AI Strategy for Incumbents: How Established Companies Should Actually Respond to AI

The Three Threat Vectors

When people say "AI is disrupting our industry," they almost never mean one thing. There are three distinct ways AI threatens an incumbent, and they require completely different responses. Most incumbent strategies fail because leadership treats them as one fuzzy problem and ships a "horizontal AI initiative" that addresses none of them.

Business Model Threat

AI changes what customers are willing to pay for. Stock photo libraries (Shutterstock, Getty) priced per image — but Midjourney generates infinite variations for $30/month. The unit of value collapsed. If your revenue model assumes scarcity that AI eliminates, this is your vector.

Distribution Threat

AI changes where customers buy or how they discover. Google search drove ~50% of e-commerce discovery; ChatGPT and Perplexity are eating the top of that funnel. If your product depends on a discovery channel that AI is reshaping, the threat is upstream of your product entirely.

UX Threat

Your core value prop is intact, but the interface is dated. Customer support tools still work — but if a competitor's AI agent resolves 70% of tickets without a human while yours requires 5 clicks, your churn rate is going to tell you. This is the most common vector and the most fixable.

The mistake nearly every incumbent makes is responding to a UX threat with a business model overhaul, or vice versa. A UX threat doesn't require repricing — it requires shipping better UX. A business model threat can't be solved by adding a chatbot. Get the diagnosis wrong and you'll burn 18 months and a lot of executive credibility.

Diagnostic: Which Vector Are You Facing?

Run this diagnostic before any AI strategy meeting. Three questions, and they must be answered in order. For deeper analysis of how disruption signals show up in your metrics, see our AI Disruption Response Playbook.

Q1: Can a customer get 80% of our value from an AI tool for less than 10% of our price?

If yes, you have a business model threat. Examples: copywriting (Jasper vs ChatGPT), basic legal review (Harvey vs $400/hr associates), customer-support FAQs. The fact that you're 'better' won't save you — most customers don't need 'better,' they need 'good enough and cheap.'

Q2: Are customers discovering and choosing alternatives through new AI-native channels?

If your category's traffic is migrating from Google to ChatGPT/Perplexity, or if buyers now arrive having already short-listed three vendors via AI research, you have a distribution threat. Marketing budget reallocation is required, not product strategy.

Q3: Are competitors shipping UX that makes our interface feel slow, manual, or dated?

If your churn cohort analysis shows accelerating losses among 'power users' citing 'workflow speed' or 'too many clicks,' you have a UX threat. The fix is fast: ship AI-native workflows inside the product you already have.

Order matters

Always answer Q1 first. If you have a business model threat, fixing UX is rearranging deck chairs. If you have a UX threat, repricing breaks the product. Misordered responses are the #1 way incumbents waste their AI budget — we see this in every Fortune 500 we advise.

Response Playbook Per Vector

Each threat vector has a different right answer. The framework below is what we use with incumbent product leaders — many of whom come into the masterclass with a vague mandate to "be more AI" and leave with a specific 12-month playbook.

Business Model Threat → Cannibalize Yourself First

What happens: If AI commoditizes your core revenue, you must ship the commoditized version before someone else does. Adobe did this with Firefly: instead of letting Midjourney eat Photoshop's image-creation use case, they shipped generative fill inside Photoshop and bundled Firefly credits into Creative Cloud. They protected the subscription by absorbing the threat.

PM Implication: Set up a separate P&L for the cannibalizing product. Protect it from the legacy revenue committee. Adobe's mistake (briefly) was pricing Firefly outside the existing subscription — they corrected within 6 months by bundling.

Distribution Threat → Become the AI Channel

What happens: If AI is eating your top-of-funnel, the only durable response is to be where the AI sends users. Intuit's TurboTax did this aggressively in 2024-2026 by becoming the structured-data answer for tax questions in ChatGPT and Perplexity via partnerships and content optimization. Booking.com and Expedia made similar moves on travel queries.

PM Implication: This is 70% marketing/partnerships and 30% product. The product change: expose your data and workflows via clean APIs so AI assistants can quote and link you. SEO-for-AI (sometimes called GEO) becomes a discrete function with its own budget.

UX Threat → AI-Native Workflows Inside the Existing Product

What happens: Microsoft Copilot is the canonical example. Excel, Word, Teams — the value prop is unchanged, but the interface gained natural language. Notion's AI does the same: blocks haven't changed, but you can now generate and edit them with natural language. The product is recognizable; the UX is faster.

PM Implication: Don't ship a separate AI product or a chatbot stapled on the side. Embed AI into the workflows your customers already do. Measure success in tasks-completed-per-session, not 'AI feature adoption.'

Build Your Incumbent AI Playbook

The masterclass walks established-company PMs through the diagnostic-to-roadmap process — taught live by a Salesforce Sr. Director PM who has run this exact exercise inside Fortune 500 product orgs.

The "AI Tax" vs "AI Moat" Decision

Every AI feature you ship falls into one of two categories. "AI tax" is the cost of staying in business — features customers expect because everyone has them, but that don't differentiate. "AI moat" is the rare AI investment that creates a durable advantage. Most incumbent AI roadmaps are 90% tax and 10% moat. The successful ones invert that ratio. For more on building moats, see our companion piece on defensive AI strategy.

AI Tax: Generic chatbot in your app

Customers expect it. Won't move the needle on retention. Ship the cheapest implementation that doesn't embarrass you — pure GPT-4o-mini with light system prompt is usually fine. Don't spend 9 months building it.

AI Tax: Auto-generated summaries

Every SaaS tool has one. Treat it as table stakes. Budget: weeks, not quarters.

AI Moat: AI that uses your proprietary data

Salesforce's Einstein advantage isn't the model — it's that the model is fine-tuned on a customer's CRM history that no competitor can replicate. Intuit's tax-advice AI is moated by 25 years of tax-filing data.

AI Moat: AI that learns from in-product user behavior

If your product creates data exhaust (corrections, edits, approvals) that improves your AI over time, every user makes the next user's experience better. This is the only true compounding advantage.

AI Moat: Workflow-level integration

Microsoft Copilot's edge isn't model quality — it's that it sits inside Excel/Outlook/Teams and operates on your live business data. The integration depth is the moat.

The discipline: for every AI feature on your roadmap, force a one-sentence answer to "is this tax or moat?" Tax features should be cheap, fast, and forgotten. Moat features deserve real investment, executive air cover, and a 3-year horizon.

Real Incumbent Moves: What Worked, What Didn't

The companies below have all been case-studied to death — but the strategic moves underneath them are still poorly understood. Here's what each one actually did, why it worked, and the lesson for your own playbook. The make-vs-buy question that runs underneath all of these is covered in detail in AI acquisition vs build strategy.

Adobe Firefly — Cannibalize early

Adobe was facing a clear business-model threat from Midjourney and DALL-E. Their response: ship Firefly trained on licensed-only data (a moat: enterprise-safe IP), bundle it into Creative Cloud at no extra cost initially, then transition to a credit system. Result: Creative Cloud renewals held, and Firefly became the enterprise-safe choice.

Microsoft Copilot — Distribution wins

Microsoft's $13B investment in OpenAI wasn't about the model — it was about distribution. Copilot ships inside Office/Windows/GitHub, products with 1B+ users. Even a mediocre AI feature inside a tool used daily by half the planet beats a brilliant standalone app.

Notion AI — Embedded, not separate

Notion didn't launch a separate AI product. They embedded AI generation and summarization inside the existing block editor. UX threat → UX response. Result: $10/user/month upsell that materially moved the per-seat ARR up.

Intuit — Vertical depth wins

Intuit's Intuit Assist (and now TurboTax Live with AI) leverages 25 years of structured tax-return data. A generalist model can't compete on accuracy because it doesn't have the training data. Vertical depth is the most durable moat for incumbents.

Chegg — Cautionary tale

ChatGPT essentially commoditized homework help. Chegg's stock fell ~90% in 18 months. Their AI response (CheggMate) came too late and was too undifferentiated. Lesson: if you have a clear business-model threat, you cannot wait. Ship the cannibalizing product on a 6-month timeline, not 24.

Salesforce Einstein — Moat from data

Salesforce's bet: customers won't move their CRM data to a generalist AI. So Einstein operates inside Salesforce, fine-tuned per customer org. This is the data-gravity play — and the reason incumbent SaaS with deep customer data has more defensible AI than pure-play AI startups.

The pattern across every successful incumbent move: they correctly diagnosed which threat vector they faced, then built a response that fit that vector — not a generic "AI initiative." If you're staring at a 40-page AI strategy deck that doesn't name your specific threat vector, the deck is wrong. For more on enterprise-scale execution, see AI enterprise strategy.