
AI Adoption
About this brief
What is the ECLC? The Executive Council for Leading Change (ECLC) is a global, peer-led community of more than 1,200 enterprise transformation leaders. Members meet in regular virtual roundtables of around 45 executives, held under Chatham House rules, to share what is working, what is not, and how they are handling change.
How Tigerhall fits. ECLC is convened by Tigerhall, the transformation delivery platform for a world of always-on change. Tigerhall turns a companyโs strategy into personalized change journeys for each employee, delivered in the flow of work (Teams, Slack, Outlook, mobile) in 33 languages, with real-time data on where the change is landing and where it is stalling. Leaders own the strategy and the plan; Tigerhall runs the delivery, so a team of two or three can deliver what used to take fifteen to twenty.
What is an executive brief? An ECLC executive brief is a short, anonymised write-up of a single roundtable. It captures what the room discussed, where leaders agreed and differed, and what to do next. Each opens with a summary, then themed sections, then eight key takeaways.
Executive Summary for Evaluating AI Risk and Ensuring Successful Implementation
Leaders from finance, operations, HR, healthcare technology and change management compared how they are scaling AI while managing its risks. The discussion started from a familiar four-part framework (data, controls, governance and model risk) and quickly widened. The risks that mattered most in practice were human: weak AI literacy, vague purpose, siloed experiments and leaders who cannot yet use what AI produces.
The consensus was that AI is an intelligence partner to be embedded deliberately, not a technology to be bolted on. The organizations managing risk best map the process first, set up governance before buying tools, keep a human in the loop where stakes are high, and test tools on real data before committing.
The Risks Leaders Named First
A four-part starting frame. Data quality and history, controls, governance and ownership, and the risk of the model itself (bias, accuracy, need for ongoing review).
Culture and capability. Poorly communicated change forces backtracking, and people may not be ready. Skills-gap analysis often comes after the tools are already deployed.
Missing performance metrics. Many companies invest in AI without measuring its impact, which slows adoption.
Whole operating model. Strong technology introduced without considering ripple effects across roles, processes and structure causes damage elsewhere.
Unclear purpose. Vendors pitch constantly, yet few can explain in plain terms what their product does or why a team would use it.
Skill erosion. Over-reliance can degrade the judgment and capability of the people it was meant to support.
Technology versus intelligence. The biggest pitfall is a technology strategy instead of asking how more intelligence could change roles, structures and how value is created.
Leaders admitting what they do not know. Intangibles such as this are harder than building governance, and siloed experiments follow when it is missing.
Where Not to Deploy AI, and Where to Keep Humans In
Never overlay AI on an undocumented process. One organization will not apply AI to any process without a full current-state and future-state map, so AI is placed at steps that actually improve the workflow, with upstream and downstream effects considered.
Fear of missing out is the wrong driver. Many are adopting AI so as not to fall behind, rather than mindfully.
Regulated areas need boundaries. Highly regulated work (privacy rules across jurisdictions, complex accounting judgments, legal research) needs clear limits on what AI may do. A story of fabricated legal precedent illustrated the danger.
Human in the loop is not the same as no deployment. Where legal or compliance risk is high, use AI but keep human review of decisions.
Too much, too soon. One leaderโs experience was the reverse of change resistance: too many people using AI for tasks it cannot do well, producing poor outputs and making them look foolish.
Align people and intelligence. One example cited was a company where the chief people officer also holds the AI portfolio, tying human capital to the intelligence layer.
Where AI Is Delivering Value With Manageable Risk
Repeatable operational workflows. Contact-center and operations processes with clear if-then logic have been highly valuable, with humans in the loop as sophistication grows.
Collections and multilingual work. An AI agent that switches languages, keeps centralized structured notes, and supports human reviewers can remove dependency on a single champion. It lets leaders compare onshore, offshore and AI pipelines.
Start broad. Introduce people to high-volume, high-frequency use cases (meeting notes, drafting, data analysis) before drilling into specific roles.
Optimization problems. With large data sets and defined outcomes, models beat gut feel and recency bias, though it took two to three years and parallel running before people trusted the output.
Product and delivery roles. Teams using AI for user stories and testing got work out faster, freed time to collaborate with the business, and gained confidence working across languages.
Role analysis. Recording work with consent and analysing it for themes helped standardize a role, move tasks to AI, and reduce burnout. Psychological safety for participants was the main risk.
Know the cost. Pair every implementation with lessons learned on what it cost, so value is weighed against risk with eyes open.
Governance That Worked, and Where It Falls Short
An AI governance council. Every new tool is assessed before sign-up, the requester signs for the risk, and the vendor is asked how the risks will be mitigated. Vendor consolidation is a side benefit and it slows teams from moving too fast.
Sequence matters. One leader implemented a tool before the council existed and had to retrofit, which cemented the value of going through it first.
Prioritization is still immature. Stack ranking tools often defaults to whoever is loudest at executive level.
Wild west reality. Where there is no enterprise direction, teams build siloed tools and self-govern, for example by hiring a tooling strategist to publish which tool fits which use case.
Biggest worry. Non-technical teams pulling unvalidated numbers from large data sets, with no enforced way of working.
The Hardest Part in Practice
Governance, adoption and change management are harder than the technology.
Demo versus reality. A vendor demo of reconciliation looked great, but the real data volume overwhelmed the tool. Test with representative volume and complexity (including multiple currencies and entities) before signing.
Decide your appetite. Be explicit about whether you are willing to pilot an immature tool with the vendor or want a polished, quick implementation.
Leaders cannot yet use the intelligence. Operations deploy tools and get output, but leaders are not AI-literate enough to act on it. An AI champion role could cascade literacy across levels.
Balancing Automation and Human Oversight
Hybrid is the destination. The strongest organizations will balance high-volume automation with AI-savvy specialists.
Build digital fluency. Teach people to sort work into data and prediction, generative, and agentic or automated buckets, and to see what is inherently human.
Looking Two to Three Years Ahead
Trusted data beats most AI usage. Differentiation will come from reliable, trustworthy data, not from who uses AI most.
Weave AI into the workforce. Round-the-clock capability across regions frees teams for higher-value work.
Connect the why of each function to the why of the organization.
Key Takeaways
Treat AI as intelligence to be embedded in the operating model, not a bolt-on technology.
Build AI literacy for leaders first, because most risks come from unclear purpose.
Document current and future-state processes before applying AI.
Stand up governance before tool purchases, with sign-off and vendor accountability.
Keep a human in the loop wherever legal, compliance or high-value decisions are at stake.
Clean the data first, and test vendor tools on real volume and complexity.
Measure results with agreed metrics, and pair each rollout with a record of what it cost.
Prepare leaders to use AI output, with champions to cascade fluency.



