Most companies won’t get much from agentic workflows by layering them onto the org they already have.
If the same approvals, same KPIs, same budget boundaries, and same handoffs still govern the work, expect faster tasks and very little enterprise uplift.
The bottleneck isn’t the model.
It’s the operating model around the work.
The Seams you are about to build. (Post 4 of 4)
Before you build the next agent, ask one question.
Does your organisation deserve to be made executable?
Because that is what you are doing. Not deploying a tool. Not automating a task. You are encoding today's version of your teams, their data, their incentives, their blindspots - and running them at scale.
If your sales team and your ops team do not share a definition of a customer today, their agents will not share one either. If nobody owns the journey end-to-end, no agent will. If your structure rewards local wins over collective outcomes, congratulations. You have now automated that.
The failures from SaaS fragmentation and data silos took years to surface. Agent sprawl will surface faster. The feedback loops are shorter. The outputs are more confident-looking. The humans in the loop are fewer.
There is no technical fix for an organisational problem. There never was.
Fix how you're organised. Then build the agents.
#OrganisationalDesign #AI #AgentArchitecture
Most enterprise AI launches follow the same script. Press release. Live demo. Someone using a chatbot in a glass-walled office.
What comes next is usually quieter.
When a vendor announces a new AI capability without disclosing active user rates, the silence is data. It tells you the product exists.
It doesn't tell you anyone is using it.
Forrester has tracked the gap between technology procurement and technology use for two decades.
It appears in CRM, in ERP, and now in AI.
Tools get bought. Habits rarely follow at the same pace.
In financial services, a customer-facing AI assistant that fails quietly doesn't just waste budget. It generates friction at the moment you were trying to remove it. Customer trust lives in what happens when someone actually uses the product - not when it ships.
The question every board should be asking of its technology vendors is not what they shipped. It's their 90-day active usage rate and how they're measuring it.
111,373 complaints to AFCA in 2025.
A record.
The top issue wasn't fraud or fees.
It was claim delays.
$643M in compensation paid out — up 120% year-on-year.
That's not a technology problem. That's what happens when you digitise the front door and leave the back room untouched.
I've seen this up close. The app is beautiful. The onboarding is slick. And when a claim drops into a queue that runs on a process map from 2011, none of that matters.
The channel is not the product.
The outcome is.
Most banks treat AI risk controls as a brake. The ones deploying fastest have discovered they're a throttle.
EY reports that 44% of UK banks now have AI systems embedded in operations, but 26% report no or limited controls ensuring regulatory adherence. That second group isn't moving faster. They're accumulating liability.
When a team can't answer "who approved this model and what were the test conditions?", the answer to every subsequent deployment question is "wait."
Clear accountability, defined testing standards and audit-ready documentation remove that hesitation. They're not overhead. They're the permission structure that lets the next deployment move at pace.
Controls that slow decisions are badly designed controls.
Well-designed controls are what let you move again.
The FCA's motor finance redress scheme is expected to cost lenders around £11 billion. The lenders built the commission structure that made it necessary.
Hidden discretionary commission arrangements let car dealers earn more by setting customers a higher interest rate. The customer didn't know this. The lender did.
Santander and Lloyds have already set aside provisions. Final scheme rules are expected in late March. Around 14 million motor finance customers are in scope.
The arc is familiar. PPI followed the same pattern: a product structure that worked for the distribution chain, not the customer, embedded deeply enough that the unwind costs multiples more than the original gain. The firms that set the commission policy are now funding the redress scheme.
The compliance question isn't just "is this legal?" It's "what happens when the customer finds out?"
In every system, complexity exists. The question is who absorbs it.
Customer first. Then colleague. Now, with agent sprawl, the user.
Hick's Law: decision time increases with the number of options.
50 agents is not empowerment. It is a design failure.
#AI#CustomerExperience
The winners will not have zero complaints.
They will run a faster complaint-to-redesign loop:
- map themes to process owners
- quantify cost and churn impact
- fix architecture before scripts
Complaints to capability, every month.
Incentives decide whether issues are solved or recycled.
If no one is accountable for root-cause complaint reduction, recurrence is predictable.
Systems behave as designed.
Your org chart is visible to customers through friction.
When someone repeats context across teams, they are experiencing internal seams as external pain.
Better design starts by removing those seams.
Customer complaints are not background noise. They are system telemetry.
If the same issue repeats, it usually points to process, policy, or handoff design debt.
Fast-improving firms learn faster from complaints than competitors.
Dubai has already shown the world how to remove friction. The next frontier is ensuring that when friction does arise, it can be resolved with the same elegance as the rest of the system @DEWAOfficial
Dubai built one of the most seamless systems of public / private integration in the world. With EmiratesID and UAE Pass, the usual red tape of residency, registration and services is almost entirely removed.
It is Service Design at scale with the end user in focus, and it works.
Service Design for the majority journey is easy. The real test is how to handle edge cases. Whether system quirks, unusual residents, those with accessibility needs, or the elderly, the absence of flexibility risks turning a model of efficiency into a point of failure.