Zapier’s latest enterprise AI survey makes it plain: 80% of organizations can’t get their AI pilots to play nicely with legacy systems. The temptation is to blame the models. The real reason is a lack of process design and the right foundations. The challenge isn’t legacy technology, it’s adapting workflows to support intelligent systems. If your data is splintered, your governance is fuzzy and your change muscle is weak, AI will amplify the chaos instead of bringing value.
The mandate is bigger than “integrate a chatbot”. You have to rebuild the value chain so that data, automation and human judgment actually work together. That means rethinking old habits, setting non-negotiable guardrails and teaching teams to move within this new world.
1. Kill the legacy logic
Declare a zero-base mindset: Too many leaders treat AI as a bolt-on feature instead of a reason to redraw the process map. Start with the customer or employee outcome you want, then design the flow that delivers it with AI as a native capability. If a step exists only because “we’ve always done it that way”, it needs to go.
Design for connected intelligence: Unified data is the only way AI can make context-aware decisions. Build a shared data layer with consistent taxonomies, business and technical lineage, and data stewardship. That foundation fuels everything else, from predictive maintenance to automated underwriting. Without it, you’re just automating blind spots.
Fix the plumbing before the AI: Minimum viable compatibility means API-first architecture, clean integration patterns, cloud elasticity and observability baked into every service. Security belongs in the same sentence. Encryption, identity management, zero-trust policies. Those are table stakes. If you can’t put your current stack through a security review on a Monday morning, don’t hand it to an AI model on a Friday night.
2. Build guardrails before you floor it
Everything should be made as simple as possible – nothing more, nothing less. AI governance works the same way: codify enough guardrails so teams can move fast without blowing up trust.
Clarify the legal terrain: Your general counsel shouldn’t be the last to hear about a new AI workflow. Formalize how privacy laws, sector regulations and emerging AI acts influence design constraints. Document the boundaries and keep the register live. Your compliance posture is only as strong as the least-informed product team.
Treat data governance as oxygen: Quality, technical lineage, retention, access, ethics – spell out the rules and automate enforcement wherever possible. Build data products with explicit owners and service-level agreements. Make it unmistakably clear where the master version lives, because maintaining data in the wrong system doesn’t just slow teams down, it compounds risk. Make transparency the norm: if people can’t see how AI reached a decision, they won’t trust it.
Assign process owners and break down siloed ways of working: Every data product needs a single accountable leader, with a clear definition of who monitors performance, signs off on changes and how often metrics are reviewed. All process owners need to work together in one end-to-end value stream to optimize the entire process. Good governance is the operating system for reliable automation.
Operationalize ethics: Bias testing, fairness audits and escalation paths belong in the delivery plan. Train your teams on responsible AI scenarios and create a cross-functional ethics council empowered to veto new releases.
Revamp the objective system: When AI processes change, incentive frameworks should follow to recognize end-to-end value creation, not just volume. It should reinforce new performance expectations and reward the outcomes that AI is designed to unlock.
3. Make transformation feel human
AI transformation fails when it’s built without empathy: for the people using it as much as the people it serves.
Lead with a change narrative. Explain what’s changing, why it matters, and how success will be measured. Communicate the upside and the safeguards. Make leaders visible, accessible and accountable. Invite feedback and show how it shapes the roadmap.
Invest in tools, but also in teams. Upskill teams in AI literacy, data interpretation, prompt design, and scenario planning. Pair training with hands-on pilots. Create communities of practice where employees can swap lessons learned. Reward experimentation and ethical judgment. Both are equally important.
Re-architect roles and rituals. AI shifts decision points, handoffs, and KPIs. Update job descriptions, performance metrics, and governance forums accordingly. Introduce new ceremonies like bias reviews, model retros and data standups as that keeps empathy and accountability front and center.
Don’t automate what should stay manual
The organizations winning with AI aren’t the ones with the fanciest models but the ones with the nerve to start from a blank sheet. They take a hard look at their processes, keeping the parts that matter, fixing what doesn’t work, and cutting the rest.
They take apart legacy workflows, rebuild them around data and automation, and hard-wire governance and empathy into the core. They implement AI where it truly makes sense – based on volume, objectives and KPIs. Automation, without AI, can be much more cost effective in certain scenarios. Certain processes might make sense to remain manual. It all comes down to a careful review of all processes and the choice of what to automate, keep manual or enhance with AI.
If you want AI to be transformational, redesign the process first, then layer in intelligence. Anything less is just decorating the past.
Contributor in this article
