How to Map a Finance Workflow for Automation
Mapping a finance workflow means writing its steps down in order, recording the hands-on time and the readiness score against each one, and naming the person or desk accountable for it. The point is not documentation. A map is the only thing that makes the steps carrying judgement visible inside a process whose two ends look like obvious candidates, and it turns automate this process into a decision about eleven separate things.
Almost every process inside a bank looks automatable from end to end when somebody describes it out loud, and looks like something quite different the moment it is written down as steps. The whole value of a map sits in how finely it is cut: too coarse and the judgement hides inside a box that scores well, too fine and nobody ever finishes drawing it. Below is the procedure Sumeru Bank Limited, an invented mid-sized Indian bank, used on its retail personal loan process before a single component of the chain was built, and the map that came out of it.
Why draw a map at all, rather than automating the obvious steps?
A household kitchen at seven in the evening makes the point. Somebody says the cooking takes an hour. The sentence is true and useless at the same time. It hides the fact that fifty of those minutes are the pressure cooker doing its own work while nobody stands over it, and ten are chopping. Anybody buying something that helped would want the ten minutes, not the hour. Described as cooking, the evening attracts the wrong purchase. Written down as steps, the ten minutes appear.
A workflow mapThe steps of a process in order, each carried with its time, its accountable name, its rule and its score. does exactly that for a process in a bank, and it is worth the days it takes for one reason. A process described in a sentence has one score; the same process written down as steps has eleven, and the low ones are never where anybody expects. The two ends of a lending process are receiving paper and sending paper, and both look like obvious candidates to anybody who has watched them. The middle is where somebody reads a statement and forms a view. Described as a process, that middle is one clause in a sentence. Drawn as steps, it is two boxes with the two lowest scores on the map.
There is a second reason, and it matters more later than it does on the day. The map is the artefact somebody argues with. A decision to automate a step that a scoring convention said not to automate is defensible or it is not, but it can only be either of those things if there is a row somewhere with a number on it and a note beside the number. Without the map there is no row, so there is nothing to disagree with and nothing to look back at.
What are the seven steps of drawing the map?
Here is the whole procedure. The order is not a suggestion. Read the whole procedure once end to end before starting on any of it. Every step consumes what the previous step produced.
- Agree where the process starts and where it stopsWrite the two edges down as a sentence and get the desks on both sides to agree with it before anything else happens. Everything outside the two edges is listed once, as excluded, and then never argued about again.Produces: the two edges, in writing, and a short list of what sits outside them.
- List the steps in order, at a size somebody can finishOne pass down the process, one line per unit of work, no times and no scores yet. Aim for the count to land somewhere between about eight and fifteen. Fewer and the map hides things; more and the drawing never gets finished.Produces: a numbered list of steps, in the order a file meets them.
- Time each step, hands-on onlySit beside the desk and time what a person actually does. Waiting, queueing and the file sitting in a tray overnight are not step time. The total is the number that can be checked against something, so record it as well as the parts.Produces: minutes a file for every step, and a total for every desk.
- Record four more things against every stepWho is accountable for it, what it takes in, what it puts out, and what rule it follows. Four short answers, written in the same words for every row so the rows can be compared.Produces: a complete row for every step, six columns wide.
- Score every step on the readiness test, before deciding anythingFive tests, nought to two each, ten at most: how often it happens, how stable the rule is, whether the input arrives in a readable form, whether the correct outcome can be stated in advance, and what happens when it is wrong. Score every row before looking at any of the numbers together.Produces: a score out of ten against every row.
- Mark each step automate, assist, or leaveTake the score, then ask the second question: does somebody have to explain this outcome to the person it was about? A step can clear the score and still not be a candidate.Produces: one of three words on every row, and no blanks.
- Write down every decision taken against the scoreWhere a step is marked automate with a score below the line, the row carries the reason and the name of whoever decided it. Not a footnote and not a conversation somebody remembers.Produces: a short written record beside every row where the score was not followed.
Where does the process start, and where does it stop?
The first step feels like paperwork before the real work, so it is the one people skip. It is not. A boundaryWhere the mapped process is agreed to start and to stop, written down before any step is listed. that nobody agreed on is what produces two maps of the same process that do not add up, and it produces them weeks later when everybody has stopped paying attention.
At Sumeru Bank Limited the two edges were set as follows. The map starts when a completed application arrives. The map stops when the disbursal instruction has been raised. Everything outside those two edges was listed once as excluded and then left alone. Listing the exclusions once is the only way a boundary argument gets settled rather than repeated. Excluded, in writing: the time an applicant spends deciding which lender to approach, the servicing of the loan after money moves, and the collection of anything that goes wrong later. None of those three is unimportant. All three are somebody else's map.
The everyday version is a household deciding to redo the kitchen. If nobody says whether the job includes the electrical work, two people quote for the same room and neither quote covers the wiring. Nobody lied. The edges were never drawn.
The map was agreed to start when a completed application arrives and to stop when the disbursal instruction is raised. Where does that put the fortnight a customer spent deciding which lender to approach?
What counts as one step rather than two?
A stepOne unit of work with a single accountable name, a single thing coming in, and a single thing going out. is one unit of work with one accountable name, one input and one output. One name, one input and one output is the whole test, and the test is stricter than it sounds. Two names means two steps, even where the two people sit at the same table and the work feels continuous. One name doing two things that produce two separate outputs is also two steps.
The test exists to control granularityHow finely a process is cut into steps, which decides what the map is able to show at all., and granularity decides whether the map is worth drawing. Cut too coarsely, a box containing both keying a field and writing an assessment scores the average of the two. The average of something easy and something hard looks moderately easy. Cut too finely, at forty or sixty steps, the map is accurate and unfinished. Nobody ever times sixty things.
Sumeru Bank Limited drew its process twice. The first drawing had four steps: receive, assess, decide, communicate. The second had eleven. Nothing about the work changed between the two drawings, and the second one is not more correct in any factual sense. The second drawing is simply cut finely enough that the two hard steps stopped sharing a box with easy ones.
What makes something one step rather than two?
How is a step timed, and what does that time exclude?
Hands-on timeMinutes a person actually spends working on a step, with waiting and queueing left out entirely. is minutes a person actually spends on the step. Hands-on time excludes waiting, queueing, the file sitting in a tray while somebody is at lunch, and the two days it takes a third party to come back with something. Elapsed timeHow long a file takes to get from one end of the process to the other, including every period of waiting. is the other number: how long the file takes to get through, waiting included.
Both numbers describe the same file and they answer completely different questions. A map that records only one of them cannot do the job it was drawn for. Minutes of work are what consume a person, so hands-on time sizes the desk. Elapsed time is what the customer experiences, and it consumes nobody. Mixing them produces one of two errors: a two day process that looks like eleven minutes of work, or eleven minutes of work costed as though it occupied somebody for two days.
Time it by sitting beside the desk with a watch, on a normal day, on files that arrive in the usual way. Take several files rather than one. The person will change and the step will not, so write the minutes down against the step rather than against the person.
A file takes two working days end to end and eleven minutes of hands-on time. Which of the two numbers sizes the desk?
What else goes in the row besides the time?
Four more things, and the map is only usable when all four are on every row. The ownerThe named person or desk accountable for a step being carried out, written as one name and never as two. is one name or one desk. Then what the step takes in, and what it puts out, in the plainest words available. Last, the rule the step follows: a written procedure somebody can be handed, or a note saying there is not one.
The rule field does more work than it looks like it does. A step with a written rule and a step where an experienced person decides by feel are different objects on a map, and the difference shows up nowhere else on the row. Both may run in three minutes. Only one of them can be handed to somebody new on a Monday morning.
There is a discipline about how the what it does field is written, and it is the discipline that keeps a map honest. Every row says what happens in the step, in a verb and an object, and never what anything works out, recognises or takes into account. Fields keyed into the system is a step. Judges whether the income looks reasonable is not a step; it is a description of somebody's inner life, and no map can hold it, time it or score it.
How does a score become a decision on the step?
Step five produces a score out of ten on every row. Step six turns that score into one of three words, and it does so with the score plus exactly one further question.
The score first. Seven or above means the step can be carried out without a person standing in it: it happens often enough to be worth the trouble, the rule behind it changes rarely, the input arrives readable, the right answer can be stated in advance, and the cost of getting it wrong is bearable. Below seven, at least one of those five is not true, and the map does not say which until somebody reads the five component scores.
Then the second question, and it is the one that catches the awkward cases. Does somebody have to explain this outcome to the person it was about? Where the answer is yes, the step is assisted rather than automated, whatever the score says: something is produced and a person signs it. A high score establishes that a step could run without a person. Whether it should is settled by who has to stand behind the result afterwards.
A step scores eight, and its outcome has to be explained to the customer it was about. Should it be marked automate?
What did the finished map say at one bank?
Here is the whole thing for the retail personal loan process at Sumeru Bank Limited, as it stood before any part of the chain was built. Eleven steps, in the order a file meets them.
| Step | What the step does | Owner | Hands-on | Score | Decision |
|---|---|---|---|---|---|
| 1 | Application received and logged | Intake and processing desk | 1.0 min | 9 | Automate |
| 2 | Identity documents checked against the application | Intake and processing desk | 2.0 min | 8 | Automate |
| 3 | Documents sorted and filed | Intake and processing desk | 1.5 min | 9 | Automate |
| 4 | Fields keyed into the system | Intake and processing desk | 3.0 min | 7 | Automate |
| 5 | Income corroborated against the statement | Credit officers | outside this total | 8 | Automate |
| 6 | Credit record pulled | Credit officers | outside this total | 10 | Automate |
| 7 | The assessment written | Credit officers | outside this total | 3 | Leave |
| 8 | The decision taken | Credit officers | outside this total | 4 | Automate, as an override |
| 9 | The decision recorded | Intake and processing desk | 1.0 min | 9 | Automate |
| 10 | The letter drafted and sent | Intake and processing desk | 1.5 min | 6 | Assist |
| 11 | The disbursal instruction raised | Intake and processing desk | 1.0 min | 8 | Automate |
| Seven steps sitting with the intake and processing desk | 11.0 min | 8 automate, 1 assist, 1 leave, 1 override |
Two things about that table are worth stopping on. The first is that the seven steps sitting with the intake and processing desk add to exactly 11.0 minutes: 1.0 plus 2.0 plus 1.5 plus 3.0 plus 1.0 plus 1.5 plus 1.0. The seven-row total is the handling time the desk was already reporting from its own records. A figure that can be checked against an independent record is what turns the map from an impression into something checkable. If the seven rows had added to seven minutes or to sixteen, the map would have been wrong somewhere and the arithmetic would have said so before anybody acted on it.
The second is the four rows marked outside this total. Steps 5 to 8 sit with credit officers elsewhere in the bank, so their time is recorded against a different desk and never inside the 11.0 minutes. Keeping the two desks apart is exactly what step one of the mapping procedure buys. Without agreed edges, somebody adds all eleven times together and sizes one desk using another desk's work.
Now read the score column downwards rather than across. Nine, eight, nine, seven at the top. Eight and ten in the upper middle. Then three and four, side by side. Then nine, six and eight to finish. The two lowest scores on an eleven step map sit next to each other in the middle, and that is what a process with all its judgement in one place looks like when it is drawn honestly. The shape is not a coincidence and it is not particular to lending. Most finance processes are handling at both ends and judgement in the middle. A map cut coarsely at the middle is comfortable to read for precisely that reason.
The two lowest scores on the map sit next to each other in the middle of the process. What does that pattern usually mean?
What has to be written down when the score is not followed?
Step 8, the decision taken, scored four. Under the rule set out above it should have been marked assist or leave. The case for building the chain at all did not work without it, so step 8 was marked automate. A decision taken against the score is an overrideA decision taken against what the readiness score advised, recorded as such with a reason and a name beside it., and the only thing the map can insist on is that it is written down as one.
Be clear about what is and is not being said here. An override is not a mistake. Somebody weighed a score of four against a case that needed the step automated and decided the second mattered more, and that is a decision people are paid to take. The map turns the override into a decision rather than a drift. A row that reads score four, decision automate, reason and name attached is something a reviewer can pick up two years later and argue with. A row that reads score four, decision automate, and nothing else looks identical from the outside and supports nothing.
The household version is a rented flat where the tenant decides to paint over a stain rather than fix the leak behind it. Perfectly reasonable when the move is three months away. The next person needs to know that the wall was painted rather than repaired, so the paint goes on the list.
Step 8 scored four and was marked automate anyway. What does the map have to hold for that to be defensible later?
Before the control below is moved. The same process drawn at four steps instead of eleven. What is the lowest readiness score on that coarser map?
Cut the same process into more boxes, and watch the advice change
One control: how finely the same eleven steps of work are cut, at four boxes, six, eight or eleven. One consequence: a box holding several steps scores the rounded average of what is inside it, so the readiness scores redraw. The eleven underlying scores never change, so the strip underneath never moves. The default below is the eleven step drawing, whose lowest score is 3 at step 7 and which marks eight steps automate, one assist, one leave and one override. Drawn at four steps, the same process has a lowest score of 7 and nothing at all is flagged.
Drag: four boxes, six, eight, eleven
Educational illustration. One invented process at one invented bank, drawn four times by the same people over the same work. The rule that a coarse box scores the rounded average of the steps inside it is a simplification. Somebody scoring a real coarse map is looking at a box and not at what is in it, so scoring the box directly lands in much the same place. Figures are the invented bank's own.
The error that gets made, and what it costs
The first map drawn at Sumeru Bank Limited had four steps: receive, assess, decide, communicate. The four step map was drawn by people who knew the process, in an afternoon, and it was a reasonable first attempt at a genuinely hard thing. A box big enough to hold both keying a field and writing an assessment scores the average of the two, so every one of the four steps scored respectably.
On that map the process looked automatable from end to end and the judgement was invisible. Not because anybody hid it, but because it had been drawn inside a box with several other things. The lowest score on the whole map was 7, and 7 clears the line.
The eleven step map is the same process. Nothing about the work changed between the two drawings. The change is that the assessment and the decision became steps of their own, scoring 3 and 4, and the conversation about the decision became possible at all. A map drawn at the wrong size does not produce a wrong answer. The answer it produces is a confident one, and a confident answer is much harder to argue with than a wrong one.
What does the map still not show, however carefully it is drawn?
Three things, and naming them is part of handing the map over.
The first is the waiting. A map records minutes of work, so the two working days a file actually takes are nowhere on it. Every gap between steps is invisible, and the gaps are where almost all of the elapsed time lives. The second is the awkward file. The file sitting on the desk on the day somebody comes to watch is the file that arrives most often, and that is the file times and scores get taken on. The rare one, where a document is in an unusual layout or two names do not match, behaves differently at every step and is not on the map at all. The third is the largest: a map records the process as somebody described it, and a described process and a running process are never quite the same thing. How the desk works at four in the afternoon on the last working day of the month may not appear on any row.
None of the three is a reason not to draw the map. Naming the three at the bottom of the map is how the next person to pick it up knows what they are holding.
Name one thing a carefully drawn map still misses.
How does a lender, a reviewer or an operations head actually use the map?
Three different people pick up the same map and read three different columns. The head of the desk reads the hands-on column and the total. A time of 11.0 minutes a file against a monthly volume is how many posts the desk needs, and no other column can size anything. Ismail Sheikh, who runs the exception desk at Sumeru Bank Limited, reads the decision column. Every step marked assist or leave is work that is still coming to him after the build, and every step marked automate is work that is not.
A reviewer reads the score column beside the decision column, looking only for rows where the two disagree. Neelima Rao, in the risk function, did the bank's later independent validation without having built any part of the chain, and a row reading score four, decision automate is exactly the row that work starts from. The map is also the thing a build gets measured against afterwards: Sumeru Bank Limited spent Rs 2,40,00,000/- to build the chain and Rs 65,00,000/- a year to run it, and the only honest way to ask whether that was worth it is to go back to the rows and see which steps actually stopped being done by people.
The household version of the same reading is a quotation for work on a flat. One person reads the total, one reads the line items, and one reads the line that says what is excluded. All three are looking at the same sheet of paper, and only the third one finds out that the wiring is not in the price.
Sources
| Source | Document | Site |
|---|---|---|
| Reserve Bank of India | Expectations on a regulated lender covering outsourcing, digital lending, record-keeping and consent | rbi.org.in |
| Securities and Exchange Board of India | Expectations where the deployer of such a process is a market intermediary | sebi.gov.in |
| Ministry of Corporate Affairs | Material on the accountability of a board for what it approves | mca.gov.in |
Sumeru Bank Limited, Ismail Sheikh and Neelima Rao are invented.
Educational material. Not advice on any investment, tax, budget or market position.
