From 5-Year Roadmap to Reality: How AI Changed the Whole Business

In 2021, when I was planning what would eventually become PF TECH, I put together a capital budget for building TERN. Product development alone: north of $300,000. That was before marketing, sales infrastructure, operational overhead, or the cost of managing the business itself. Those were separate line items, each requiring either hired staff or paid agencies.

I looked at that number and concluded the timeline was "someday, if the organisation achieves significant scale." A real plan, but a slow one.

By 2025, I had built the core of what that plan described. The product development estimate essentially disappeared. And it turns out that development speed is actually the least interesting part of what happened.

The headlines tend to focus on development speed: AI lets one person build what used to require a team. Features ship in days instead of months. That is true, and the implications for product development are significant. But the main story is elsewhere.

What AI has done to the economics of PF TECH is an operating model story that reaches well past development. The compression extends across every function a business needs to run, and the cumulative effect of that compression is structural rather than incremental.

Here is what those budget lines look like in 2025:

• Product development: AI-assisted engineering made the $300,000 estimate functionally obsolete. Custom websites in under a week at under $100 in compute. Core integrations that previously took two weeks to configure now deploy in a day. Planning tools built in eight hours for less than $40 in compute, replacing $2,000/month in labour overhead.
• Marketing: I used to pay a marketing and Google Ads agency to manage our digital presence. I have replaced that engagement with custom AI agents that handle content strategy, ad optimisation, and performance monitoring — continuously, not on a monthly billing cycle.
• Executive operations: Our entire executive assistant function — scheduling, inbox management, meeting preparation, follow-up coordination — is now handled by agents I built. Tasks that used to require dedicated administrative support now run automatically.
• Managed IT: We have significantly reduced our reliance on managed IT support by building internal automation for monitoring, incident response, and routine maintenance.
• Sales: Custom AI agents for prospect research, proposal generation, and follow-up sequencing.
• Level 1 support: The website's AI chat assistant handles first-contact questions and escalation routing without human intervention.
• Digital media and web development: In-house, AI-assisted, at a fraction of the cost of the agency relationships we used to maintain.

This is not a story about one function getting faster. The cost structure of operating a knowledge-intensive business has been fundamentally restructured. The overhead that previously justified a minimum viable business size of several hundred thousand dollars annually is now accessible at a fraction of that cost.

Here is the part the AI productivity discourse tends to skip: the compression is not uniform.

The domain knowledge required to build TERN is not something AI can acquire or approximate — it has to be earned. Two decades, six operational domains:

• Non-profit fund accounting
• CRA compliance
• Grant management
• Donor reconciliation
• Payroll systems
• Governance practice

It can help someone with that knowledge build faster. It cannot substitute for the knowledge itself.

This is why training has become more important as AI tools have proliferated.

The tech giants have already made several of what might have been PF TECH's early product ideas obsolete. They will make more. General-purpose AI tools commoditise general-purpose problems at a pace that no small operator can match. The defensible position is not in building another general tool. It is in the specific intersection where general AI tools fail: the intersection of AI capability, information management discipline, non-profit operational knowledge, governance and risk frameworks, and sector-specific regulatory compliance.

That intersection is what PF TECH occupies, and what TERN is built for. It is also what the Mission Multiplier Program is designed to cultivate in practitioners across the sector.

The force multiplier metaphor is accurate but incomplete. A force multiplier amplifies force in proportion to the force that is input. If the input is twenty years of sector-specific knowledge, the output is significant. If the input is general curiosity and no domain depth, the output is still general.

This is why the training and advisory work is now as central to the Multiplier Model as the technology itself. TERN's technical capabilities only matter if the practitioners using them understand what the outputs mean, what the governance requirements are, and how to interpret the data in the context of their specific organisation's operational and regulatory situation.

The sector that will benefit most from AI is the one that builds the deepest competency at the intersection of AI capability and sector-specific knowledge. That competency does not come from a product. It comes from deliberate practice, ongoing learning, and the kind of peer community that compounds knowledge over time.

What I described as a five-year roadmap in 2021 is now a present-day reality. The harder question — and the one that matters more — is whether the sector is ready to use it.