Part 3: Assess AI Capabilities
Know what AI can actually do before you buy what a vendor says it can do.
A vendor promises:
“Our AI will write all your social media content, personalize it for each audience segment, and automatically post at optimal times.”
Sounds great. But can AI actually do that?
The AI Capability Ladder
AI can do some things brilliantly. Some things poorly. And some things not at all.
AI Does This Better Than Humans
What AI can do:
- Data pattern recognition across massive datasets
- Image classification and object detection
- Transcription and language detection
- Generating variations at scale
Real-world examples:
What it still can't do:
- Understand WHY patterns exist
- Make strategic decisions based on those patterns
AI Does This Well (With Human Oversight)
What AI can do:
- Drafting content from outlines or briefs
- Summarizing long documents
- Generating creative concepts and brainstorming
- Answering frequently asked questions
Real-world examples:
What it still can't do:
- Understand brand voice without extensive training data
- Know what NOT to say in sensitive contexts
AI Does This Inconsistently
What AI can do:
- Long-form strategy documents
- Original creative concepts
- Predicting new market opportunities
- Maintaining consistent brand voice across formats
Real-world examples:
What it still can't do:
- Replace a senior strategist's judgment
- Understand unwritten brand rules and cultural context
Note: Depends heavily on prompt quality. Requires significant human editing.
AI Does This Poorly
What AI can do:
- Navigate company politics and internal dynamics
- Reading between the lines of stakeholder feedback
- Knowing when to break brand rules intentionally
- Generating truly novel, breakthrough ideas
Real-world examples:
What it still can't do:
- Access implicit organizational context
- Read the room in real-time situations
AI Can't Do This At All
What AI can do:
- Building genuine relationships with stakeholders
- Negotiating budgets, timelines, and priorities
- Defending creative decisions under pressure
- Knowing when to kill a project
Real-world examples:
What it still can't do:
- Replace human judgment in ambiguous situations
- Provide empathy or leverage political capital
Test Your Vendor's Claims
Use the capability ladder to evaluate what vendors are really selling you.
Vendor Claim:
“Our AI will fully automate your content calendar—no human input needed.”
Content writing = Level 4 (AI does this well with oversight)
Content strategy = Level 2-3 (AI does this poorly to inconsistently)
Verdict: Overselling.
AI can draft content, but “fully automate” with “no human input” ignores the strategy, planning, and quality control that require human judgment.
Vendor Claim:
“AI analyzes all your customer data and tells you exactly which segments to target.”
Pattern recognition = Level 5 (AI does this better than humans)
Strategic prioritization = Level 2 (AI does this poorly)
Verdict: Half true.
AI excels at finding patterns in data, but deciding which segments to prioritize requires strategic judgment, budget context, and business goals that AI cannot evaluate.
Vendor Claim:
“Our AI writes email campaigns that match your brand voice perfectly.”
Content drafting = Level 4 (AI does this well with oversight)
Brand voice matching = Level 3 (AI does this inconsistently)
Verdict: Possible—but requires training data and human editing.
With enough examples of your brand voice and a human editor reviewing output, this can work. But “perfectly” is an overstatement.
Key Takeaway
Before buying any AI tool, map the vendor's promises to the capability ladder. If they claim Level 4-5 results for Level 2-3 tasks, they're overselling.
AI is best at data-heavy, pattern-based tasks. It struggles with strategy, nuance, and anything that requires organizational context.
The right question is never “Can AI do this?” It's “At what level can AI do this, and what human support does it need?”