Case Study: Alex's Competitive Analysis Sprint — Vision + Document AI in 90 Minutes

The Quarterly Brief

Every quarter, Alex produces a competitive analysis brief for the Brightleaf leadership team. The brief covers three major competitors' recent marketing activity: ad campaigns, product launches, pricing changes, and messaging shifts. It is one of the more time-intensive pieces she produces — not because the analysis is difficult, but because the research phase requires manually reviewing a lot of content across a lot of sources.

The old process looked like this: spend 2-3 hours browsing competitors' social feeds, advertising libraries, press release archives, and news coverage. Take notes manually. Consolidate into a structured brief. Write the synthesis and strategic implications. Total time: 5-6 hours, spread across two days.

In Q3, Alex ran an experiment. She redesigned the research phase around AI tools — specifically vision capabilities for ad screenshots and document analysis for press releases and news articles. The goal was to compress the research phase into 90 minutes and still produce a brief of equivalent quality.

Here is exactly what she did and how she did it.

Phase 1: Ad Analysis via Vision (40 minutes)

Alex's research for each competitor starts with their advertising. She monitors three ad libraries: Meta's Ad Library, LinkedIn's ad transparency feature, and a third-party display ad tracking service she subscribes to.

For this quarter's brief, she collected screenshots of all ads from each of her three competitors that had been running for more than two weeks (a proxy for "working" ads the competitor is investing in). Total: 31 screenshots across three competitors.

Setup prompt for each competitor batch:

She organized the screenshots into three batches — one per competitor — and wrote a structured analysis prompt for each batch. Here's the prompt for Competitor A:

These are screenshots of [COMPETITOR A]'s currently running digital advertising,
collected from their Meta Ad Library and LinkedIn ad campaigns. All ads have
been active for 2+ weeks.

Context: I'm preparing Brightleaf's Q3 competitive marketing brief. I need to
understand [Competitor A]'s current campaign themes, audience targeting signals,
and messaging strategy.

For each ad:
1. Primary message or value claim (what is the ad saying?)
2. Audience signal (who is this aimed at, based on imagery, copy, and placement context)
3. Product focus (what product or product category is featured?)
4. Call to action (what is the viewer being asked to do?)
5. Emotional/motivational appeal (what feeling is this trying to create?)

After analyzing all ads in this batch, provide:
- The 2-3 dominant campaign themes this quarter
- Any messaging angles that represent a shift from typical positioning
  (I'll add prior quarter context after your analysis)
- A plain-language summary of this competitor's apparent marketing strategy
  this quarter in 3-4 sentences

Format: Individual ad analyses in a compact table, followed by the synthesis section.

Results per competitor:

Old process for the same research: approximately 90 minutes of manual review, note-taking, and pattern identification.

What she found and how she used it:

The AI-generated tables gave her a structured catalog of all 31 ads — something she had never had in prior quarters. In the past, she reviewed ads mentally and noted the ones that stood out, inevitably missing patterns that only became visible across the full set.

Looking at Competitor B's full ad catalog, the AI synthesis identified that 7 of 11 ads featured user testimonials from a specific industry segment (small business owners in retail). Alex had noticed this in a few ads but hadn't registered it as a systematic strategy. This insight became the lead finding in the Competitor B section of her brief: a targeted strategy shift toward a specific customer segment.

Alex's review time was spent verifying the AI's ad descriptions against the actual screenshots (accuracy rate was high — she found errors in 2 of 31 ad descriptions, both involving misread small text in CTA buttons), adding context from prior quarters, and making judgment calls about strategic significance.

Phase 2: Press Release and News Analysis via Document AI (30 minutes)

For the document phase, Alex collected the following for each competitor: - 2-3 press releases from the last quarter (sourced from their newsroom pages) - Any major news coverage (product launches, partnerships, executive hires) - Any public investor communications where competitors are publicly traded

Total: 11 documents across three competitors, ranging from 400-word press releases to 2-page analyst reports.

Document analysis prompt for each competitor:

These are [COMPETITOR B]'s press releases and news coverage from the last quarter.

Context: I'm building Brightleaf's Q3 competitive brief. I need to understand
what [Competitor B] has announced, launched, or changed in the last 90 days.

Extract and organize:

1. PRODUCT/FEATURE ANNOUNCEMENTS: What new products or features have they launched?
   For each: name, brief description, target audience signal, announced date

2. PARTNERSHIP OR DISTRIBUTION CHANGES: Any new partnerships, retail relationships,
   or distribution agreements

3. PRICING OR PACKAGING CHANGES: Any changes to how they price or package products

4. MESSAGING SHIFTS: Key quotes from executives or marketing materials that indicate
   positioning changes

5. STRATEGIC SIGNALS: Any indication of where they are investing (hiring, acquisitions,
   new markets, new product categories)

Rules:
- Extract only from the provided documents — no outside knowledge
- Note the source document for each item
- If a category has no items in these documents, write "None in provided documents"

Output format: Structured sections by category, with bullet points under each.

Results:

The document extractions ran faster than the image analysis — about 6 minutes per competitor batch, with 5-8 minutes of Alex's review time per competitor.

The most significant finding from the document phase: Competitor C had quietly announced a partnership with a major grocery chain that would expand their retail distribution significantly. The partnership had been mentioned in a single paragraph of a broader press release announcing Q3 results. Alex had read this press release when it came out but hadn't registered the distribution announcement as strategically significant.

The AI, asked specifically to extract "partnership or distribution changes," surfaced it as a category-one finding. Alex's interpretation of its significance — this will materially change the competitive landscape in the grocery channel where Brightleaf has recently been growing — was the human expertise layer. The AI found the fact; she provided the strategic weight.

Phase 3: Synthesis (20 minutes)

With structured data from both phases — 31-ad analysis + 11-document extractions — Alex wrote a synthesis prompt:

I've completed competitive research on three competitors. Based on the ad
analyses and document extractions I've compiled, help me identify:

1. Cross-competitor themes: Are any of the three competitors moving in the same
   direction simultaneously? What might this signal about the market?

2. The most significant competitive development this quarter (single biggest thing
   we need to respond to)

3. Any gaps in Brightleaf's current messaging that our competitors are successfully
   occupying

I'll provide the summarized findings from each competitor. Please synthesize:
[paste of key findings from phases 1 and 2]

Note: These findings are AI-extracted but I've reviewed them. Flag any place
where you'd recommend I verify more carefully before including in an executive brief.

The synthesis took 5 minutes to run and 15 minutes of Alex's review and editing.

The Final Brief

Time breakdown:

New process: 84 minutes total (Alex's time: 51 minutes). Old process: ~195 minutes.

The quality assessment was harder to measure — but Alex shared both versions (without identifying which was AI-assisted) with a colleague and asked for a comparison. The colleague noted that the new brief was more systematic — it covered more competitors in more consistent depth — and surfaced one finding (the grocery chain distribution partnership) that was more significant than anything in the prior quarter's brief.

What She Learned

Vision AI is excellent for catalog-level analysis. When you need to process many similar items (ads, screenshots, designs) systematically, vision AI produces structured catalogs that manual review cannot match for consistency and completeness. The pattern-identification capability across a full set is particularly valuable.

Document AI finds what you weren't specifically looking for. Because the extraction prompts had explicit categories, the AI surfaced information in secondary document sections that Alex would have missed with active-reading-only attention. The structured extraction format forces complete coverage of a document in a way that skimming does not.

The human value shifts from gathering to interpreting. Alex spent zero time in the new process copying ad descriptions into a spreadsheet or searching press releases for key phrases. Her 51 minutes were almost entirely spent on analysis, interpretation, and judgment — the work that required her expertise about the market, Brightleaf's strategy, and what the findings meant. This is the shift multimodal AI enables: from information gathering to information interpretation.

Error verification is still necessary for critical claims. She found 2 AI errors in ad descriptions and verified 3 document extractions before including them in the brief. The error rate was low but not zero. For an executive brief, 100% accuracy is required on claims that drive strategic decisions. Alex's rule: verify anything that will be quoted or cited as a specific factual claim; trust AI for pattern-level observations.

"The brief is better than before," she said. "Not just faster. The systematic coverage that AI makes possible surfaces things I would have missed. But my judgment is still the product — I just spend my time on the judgment instead of the information gathering."