Why ESSAM was built with a human in the loop — and why Harvard just published the framework that proves it
Share
Why ESSAM was built with a human in the loop — and why Harvard just published the framework that proves it
Every serious AI product has to answer one question before it earns the right to operate inside a real enterprise: who makes the decision when the model is wrong?
Most enterprise AI tools shipping in 2026 dodge this question. They ship as if the model is always right, hide the confidence problem behind clean UX, and quietly transfer the accountability to whoever pressed the button.
ESSAM was designed on the opposite assumption. In process transformation — the layer where strategy either becomes real or stays on a slide deck — most decisions cannot be delegated to a model. The stakes are too high and the knowledge required is too tacit. So we built ESSAM as a human-in-the-loop system from the first design meeting. Not as a compliance patch. As the architecture.
Then, in November 2025, two Harvard Business School professors published the framework that explains, with academic precision, exactly why that design choice was the right one.
The framework Harvard just gave every executive
In The Gen AI Playbook for Organizations (HBR, November 2025), Bharat N. Anand and Andy Wu argue that every task in an enterprise can be located on a 2×2 built from two questions:
- What is the cost of an error? If AI gets it wrong, is the consequence a rewritten paragraph — or a regulatory fine, a reputational crisis, a broken process nobody can unwind?
- What type of knowledge does the task require? Explicit data that can be documented, stored, and retrieved — or tacit knowledge, the kind built through lived experience and difficult to codify?
Plot those on the same page and four zones emerge — each with a fundamentally different rule for how AI should be used.
The four zones:
- No regrets (explicit data, low cost) — AI does it all. Résumé screening, meeting summaries, low-dollar approvals.
- Quality control (explicit data, high cost) — AI drafts, humans verify. Contracts, code, compliance documents.
- Creative catalyst (tacit knowledge, low cost) — AI generates options, humans select. Marketing copy, product concepts, brainstorming.
- Human-first (tacit knowledge, high cost) — Humans lead, AI assists with minor tasks. Setting strategy. Integrating enterprise systems. Making disciplinary decisions.
Now here is the strategic question every enterprise transformation program has to answer: which zone does process transformation actually live in?
Process transformation is a Zone 4 problem — mostly
The honest answer is that process transformation lives across more than one zone. But most of the decisions that make process transformation succeed or fail sit squarely in Zone 4 — the human-first zone.
Consider what actually happens when we redesign a real process inside a bank, a healthcare provider, or a manufacturer:
- An "extra approval step" looks non-value-adding on paper — but exists because a regulator requires it and the frontline team is the only source of that context.
- A step gets flagged as waste — but the operator knows it prevents an error class that only surfaces once a year.
- A proposed redesign looks cleaner — but violates a cultural norm that would tank adoption on day one.
- A future-state design looks efficient — but requires a handoff between two departments who have not spoken in five years, and no AI can read that political map.
All of these are tacit-knowledge decisions with high error costs. A process transformation platform that acts autonomously in this zone will confidently delete the wrong step, propose designs that cannot be implemented, and hand the organization a beautifully documented failure. This is exactly why so many process-mapping tools produce impressive artifacts that never get adopted.
ESSAM was designed for the reality Anand and Wu are describing. Its architecture treats every process-transformation task as Zone 4 by default — and only relaxes that assumption when the user explicitly says the specific task is safe to automate.
How the human-in-the-loop is wired into ESSAM
Design principles are cheap. The test is whether they show up in the product's actual mechanics. Here is where humans stay at the center of the decision inside ESSAM, step by step.
1. ESSAM asks before it acts.
When a user brings a process to ESSAM for optimization, the first thing ESSAM does is decide whether it has enough information to do the job responsibly. It reviews what the user has provided, identifies gaps, and asks for more input if the picture is incomplete. It will not proceed to map a process it does not adequately understand — because a confidently-mapped process built on missing information is worse than no process map at all.
This is Zone 4 discipline. The cost of an incorrect map is high, and the knowledge required to correct the picture is tacit — it lives with the user, not the model.
2. Every step is confirmed before ESSAM moves on.
As ESSAM baselines the current-state process, it maps each step and shows it to the user for confirmation before advancing. Not at the end. Not as a batch review. Step by step. The user confirms the sequence, the actors, the inputs, the outputs, and the handoffs before the model treats any of them as fact.
This structurally prevents the failure mode where a hundred-step process gets mapped from a single user prompt, then the user is asked to review a wall of text they will inevitably skim.
3. Users can inject tacit knowledge that overrides the model's logic.
This is the feature that matters most in Zone 4, and the one most process-mapping tools get wrong. When ESSAM flags a step as non-value-adding waste, the user can respond: "Keep this step. It is mandatory. Regulator requires it. Non-negotiable."
ESSAM accepts that override, tags the step with the user's rationale, and preserves it in every downstream artifact — the SOP, the RACI, the process map, the FMEA. The tacit knowledge the user carries — regulatory context, institutional memory, cultural constraint — never gets lost inside the transformation. It becomes part of the design record.
This is exactly what Anand and Wu describe as the correct Zone 4 pattern: AI expands the human's capacity to perform the task without undermining that person's control of the decision.
4. ESSAM drafts. Humans add. The human owns the document.
When ESSAM produces a procedure or SOP, it drafts it — it does not finalize it. Users can add sections, insert steps, add exception handling, override formatting choices, and expand any part of the document the model wrote too tersely. The final artifact is a document the user shaped, not a document the model handed them.
Why does this matter? Because SOPs that staff will actually follow are almost never the SOPs the model wrote alone. The user knows which edge cases matter, which language matches the organization's tone, and which sections need more detail than a general-purpose model would allocate.
5. Reviewer additions get pressure-tested — in both directions.
During procedural review, reviewers can add steps ESSAM missed. But ESSAM does something most tools do not: it flags additions that look like non-value-adding waste and asks the reviewer to justify them.
This is the two-way version of human-in-the-loop. The human can override the model — as they should. But the model can also push back on the human when the human is adding process debt. Neither party is a rubber stamp for the other. Both are accountable.
The organizational effect is significant. Process bloat — the slow accretion of "just add a check here" steps that erodes efficiency over years — is one of the most expensive failure modes in enterprise operations. ESSAM makes that accretion visible in the moment it happens, not five years later during a Lean audit.
6. Risk identification runs the same collaborative loop.
Before ESSAM performs FMEA (Failure Mode and Effects Analysis) on a new process design, it identifies the implementation risks it can infer from the process structure and the user's context. Then it stops and asks: "Are there other risks you want to add before we run FMEA?"
Because the operator knows things the model does not. That the country manager who owns this process is retiring in Q3. That the IT system that carries step 7 is scheduled for migration. That the regulator flagged a similar process at a competitor last month. That the team lead responsible for step 4 is a single point of failure.
None of that is in the process data. All of it is Zone 4 knowledge. ESSAM asks — because asking is what a human-in-the-loop system does before running a critical calculation.
Where ESSAM lets AI run — and where it stops
Not every part of ESSAM's cycle requires the same level of human oversight. The seven-step AI Lean Transformation Cycle deliberately maps to the different zones of the Anand-Wu framework:
- Baseline the current process — Zone 4. Human-led. ESSAM asks, confirms, and preserves tacit context.
- Analyze for waste — Zone 2/4 hybrid. ESSAM produces the Lean waste audit; the user validates each flagged waste against tacit context (regulatory, cultural, operational).
- Optimize the design — Zone 4. ESSAM proposes; the user shapes, approves, and owns the future-state design.
- Document and approve — Zone 2. ESSAM drafts the SOPs, maps, and RACI; humans verify and add.
- Deploy via WhatsApp — Zone 1. Once the human has approved the process, deployment is mechanical and automated.
- Collect staff feedback — Zone 1. Structured collection through the WhatsApp agent, no judgment call required by the tool.
- Repeat the cycle — Zone 4. The re-baselining step returns to human-in-the-loop, because the reasons a process drifted are always tacit.
The zones are not decorative. They are the operating architecture. Every automated step is automated deliberately. Every human-in-the-loop step is there because the cost of a wrong decision at that step is too high — and the knowledge required is too tacit — to be handed to a model.
Why this matters for the buyer's decision
If you are evaluating an enterprise process-transformation platform right now, the single most important design question is not "how good is the model?" It is: where does the platform draw the line between AI autonomy and human authority, and can it defend that line?
The Harvard framework gives every buyer the vocabulary to ask that question rigorously. And it validates a design principle we committed to before the framework existed: process transformation is a Zone 4 problem, so ESSAM is a Zone 4 tool.
A platform that acts autonomously on Zone 4 decisions will damage the operation it was supposed to fix. A platform that treats every task as Zone 1 will produce impressive artifacts that never get adopted. And a platform that pretends AI is always right will quietly transfer accountability to whichever executive signed the contract.
ESSAM was built on the opposite premise. AI expands the human's capacity. The human owns the decision. The record shows who decided what and why. And every step of the transformation carries the tacit knowledge that made the design defensible in the first place.
That is not a compliance feature. It is the reason organizations that adopt ESSAM end up with transformations that actually stick — while 75% of business transformations, according to BCG, fail to deliver long-term change. The difference is not the model. The difference is where the human sits in the loop.
See it in your own operation
If you want to see exactly where ESSAM keeps you in the loop and where it accelerates the work — mapped against a real process from your own operation — a 60-minute demo is the fastest way.
60 minutes. One real process from your operation. A concrete redesign proposal — and a walkthrough of exactly where the human owns the decision — by the end of the call.
About the author. Abdulla Al-Awadi is founder of TheStrategist.me and ESSAM.AI, and Chief Strategy Officer at Kuwait International Bank. TheStrategist.me gives business leaders the strategy OS they need to develop rigorous strategy. ESSAM.AI gives them the process transformation platform they need to execute it.
Source. Anand, B. N., & Wu, A. (2025, November). The Gen AI Playbook for Organizations. Harvard Business Review. hbr.org/2025/11/the-gen-ai-playbook-for-organizations