HFS Research recently published an article titled “The AI balloon is bursting and services must be ready to pick up the pieces” and I’ve dubbed it “Wall Street versus Main Street”. In it, they identified four categories of debt that separate distinguish enterprises forging ahead in implementing AI at scale from the ones stuck in POC Purgatory: technical debt, financial debt, process debt, and data debt.

Process debt and data debt live in GBS. They live in your order to cash processes, your accounts payable workflows, your customer master data, and your reconciliation procedures.

I asked Phil Fersht , CEO of HFS Research , how is research article applies to GBS leaders specifically. I asked “What should a GBS leader do differently in the next 90 days based on HFS’s research” He said this:

“The next 90 days shouldn’t be about launching more AI pilots. They should be about identifying one or two business domains where AI can fundamentally change how work gets done, capturing that expertise into repeatable operating capability, and proving measurable business outcomes. The GBS organizations that move first from delivering services to building Services-as-Software will define the next generation of enterprise value.”

What the Four Debts Look Like in Your GBS Operations

Technical Debt is when your systems do not talk to each other:

• Order to cash: AR system does not integrate with GL, so someone manually reconciles daily
• Payables: Multiple AP platforms across regions, invoices flowing through three different systems
• Reconciliation: Data pulled from six systems into a spreadsheet every month

You cannot capture this into repeatable operating capability because the capability does not exist reliably. Every month you are inventing workarounds.

Financial Debt is when you maintain manual work because automation feels too risky or expensive relative to the current cost:

• AR exception handling requires dedicated headcount but everyone treats it as a given cost of doing business
• Invoice keying continues because the business decided at some point that capture technology was too risky to implement
• Reconciliation staff stays the same size even though the volume that justified that staffing level disappeared years ago

You are paying compound interest on a decision someone else made. You cannot prove measurable business outcomes of an AI solution until you stop paying that interest.

Process Debt is when your procedures vary without business reason:

• Credit approval workflows are different by region because of acquisitions, not because the business demands it
• Invoice matching rules evolved over time and nobody remembers why some of them exist
• Month-end close procedures exist because they always have

This debt makes repeatable capability impossible and according to HFS Research, is the most expensive. How do you apply AI to a process that is not actually a process, but a learned behavior that is done on auto-drive every month?

Data Debt is when you cannot trust the underlying information:

• Customer masters have quality issues causing holds and manual overrides
• Vendor masters are inconsistent, causing payment delays and reconciliation work
• GL structures do not map cleanly across the enterprise, requiring translation before consolidation

This is the hardest debt to pay down because fixing it requires discipline and disruption. But this is also where most AI deployments fail. You cannot prove outcomes with bad data and you can’t scale a capability built on information you do not trust.

You cannot capture expertise into repeatable operating capability if these debts exist without being addressed. You cannot prove measurable business outcomes if you are solving the wrong problem because the data is bad or the process is undocumented or doesn’t make sense.

What GBS Leaders Should Do To Address The Debt

Start by understanding what you do and why you do it that way. Before you touch AI, before you select a domain, and certainly before you talk to vendors or build a business case, you need to know your own operations. Ask yourself:

• What does your order to cash process actually look like? Not what the documentation says. What does it actually do?
• How many exceptions flow through each step?
• How much of your work is following a documented procedure versus workaround and judgment call?
• Where is your data actually coming from and what are you doing to validate it?

Many GBS functions operate on muscle memory and inherited practice. That works when conditions are stable but when you try to change something or apply AI, the gaps become visible quickly.

Start with a single process

• Map it completely
• Document the rules
• Identify the exceptions
• Find the data quality issues
• Understand where the time actually goes and what consumes the most intervention

What typically floats to the surface:

• A gap between the documented procedure and the real procedure
• Data quality worse than expected
• Work exists only because the previous step failed to deliver what it promised
• Systems do not integrate the way you thought they did

Address the debt in priority order and don’t try to fix everything at once. Start with technical debt that is causing the most manual work. A single integration or data consolidation effort often removes half the exceptions in a process. Then fix the data quality issues that prevent you from even measuring the current state accurately. Once you have reliable data flowing through a connected system, the process debt becomes visible and fixable. Financial debt can takes care of itself once the first three are resolved because you no longer need the unnecessary headcount to manage workarounds.

Once you understand what you actually do and you have paid down the debt, then you can decide where AI makes sense. It will likely not be in low volume and/or judgment heavy work. It will likely be in the high-volume, repeatable processes where quality is currently impacted by manual effort and inconsistency. That is where you find the one or two domains worth building for scale.

I want to thank Phil Fersht and Saurabh Gupta for writing the article used on the basis for this newsletter and Phil for the quote. Here is a link to the original article: https://www.horsesforsources.com/wall-street-betting-on-an-ai-future-main-street-cannot-deliver_062526/