Decision intelligence before automation
16 Sep 2026
As digital labour moves onto the Sibos agenda, CIBC Mellon’s chief operations officer Catherine Thrasher argues that the most important foundation for the agentic future remains data
Image: CIBC Mellon
Automation has been an unavoidable component of operating model modernisation in securities services for decades. Institutions continue to face growing pressure to reduce manual effort, manage rising complexity, and meet increasingly stringent regulatory and client expectations.
We all continue to invest heavily in technology and tools to future-proof our operations. Yet, for too many firms, outcomes remain uneven: many initiatives deliver localised efficiency gains, while far fewer translate into sustained improvements across the broader operating model.
That gap rarely comes down to the technology. More often it reflects how automation decisions are prioritised, sequenced, and governed. Too frequently, organisations automate based on the tools they already have or the capabilities vendors promote, rather than on what operational data reveals about how work is actually performed.
The numbers bear this out. 70 per cent of transformation projects fall short of their objectives, and EY reports that two-thirds of capital markets leaders have experienced at least one underperforming transformation in the past five years.
At Sibos this year I will be joining a panel on the rise of digital labour and agent economics. The premise is that we are entering a period in which digital labour is recruited, governed, and scaled alongside human talent, and firms are redefining their operating models accordingly. I want to make a narrower, more practical point: none of that works without a factual baseline of how work is performed today. Agentic workflows will accelerate the collaboration between people and tools, but trust, regulatory compliance, and judgement remain critical to outcomes. The discipline that makes conventional automation succeed is the same discipline that will make agents safe to deploy.
Decision intelligence as a control layer
We describe that discipline as decision intelligence: the systematic use of operational data, workforce analytics, and process-level metrics to understand how work is executed, where capacity is consumed, and how effort is distributed across teams and activities. In place of lagging indicators and anecdotal feedback, it provides timely insight into task volumes, time allocation, and process variability, creating a baseline against which interventions can be identified, prioritised, and later validated. This is the data that should rest at the foundation of the future: agentic and more.
The approach also shaped our guiding questions. How can automation amplify, rather than replace, professional judgement and expertise? What characteristics define a process that is suitable for automation, recognising that poorly-designed processes rarely improve when automated? Where do friction points exist between human and digital touchpoints? How can teams be supported through reskilling in ways that encourage participation from the outset?
What the data showed in fund accounting
Fund accounting has historically relied on manual processes designed to confirm accuracy and control. Spreadsheet-based reconciliations, email-driven exception handling, and client-specific workflows were common.
While effective in maintaining accuracy, these approaches were labour-intensive and difficult to scale.
Our operational leaders had long observed that reconciliation represented a significant investment of effort, but lacked the precise data to quantify its scale or understand how that effort was distributed. Workforce data provided that visibility, revealing that reconciliation activities consumed approximately 40 per cent of fund accountants’ time, representing over 10,000 hours of effort each month. Three reconciliation types — cash, asset, and capstock — accounted for roughly 75 per cent of that activity and approximately 30 per cent of total daily work effort.
The data also pointed to a broader issue: the reconciliation process itself was highly fragmented. Files were collected at the client level, loaded into spreadsheets, and processed using macros, with exceptions recorded manually and consolidated into master spreadsheets for review.
Crucially, these insights did not merely confirm that automation was needed. They identified precisely where intervention would have the greatest impact.
From insight to intervention
Rather than pursuing broad automation across multiple activities simultaneously, we prioritised the areas where the concentration of effort and operational friction were most pronounced. Based on the analysis, we established a centralised NAV Construction team aligned with our global operating model.
13 activities were identified for migration into this function, with reconciliation selected as the initial focus because of its disproportionate consumption of time and its suitability for standardisation.
Two principles guided that decision. First, automation efforts should be informed by measured operational need rather than perceived inefficiency. Second, automation is most effective when paired with organisational design changes that support scale, control, and consistency. Centralisation allowed reconciliation to be managed as an enterprise capability rather than a series of client-specific tasks.
The choice of reconciliation platform followed an operational evaluation of tools capable of supporting the standardised workflows that centralisation required. As the centralised team expanded, asset and capstock reconciliations were migrated into consolidated workflows. This represented a shift from client-by-client reconciliation toward a fund-level approach: where multiple clients were invested in the same fund, the reconciliation could be performed once rather than repeated for each client.
Measuring what changed
A defining feature of this approach was the continued use of decision intelligence after implementation. The same platform that identified where to intervene was used to validate whether the intervention had achieved its intended effect, closing the feedback loop between insight and outcome.
Early results from proof-of-concept implementations indicated a reduction of approximately 1,400 hours of manual effort and a 15 per cent decrease in reconciliation-related workload for the activities in scope. Automated workflows reduced the need for manual file preparation and exception tracking, allowing teams to redirect effort toward analysis, review, and client-facing activities.
These changes did not eliminate oversight. They shifted human involvement toward judgement-based tasks and strengthened control and governance.
Ongoing visibility into how work evolves also reduces the risk of automation drift, where processes gradually diverge from their intended design without detection.
Measurement is not simply a precursor to automation. It is an ongoing operating model capability.
People are the operating model
The sustainability of this change depends heavily on workforce engagement and capability development. As reconciliation was centralised and automated, roles within fund accounting evolved from transaction-oriented execution toward analysis, exception management, and insight generation. That shift required new skills in data interpretation, problem-solving, and collaboration across teams. Programmes such as Skill Sprints and technology showcases gave employees structured opportunities to build digital and analytical capability while contributing directly to operational improvement.
Internal survey results within our operations function indicate that 91 per cent of employees reported feeling engaged and 90 per cent reported feeling empowered to contribute ideas for improvement, six and seven points above the financial services industry norm respectively, according to employee engagement surveys.
For clients, the gains were second-order effects of a more disciplined operating model: teams better positioned to meet deadlines, manage exceptions proactively, and stand behind the integrity of reported data.
Sustainable automation is not achieved through technology alone. It requires evidence-based decision-making, thoughtful organisational design, and continued investment in people. As digital labour enters our operating models, the institutions that treat automation as an operating model discipline rather than a technology initiative will be the ones positioned to scale it responsibly.
The agentic future offers significant opportunities to transform how operations are designed and delivered. But realising that potential starts with understanding the work that exists today. Foundational workforce and process data, combined with decision intelligence, provide the visibility required to distinguish where technology can eliminate effort, where processes need to be redesigned, and where human expertise remains essential.
The question is no longer simply what can we automate — but what should we redesign — and what does the data tell us about why?
We all continue to invest heavily in technology and tools to future-proof our operations. Yet, for too many firms, outcomes remain uneven: many initiatives deliver localised efficiency gains, while far fewer translate into sustained improvements across the broader operating model.
That gap rarely comes down to the technology. More often it reflects how automation decisions are prioritised, sequenced, and governed. Too frequently, organisations automate based on the tools they already have or the capabilities vendors promote, rather than on what operational data reveals about how work is actually performed.
The numbers bear this out. 70 per cent of transformation projects fall short of their objectives, and EY reports that two-thirds of capital markets leaders have experienced at least one underperforming transformation in the past five years.
At Sibos this year I will be joining a panel on the rise of digital labour and agent economics. The premise is that we are entering a period in which digital labour is recruited, governed, and scaled alongside human talent, and firms are redefining their operating models accordingly. I want to make a narrower, more practical point: none of that works without a factual baseline of how work is performed today. Agentic workflows will accelerate the collaboration between people and tools, but trust, regulatory compliance, and judgement remain critical to outcomes. The discipline that makes conventional automation succeed is the same discipline that will make agents safe to deploy.
Decision intelligence as a control layer
We describe that discipline as decision intelligence: the systematic use of operational data, workforce analytics, and process-level metrics to understand how work is executed, where capacity is consumed, and how effort is distributed across teams and activities. In place of lagging indicators and anecdotal feedback, it provides timely insight into task volumes, time allocation, and process variability, creating a baseline against which interventions can be identified, prioritised, and later validated. This is the data that should rest at the foundation of the future: agentic and more.
The approach also shaped our guiding questions. How can automation amplify, rather than replace, professional judgement and expertise? What characteristics define a process that is suitable for automation, recognising that poorly-designed processes rarely improve when automated? Where do friction points exist between human and digital touchpoints? How can teams be supported through reskilling in ways that encourage participation from the outset?
What the data showed in fund accounting
Fund accounting has historically relied on manual processes designed to confirm accuracy and control. Spreadsheet-based reconciliations, email-driven exception handling, and client-specific workflows were common.
While effective in maintaining accuracy, these approaches were labour-intensive and difficult to scale.
Our operational leaders had long observed that reconciliation represented a significant investment of effort, but lacked the precise data to quantify its scale or understand how that effort was distributed. Workforce data provided that visibility, revealing that reconciliation activities consumed approximately 40 per cent of fund accountants’ time, representing over 10,000 hours of effort each month. Three reconciliation types — cash, asset, and capstock — accounted for roughly 75 per cent of that activity and approximately 30 per cent of total daily work effort.
The data also pointed to a broader issue: the reconciliation process itself was highly fragmented. Files were collected at the client level, loaded into spreadsheets, and processed using macros, with exceptions recorded manually and consolidated into master spreadsheets for review.
Crucially, these insights did not merely confirm that automation was needed. They identified precisely where intervention would have the greatest impact.
From insight to intervention
Rather than pursuing broad automation across multiple activities simultaneously, we prioritised the areas where the concentration of effort and operational friction were most pronounced. Based on the analysis, we established a centralised NAV Construction team aligned with our global operating model.
13 activities were identified for migration into this function, with reconciliation selected as the initial focus because of its disproportionate consumption of time and its suitability for standardisation.
Two principles guided that decision. First, automation efforts should be informed by measured operational need rather than perceived inefficiency. Second, automation is most effective when paired with organisational design changes that support scale, control, and consistency. Centralisation allowed reconciliation to be managed as an enterprise capability rather than a series of client-specific tasks.
The choice of reconciliation platform followed an operational evaluation of tools capable of supporting the standardised workflows that centralisation required. As the centralised team expanded, asset and capstock reconciliations were migrated into consolidated workflows. This represented a shift from client-by-client reconciliation toward a fund-level approach: where multiple clients were invested in the same fund, the reconciliation could be performed once rather than repeated for each client.
Measuring what changed
A defining feature of this approach was the continued use of decision intelligence after implementation. The same platform that identified where to intervene was used to validate whether the intervention had achieved its intended effect, closing the feedback loop between insight and outcome.
Early results from proof-of-concept implementations indicated a reduction of approximately 1,400 hours of manual effort and a 15 per cent decrease in reconciliation-related workload for the activities in scope. Automated workflows reduced the need for manual file preparation and exception tracking, allowing teams to redirect effort toward analysis, review, and client-facing activities.
These changes did not eliminate oversight. They shifted human involvement toward judgement-based tasks and strengthened control and governance.
Ongoing visibility into how work evolves also reduces the risk of automation drift, where processes gradually diverge from their intended design without detection.
Measurement is not simply a precursor to automation. It is an ongoing operating model capability.
People are the operating model
The sustainability of this change depends heavily on workforce engagement and capability development. As reconciliation was centralised and automated, roles within fund accounting evolved from transaction-oriented execution toward analysis, exception management, and insight generation. That shift required new skills in data interpretation, problem-solving, and collaboration across teams. Programmes such as Skill Sprints and technology showcases gave employees structured opportunities to build digital and analytical capability while contributing directly to operational improvement.
Internal survey results within our operations function indicate that 91 per cent of employees reported feeling engaged and 90 per cent reported feeling empowered to contribute ideas for improvement, six and seven points above the financial services industry norm respectively, according to employee engagement surveys.
For clients, the gains were second-order effects of a more disciplined operating model: teams better positioned to meet deadlines, manage exceptions proactively, and stand behind the integrity of reported data.
Sustainable automation is not achieved through technology alone. It requires evidence-based decision-making, thoughtful organisational design, and continued investment in people. As digital labour enters our operating models, the institutions that treat automation as an operating model discipline rather than a technology initiative will be the ones positioned to scale it responsibly.
The agentic future offers significant opportunities to transform how operations are designed and delivered. But realising that potential starts with understanding the work that exists today. Foundational workforce and process data, combined with decision intelligence, provide the visibility required to distinguish where technology can eliminate effort, where processes need to be redesigned, and where human expertise remains essential.
The question is no longer simply what can we automate — but what should we redesign — and what does the data tell us about why?
NO FEE, NO RISK
100% ON RETURNS If you invest in only one asset servicing news source this year, make sure it is your free subscription to Asset Servicing Times
100% ON RETURNS If you invest in only one asset servicing news source this year, make sure it is your free subscription to Asset Servicing Times
