with $[X]B revenue and a [GROWTH/MATURE] business profile.

Propose 12 comparable public companies. Group them into:

A. CORE PEERS — same business model, same geography, similar size
B. ADJACENT PEERS — same industry but different business model or
   scale (justify why each one belongs)
C. STRETCH PEERS — companies often cited but probably should NOT
   be included (explain why)

For each company, give: ticker, primary exchange, one-sentence
business description, latest reported revenue, and one reason
they belong (or don't).

You now have an over-inclusive list with rationales. Your job is to argue with it. Ask follow-ups:

Remove peers with less than $500M revenue. Add European peers with similar margin profiles.

By the third iteration you have a sharper peer set than most analysts produce, and you have done the thinking in 15 minutes instead of an afternoon.

Computing and Cleaning Multiples

Pull the raw data from your standard source (Capital IQ, FactSet, Refinitiv, or even a free aggregator for practice). Export to CSV with these columns: ticker, market cap, enterprise value, LTM revenue, LTM EBITDA, NTM revenue, NTM EBITDA, NTM EPS.

Upload to ChatGPT with Advanced Data Analysis enabled:

The attached CSV is my public comparables set for [COMPANY].
Compute and add these columns:

- EV / LTM Revenue
- EV / NTM Revenue
- EV / LTM EBITDA
- EV / NTM EBITDA
- P / NTM EPS
- LTM EBITDA margin (%)
- NTM revenue growth (%)

Then:
1. Report the median, mean, 25th percentile, and 75th percentile
   for each multiple, excluding any value above 3x the interquartile range.
2. Identify which companies are statistical outliers and explain
   in one sentence what likely makes each one an outlier
   (e.g. M&A pending, accounting one-time, business model mix).

Output two markdown tables: the augmented company-level table and
the summary statistics table.

Advanced Data Analysis runs real pandas on the file and returns clean output. You can copy the tables into your model.

Spotting and Explaining Outliers

Outliers in comp tables are not always bad. Sometimes they are the entire point — the target trades at a discount because peers include a special situation. The analyst's job is to explain. Prompt:

Looking at my comp set, [TICKER A] trades at 35x EV/NTM EBITDA
while the median is 15x. What is the most likely explanation?
Consider:

- Pending M&A announcements
- Accounting distortions (one-time impairment recovery, etc.)
- Business model differences (subscription vs transactional, etc.)
- Growth profile or guidance changes
- Premium for category leadership

Give me your top three hypotheses with what I would need to check
to confirm each.

AI is good at generating hypotheses. You verify them via the news search in Bloomberg or a quick FactSet pull. This is the kind of work that used to require a senior analyst's intuition.

Identifying the Right Multiple to Anchor On

Different multiples matter for different sectors. AI helps you pick. Ask:

For valuing [COMPANY] in the [SECTOR] industry at this stage of its
lifecycle (growth / mature / decline), which multiple should be the
primary anchor, and which two should be secondary? Justify briefly.
What multiples are typically misleading for this sector and why?

Example outputs:

• For consumer staples: EV/EBITDA primary, P/E secondary, EV/Revenue is misleading
• For early-stage SaaS: EV/Revenue or EV/ARR primary, Rule of 40 secondary, EBITDA multiples often unusable
• For banks: P/Tangible Book primary, P/E secondary, EV multiples are unusable due to capital structure

Writing the Comp Commentary

The output of comp work is usually a one to two paragraph writeup in a research note. Prompt:

Draft the comparable companies commentary for [COMPANY] in our
research note. Reference these facts:

- Target trades at [X]x EV/NTM EBITDA vs peer median of [Y]x
- Discount/premium is [Z]%
- Top three reasons for the discount/premium based on your earlier
  outlier analysis
- Conclude with whether the multiple is justified or whether we see
  re-rating potential

200 words. Sell-side research tone. Avoid the words "strong" and
"robust." Reference each peer by ticker on first mention.

This becomes 70% of the section in your research note. You edit for house style and verify every number.

Building the Football Field Chart

A football field — the range of values implied by different valuation methods — is the closing exhibit of most banker decks and pitchbooks. AI helps you build the inputs:

For [COMPANY], compute the implied equity value range using:

1. Trading comps: peer 25th to 75th percentile EV/NTM EBITDA applied
   to our NTM EBITDA estimate of $[X]M
2. Transaction comps: median deal multiple of [Y]x applied to LTM EBITDA
3. DCF base case: $[A]M
4. DCF sensitivity: $[B]M to $[C]M
5. 52-week trading range: $[low] to $[high]

Output as a table with each method's low, midpoint, and high, plus
the resulting per-share range using a current share count of [N]M.

Drop the table into PowerPoint and your football field is ready.

Key Takeaways

• Start with AI-generated peer universe in three buckets (core, adjacent, stretch) and then prune
• Use Advanced Data Analysis to compute multiples, summary statistics, and outliers from a CSV
• Always ask AI to explain outliers — it generates better hypotheses than most associates
• Sector dictates which multiple anchors — ask AI explicitly before defaulting to EV/EBITDA
• AI drafts the comp commentary in your house style; you edit and verify