Summary of "Webinar: The AI-first workflow redesigning for growth"
Who/what this is about
- Stefan Ohran (Board of Innovation) presents an approach to AI transformation centered on redesigning knowledge work, using an analogy to how factories re-engineered workflows for robots.
- Board of Innovation positions itself as an AI transformation studio focused on growth, with three topic areas:
- Value creation (how AI changes products/services/business models)
- Work redesign (today’s focus): redesigning knowledge work for AI capability
- Operating model (how to organize to scale opportunities)
Core concept: AI-enabled vs AI-first (two transformation paths)
AI-enabled (retrofit)
- AI is added onto the existing human-driven workflow
- Gains are often limited to task-level efficiencies
- Does not scale well over time; “missed opportunities”
AI-first (redesign)
- Work is redesigned around AI capabilities
- The engine shifts:
- AI-enabled: human = engine, AI accelerates steps
- AI-first: AI = engine, human intervenes at key moments
- Requires changing growth, operations, and competition with AI “at the core”
Concrete example: RFP response process redesign
Traditional (AI-enabled retrofit)
- Linear, SME-heavy workflow:
- RFP intake (inbox or active BD)
- Context gathering
- Qualification (capabilities/credentials/selection criteria)
- Compliance checks
- Bid/no-bid decision (senior meeting)
- Proposal definition: pricing, drafting sections, QA, submission
- Headcount scales similarly to the old model
- Example impact mentioned: cycle time could improve from ~2 weeks to ~1 week (marginal gains vs structural redesign)
AI-first workflow (three phases; throughput ~3 hours total)
-
Instant intake (fully autonomous) ~1 hour
- Parse RFP, extract requirements
- Cross-reference CRM + past wins for context
- Generate scorecards/elements needed for bid/no-bid
- Run compliance checks, etc.
-
Parallel generation (AI + human in the loop) ~1 hour
- Start proposal drafting early (cheap to iterate)
- Draft all sections; build pricing model from past data
-
Review & submit (human driving + AI assistance) ~1+ hour
- Human reviews for fit/comfort and prepares final submission
- AI performs post-facto consistency audits
Frameworks / playbooks highlighted
“Factory lessons” translated to knowledge work
-
Value stream mapping
- Map flow between “workstations” → in knowledge work, map decisions and information dependencies
- Identify what decisions are waiting for (reveals real waste)
-
Task allocation
- Decide for each task: human vs machine
- Framed as a multi-dimensional allocation problem
-
Jidoka (Toyota principle) / escalation logic
- Machine handles the norm, human handles exceptions
- Build triggers for anomalies so AI stops/escalates when needed
6-step methodology to move from workflow → AI-first workflow
Phase 0: Prioritize
- Step 0: Set priorities for which workflow to redesign first
- Use a 2x2 matrix:
- Strategic leverage vs Readiness
- Target: high readiness + high leverage opportunities (argued to build a competitive “moat”)
Phase 1: Map decisions
- Step 1: Map the decision flow
- Trace decision inputs/outputs and transitions (not just activities)
- For RFP: bid/no-bid, pricing, quality, sign-off—each waits on specific information
Phase 2: Allocate + govern
-
Step 2: Define constraints + allocate work to one of three archetypes:
- Autonomous AI (AI executes alone)
- AI + human in the loop (AI drives; human provides feedback)
- Human + AI assistance (human drives; AI assists)
-
Allocation is guided by questions:
- Determinism: how specifiable is the task, and what’s the cost of error?
- Judgment: does it require tacit knowledge / judgment?
- Recurrence: is it recurrent vs non-recurrent?
-
Step 3: Define escalation logic (Jidoka)
- Examples of escalation triggers in RFP compliance:
- Unknown requirement
- Borderline match (e.g., coverage limit 4.8M vs requirement 5M)
- Contradictory clauses in the RFP
- Examples of escalation triggers in RFP compliance:
Phase 3: Design human role + close the loop
-
Step 4: Design the human role (upgrade, not demotion)
- Use a Toyota-inspired guardrail:
- Automation needs to be < ~8% for certain key processes (to keep human engagement/expertise)
- Human responsibilities shift toward:
- Maintaining domain expertise
- Concentrating on judgment-intensive work
- Providing feedback loops to improve the system
- Use a Toyota-inspired guardrail:
-
Step 5: Measure the process and reallocate continuously
- Metrics remain similar; observability improves
- As models improve, the boundary of AI autonomy shifts, so allocations must be updated
Metrics / KPIs mentioned
No explicit numeric targets were provided, but the talk emphasizes central metric types (especially for RFP):
- Win rates
- Gross margin
- Pricing quality / pricing decisions
Also implied in task allocation:
- Risk/cost of error (used as part of the cost model, not just time savings)
Key takeaways (business implications)
- AI isn’t just workflow optimization—it’s workflow redesign (decision-making and system design).
- The real gains come from allocation: assigning decisions to the right agent (human vs AI) to maximize quality vs cost (including risk).
- Competitive advantage comes from system design, not merely adopting newer tools.
Actionable “tomorrow morning” recommendations
- Pick one workflow and run an AI-first redesign thought experiment:
- Does it shrink drastically and become end-to-end AI-handled?
- Is the human role elevated or minimized/removed for some segments?
- Does the workflow change form (e.g., shift to AI-aided exception triage)?
- Are there new workflows now possible because AI can act as a cognitive agent?
Presenters / sources
- Presenter: Stefan Ohran, Managing Director, Board of Innovation (AI transformation studio)
Category
Business
Share this summary
Is the summary off?
If you think the summary is inaccurate, you can reprocess it with the latest model.
Preparing reprocess...