Orchestration is the only way to join up your people, processes and AI
From what I hear from organisations I speak to, and from my own experience, there's a wide gap between AI adoption and AI value. Many initiatives stall and deliver little to no measurable impact on the P&L, largely because AI is bolted onto workflows built in a different era and lacking alignment with how work actually happens day to day.
MIT research published in late 2025 found that AI is currently economically viable to replace the equivalent of just 11.7% of US jobs, the tasks where AI can perform the same work at a cost competitive with or cheaper than a human. For everything else, humans remain the better option.
We've been here before. RPA promised to transform operations too, and for many firms it became shelf-ware, for exactly the same reason: deploying a technology without redesigning the work around it.
The other pattern I keep seeing is how this gets funded. Too many organisations are paying for their AI programmes out of savings from the last automation wave, which means they're running a self-funding efficiency cycle, not making a genuine growth investment case. That frames the whole conversation around cost, and cost is the wrong question. The one I'd rather we asked is this: What do our best people do that nobody else can, and how do we give them more time to do it, not fewer of them to do it?
Nobody is redesigning for how clients actually work
The workflows inside most service firms reflect internal org structures, not the business flows of the clients they serve. That gap is widening as AI gets pushed into broken processes rather than rebuilt ones.
When I challenged my own teams to stop bolting automation onto existing roles and instead redesign the outcome itself, the work split naturally into three distinct human functions:
- Exception handling, where the automation couldn't complete a step it was meant to
- Capability improvement, refining what automation could take on next
- Client interaction, the work nobody seriously believes AI is best placed to own
None of that maps to a department on an org chart. It maps to where people add most value inside the client's actual workflow.
That's exactly the shift I see most Trust and Corporate Services providers, Fund Administration providers, accountancy firms and the wider BPO sector struggling to make. They're combining people, often sitting in shared services, with automation built around how work is currently structured, not how it needs to be reimagined for a new era.
Resistance is rarely about the redesign logic itself. People grasp it quickly once they see their own work mapped this way. The barrier is the fear of not having a job, reinforced by the cost-out framing every automation conversation defaults to. The version that actually lands is a growth one: the same team serving more clients, doing more of the work only they can do, not fewer of them doing the same work as before. That's borne out in the data.
Research from the Ramp Economics Lab, across more than 21,000 US firms, found that heavy AI adopters grew headcount by 10% in the two years following adoption, with entry-level hiring up 12%. Low adopters saw no statistically significant change.
AI can’t solve this alone, and neither can people
AI isn't one thing and treating it as such is part of why so many conversations stall. Rules-based automation, predictive models, generative AI and agentic AI are four distinct technologies with different capabilities, different risk profiles and different fitness for any given task.
A good deal of what gets pitched as an AI problem is better solved with simpler automation and a cleaner process. And even where generative or agentic AI genuinely fits, outputs still need checking and capability still needs investment. There's no silver bullet, whatever the deck says.
The real gap isn't a missing automation technology. It's the absent layer that decides which tool handles which step, when a person needs to step in, and who owns the handoff between them. Most vendors claiming to deliver this end-to-end are doing what analysts now call agent washing, rebadging existing automation and chatbots with agentic language because that's the story the market currently rewards.
Try asking them "what happens when the tool hits an exception it can't resolve", "who is accountable when it gets something wrong inside a live client process", and "whether the demo reflects a narrow well-scoped task or the actual mess of a real workflow". Most have a confident answer to the first question and go quiet on the second and third.
Until this coordination work is named and built as an orchestration product in its own right, every AI conversation in this sector will keep going the same way. Service providers will keep bolting new tools onto broken processes, lean on people to invisibly glue it all together, and then wonder why the ROI never shows up.
What good actually looks like
The firms that are getting AI right treat the client's case as the unit of design, not the individual task. They sequence it properly too, starting by asking what the work should look like in a human and AI era, not what it looks like today. AI is part of the design from the outset, not something bolted on once the process has been tidied up. Once that foundation is in place, people stop executing tasks. They start handling exceptions, improving what the automation can take on next, and owning the client relationship, the same three-function model I described earlier, arrived at independently across different organisations, which tells me it's the shape this work takes wherever it is done properly.
Orchestration isn’t optional
This all circles back to the same fork in the road: Is AI here to take cost out, or to help your best people do more for more clients? That tension doesn't resolve itself, and orchestration is the only thing that actually resolves it, because without it you can't tell which one is happening. You won't know whether your AI spend is replacing cost or quietly becoming a new cost line nobody is tracking.
The firms treating orchestration as optional today are the same firms who, in six months, will have no idea what their AI is costing them per client, per case, or per exception handled versus automated. RPA left behind a graveyard of licences nobody could account for.
AI's about to do the same thing, faster and on a much bigger bill, unless someone owns that control point properly. That’s before the next problem arrives, because the bills for API calls, token usage and cloud infrastructure are already racking up, and most organisations have no idea how to manage them. AI cost optimisation will be the subject of my next article.
