When AI runs the back office, where does the margin move?
AI changes the cost base of banking and insurance directly. Transaction processing automates toward the electronic end of the cost curve, AI absorbs tiers of risk and compliance work, and agents handle servicing that used to fill a call centre. As manual cost falls, the spread between customers narrows in some places and widens in others, and only a per-customer, per-channel cost model can tell you where.
Cost and Profitability Consulting · 150+ models since 2010 · TDABC
AI automates transaction processing toward the cheap, electronic end of the cost curve, absorbs tiers of risk and compliance work, and handles servicing that used to fill a call centre. As manual cost falls, the spread between customers narrows in some places and widens in others, and only a per-customer, per-channel cost model can show where the margin actually moves. AI changes the shape of the cost curve but not where margin lands, so it makes cost-to-serve modelling more important, not less. The institutions that win are the ones that already price on cost to serve, not on a blended average.
Four shifts, one dependency.
Transaction processing
Automation pushes manual transactions toward the electronic end of the curve, compressing the most expensive minutes.
Risk & compliance
AI absorbs tiers of review and monitoring work, changing the load each relationship carries.
Servicing & agents
Conversational agents handle contact that used to fill a call centre, moving cost off the expensive channel.
The cost curve reshapes
As manual cost falls, the 5 to 10 times spread changes shape. Only a per-customer model shows the new picture.
ONE TRANSACTION, FOUR CHANNELS, FOUR COSTS
Illustrative. AI compresses the expensive end of this curve, branch and call, toward the cheap end. The relative gaps change, and so does where each customer's margin sits.
Automation changes the curve. It does not tell you where margin lands.
An institution can automate enthusiastically and still not know whether it improved margin, because automation reshapes cost without revealing profit. A per-customer, per-channel model on a TDABC base is what turns AI from a cost-cutting story into a margin decision: it shows which relationships got cheaper, which barely moved, and where to deploy automation next. This is a question of decision quality, not a regulatory countdown, major obligations are phased toward 2027, so the case is defensibility and clarity, not a deadline. Budget the human side honestly: the teams reading the model need to understand cost to serve well enough to act on it. The institutions that win already price on cost to serve, not on a blended average.
Frequently asked questions
- How is AI changing cost in banking and insurance?
- AI automates transaction processing toward the cheap, electronic end of the cost curve, absorbs tiers of risk and compliance work, and handles servicing that used to fill a call centre. As manual cost falls, the spread between customers narrows in some places and widens in others, and only a per-customer, per-channel cost model can show where the margin actually moves.
- Does AI remove the need for cost-to-serve modelling?
- No, it makes it more important. AI changes the shape of the cost curve but does not tell you where margin lands; that still requires a per-customer, per-channel model. The institutions that win are the ones that already price on cost to serve rather than on a blended average, so they can see how automation shifts each relationship.
- Is the EU AI Act a deadline for banks on this?
- This is a defensibility question, not a countdown. With major obligations phased toward 2027, the case for a per-customer cost model is decision quality, not a regulatory deadline. Knowing the true cost to serve is what lets a bank deploy AI where it improves margin and prove the basis of its pricing, regardless of timing.
Get the cost model AI needs underneath it.
The Profit Check takes five minutes and no data upload. It shows whether your cost data can tell you where automation moves margin, and what to fix first.