How a Series B SaaS Company Fixed AI Categorization in 6 Weeks

The problem: AI was sending the wrong buyers

A Series B enterprise SaaS company (identity withheld at their request) came to Optimly with a pipeline quality problem. Their product serves mid-market and enterprise teams with an average deal size of $85K. But nearly 40% of their inbound demos were coming from small businesses and solopreneurs who would never close.

The company's marketing team had spent months optimizing their website, refining ICP messaging, and tightening their paid media targeting. None of it moved the needle on misfit demo volume.

The cause wasn't their marketing. It was AI.

When we ran the company through Optimly's Brand Strategist audit, the results were stark:

• ChatGPT described them as "a project management tool for small teams and freelancers"
• Claude categorized them as "a productivity app similar to Notion and Trello"
• Gemini placed them in "small business software" and compared them to tools with $10/month price points
• Perplexity pulled mixed signals — sometimes enterprise, sometimes SMB

Their actual positioning: enterprise workflow automation for operations teams at companies with 500+ employees. AI wasn't just wrong — it was pointing the wrong buyers directly at their sales team.

Why AI got the category wrong

AI models form beliefs about companies from thousands of sources ingested during training. When those sources disagree — or when the loudest signals come from the wrong era of a company's life — the model locks in a distorted view.

For this company, we traced the categorization problem to three root causes:

1. Legacy positioning artifacts. The company had launched three years earlier as a small-team productivity tool before pivoting upmarket to enterprise. Their original Product Hunt launch page, early TechCrunch coverage, and first-generation G2 listing all described the pre-pivot product. These sources carried disproportionate weight in model training because they had high domain authority and backlink profiles.

2. Competitor comparison pages. Several "best project management tools" roundup articles included the company alongside Trello, Asana, and Monday — tools that serve a completely different segment. AI models learned the association: if you're in a list with Trello, you must be like Trello.

3. Inconsistent self-description. The company's own marketing materials used different category language across different channels. Their website said "enterprise workflow automation." Their LinkedIn said "team productivity platform." Their Crunchbase profile said "project management software." Each inconsistency gave AI another conflicting signal.

None of these problems would have been caught by a traditional SEO audit. The website was fine. The problem was in the broader information ecosystem that AI models learn from.

What Optimly did: Define, Detect, Deploy

Phase 1: Define the ground truth (Week 1)

Before fixing anything, we needed a single, canonical definition of what this company is. We built their Brand Vault — a structured document defining:

• Category: Enterprise workflow automation (not project management, not productivity)
• ICP: Operations leaders at companies with 500–5,000 employees
• Use case: Cross-department workflow orchestration, compliance automation, process standardization
• Competitive set: ServiceNow, Pega, Appian (not Trello, Asana, Monday)
• Price signal: Enterprise contracts starting at $50K ARR

This became the measuring stick for everything that followed.

Phase 2: Detect the poisoned sources (Weeks 2–3)

Using the Brand Vault as ground truth, we systematically identified every source contributing to the miscategorization. The audit surfaced 23 sources that were actively teaching AI the wrong category, including:

• 4 directory listings with incorrect category tags
• 7 "best of" comparison articles placing them in the wrong competitive set
• 3 of their own pages with inconsistent positioning language
• The original Product Hunt listing (still live, still describing the pre-pivot product)
• Several outdated review site profiles

Each source was scored by estimated impact on model beliefs, based on domain authority and likelihood of inclusion in training data.

Phase 3: Deploy fixes at the source (Weeks 3–6)

This is where the work gets specific and unglamorous. Source by source:

• Updated Crunchbase, G2, and Capterra profiles to reflect current enterprise positioning
• Worked with the company to update their Product Hunt listing
• Submitted correction requests to publications with outdated descriptions
• Rewrote the company's own inconsistent pages (LinkedIn, about page, integrations page)
• Created structured brand-facts content on their domain designed to give AI a clear, machine-readable signal
• Published content that explicitly positioned them against their actual competitive set (ServiceNow, Pega) rather than the false one (Trello, Asana)

Results: 6 weeks to accurate AI categorization

Categorization accuracy across models:

Overall categorization accuracy: 22% → 87% at 6 weeks, 94% at 12 weeks.

Pipeline impact:

The downstream effect on sales was immediate. Once AI started describing the company correctly, the type of buyer arriving through AI channels changed.

• Misfit demo rate: 40% → 11% (reduced by 73%)
• Demo-to-qualified rate: 58% → 89%
• Average deal size from AI-referred leads: $72K (vs. $34K pre-fix)
• Sales cycle for AI-referred leads: 28 days shorter than average

The VP of Sales told us: "It was like someone fixed the address on our building. The right people started showing up."

Why this matters for your company

Wrong categorization isn't a minor nuisance — it's a direct pipeline tax. Every misfit demo that comes through your door because AI described you wrong costs real money: sales rep time, demo prep, follow-up sequences that go nowhere.

And the problem is getting worse. As more B2B buyers use AI for vendor research, the cost of AI miscategorization compounds. You don't just lose one deal — you lose every deal where the buyer's first impression of your company came through a confused AI response.

The fix isn't more content. It's accurate content, placed at the sources where AI forms its beliefs, measured against a structured ground truth.