Most business cases for KYC automation fail at the same meeting. The vendor’s slide says onboarding gets faster and analysts get more productive, the CFO asks where the numbers came from, and the answer is another bank, another product mix, another year. This post does not give you numbers. It gives you a model, the measurements that fill it, and the questions a finance team will ask, so the case you take forward is built from your own operation.
In short
- Measure before you model. A two-week sample of where analyst time goes is worth more than any industry benchmark, including ours.
- Value comes from four places. Analyst hours on preparation, onboarding time to revenue, the review backlog, and the cost of rework and findings.
- Cost is more than the licence. Integration, validation, change management and the running cost of models belong in the case from the start.
- The sensitive variable is review effort. How long an analyst spends checking a prepared case, rather than building one, decides most of the outcome. Test the case against a pessimistic value.
Start with a time sample
Before any model, find out where time goes today. Pick a representative fortnight. For a sample of cases of each type (individual onboarding, business onboarding, periodic review, event-driven review), ask analysts to log minutes against a small set of activities. Keep the categories coarse; precision matters less than honesty.
| Activity | Preparation or decision? | Measure |
|---|---|---|
| Gathering: documents, registry extracts, list results, prior files | Preparation | Minutes per case |
| Reading and comparing: names, dates, addresses, expiry, MRZ | Preparation | Minutes per case |
| Ownership: walking the structure, calculating percentages | Preparation | Minutes per business case |
| Screening review: clearing possible matches | Mostly preparation; decision for real candidates | Minutes per case, and matches per case |
| Chasing the customer: drafting requests, re-checking what arrives | Preparation | Minutes per case, and days waiting |
| Writing the memo | Preparation | Minutes per case |
| Deciding, and second-line approval where needed | Decision | Minutes per case |
| Rework after QA | Both | Share of cases returned, and minutes to fix |
Alongside the time sample, pull four numbers from your systems: cases opened per year by type, median days from application to decision, the number of periodic reviews currently overdue, and the share of cases returned by quality assurance. These are the baseline the case is measured against later.
The model
Write the case as formulas with named inputs, so finance can change any input and see the effect.
For each case type t:
# inputs you measure N_t cases per year of type t P_t preparation minutes per case today D_t decision minutes per case today R_t minutes to review a prepared case # estimate, then measure in a pilot C fully loaded cost per analyst minute # analyst capacity released minutes_today = Σ_t N_t × (P_t + D_t) minutes_after = Σ_t N_t × (R_t + D_t) capacity_value = (minutes_today − minutes_after) × C # onboarding time to revenue (business lines that can attribute revenue to an account) V revenue per new relationship per day once open Δdays reduction in median days to decision # measured in the pilot acceleration = N_onboarding × Δdays × V × share_not_lost_anyway # the case annual_value = capacity_value + acceleration + backlog_value + rework_value annual_cost = licence + run_cost + support + share_of(implementation + validation)
A model, not a forecast: every input is yours. Leave any line at zero if you cannot measure it.
Where the value comes from
1. Analyst capacity
The largest line in most cases. Preparation minutes (P_t) are replaced by review
minutes (R_t); decision minutes (D_t) stay the same, because a person
still decides. Released capacity can absorb growth without hiring, clear a backlog, or reduce
reliance on contractors. Be explicit about which: a CFO will discount “productivity”
that does not change a budget line.
2. Onboarding time
For commercial and private banking, days to open an account delay revenue, and long onboarding loses customers to competitors. Only count this where the business can attribute revenue to an opened relationship, and apply a factor for customers who would have stayed anyway. Much of the delay is waiting on the customer, so measure how much of it is the bank’s.
3. The review backlog
Overdue periodic reviews are a regulatory exposure. The value of clearing them is usually avoided cost: the contractor team or remediation programme the bank would otherwise fund. Put that cost in, if there is one planned; otherwise leave this line at zero.
4. Rework and findings
Cases returned by quality assurance cost the minutes to fix them. Supervisory findings cost more, but are hard to put a number on honestly; most finance teams prefer them described as a risk reduction rather than counted as value.
Estimate review effort, then replace the estimate with a measurement from a pilot before asking for the full budget.
Costs to include
| Cost | What it covers | Often forgotten? |
|---|---|---|
| Licence or subscription | The platform itself | No |
| Run cost | Hosting in your cloud or the vendor’s; model inference if a language model is used | Inference, often |
| Integration | Connecting the application system, document store, company registry, screening lists, core banking and the case history | The number of systems, often |
| Validation | Model risk review, testing against planted scenarios, independent validation where your policy requires it | Yes |
| Policy translation | Turning the written risk methodology and procedures into rules and settings, and signing them off | Yes |
| Change management | Training analysts to review prepared cases rather than build them; QA during the transition | Yes |
| Parallel running | Working cases both ways during the pilot | Yes |
The variable that decides it
Almost every KYC automation case is most sensitive to one input: R_t, the minutes an
analyst spends reviewing a prepared case. If the preparation is trustworthy and shows its sources,
review is quick. If analysts have to redo the work to trust it, R_t approaches
P_t and the case disappears.
A score, a rating and a recommendation with no sources. The analyst re-reads the documents and re-runs the checks to be safe. Review takes nearly as long as preparation did.
Every value quoted from its line, every check with what it compared, every match with its components. The analyst reads the reasoning and decides.
Run the model at three values of R_t: your estimate, double it, and the value at
which the case breaks even. If it only works at the optimistic value, say so; the pilot is then
the decision point, not the business case.
Questions finance will ask
- Which budget line changes, and when? Hiring avoided, contractors released, a remediation programme not funded.
- What happens if volume falls? A per-case cost structure looks different from a fixed one when volume drops.
- What is the exit cost? Where the data lives, in what format, and whether the decision records stay readable if the platform is replaced.
- What did the pilot measure? Real
R_t, real days to decision, real agreement rates, not the vendor’s.
How we would run the pilot
One product line, one case type, a few weeks, with cases worked both ways on a sample. InsightKYC records how often analysts accept, reject or question each prepared decision, and every step’s timing, so the pilot produces the numbers the model needs. The demo bank in our screenshots is fictional; your pilot’s numbers will be your own.
What it means for each audience
For the CCO & MLRO
Lead with control as well as cost: consistent ratings, complete files and reviews on time are part of the case, even where they are not counted as money.
Insist that the pilot measures agreement and quality, not just speed.
For the KYC operations lead
Your time sample is the foundation. Make it honest, including the minutes spent chasing customers and fixing QA returns.
Be clear about what released capacity will be used for.
For the board
Ask for the model, not the answer: named inputs you can challenge, a pessimistic case, and a pilot that replaces estimates with measurements before the full spend.
A case that only works with a vendor’s benchmark is not a case.
The best business case for KYC automation contains no number you did not measure yourself.
Sample the work, model it with your inputs, and let a pilot settle the one variable that matters: how long a person needs to review a prepared case.