Operational excellence · Operational value pools

Where I'd look for hidden value in your operation before proposing AI

How recurring process patterns reveal value pools, how to test the economics and why the intervention should come after the evidence.

Leania · Operational value creation

VALUE POOLS

Recognise the pattern. Test the cause. Size what can change.

Repeat effort
Waiting
Duplication
Expert queues
Illustrative taxonomy of recurring process patterns. These are hypotheses to test, not findings about a specific organisation.

Most businesses I speak to already have a list of things they want to automate. There is usually a spreadsheet full of ideas, a few promising AI pilots and at least one system everybody wants to replace.

What interests me more is what sits behind that list. Which processes are taking far more effort than they should? Where is work getting sent backwards? And which of those problems is actually worth solving first?

After years in process improvement and automation, I've found that you can walk into an unfamiliar operation with some fairly useful hypotheses. You might have only the company's sector, how it makes money, the systems it uses and a few details about its recent growth or acquisitions. That's enough to recognise some recurring process patterns and make a sensible start.

Of course, a hypothesis is an invitation to check the evidence. I've been wrong before. But turning up with a reasoned view of where to look is better than spending the first 15 working days asking everyone for their automation wish list.

I think of the places where those problems concentrate as operational value pools.

The clues I'd look for before the first workshop

Imagine a financial-services group that's acquired six smaller firms over the past two years. There are signs of back-office growth, several inherited platforms and a plan to use AI to improve productivity.

Before I see a single transaction, I would have three working hypotheses.

First, the same job is probably being done in several different ways. Six acquired businesses often means six intake forms, approval habits, exception rules and versions of the truth. That may create duplicated checks, re-keying and inconsistent service levels. I'd want to look at a small sample of the same transaction type across firms, rather than jump straight into a system-replacement discussion.

Second, the cost may be hiding in the work that comes back. A broker submission gets chased for missing evidence. A tax return is opened, put aside, reopened and checked again. A payment instruction passes the automated checks but needs a specialist to repair an exception. These are small moments individually. At volume, they become whole teams' worth of effort.

Third, the real constraint may be five experienced people. Routine cases can flow through a new tool while difficult decisions still wait for somebody who knows the unwritten rules. Automating the simple path might improve a headline metric without moving the backlog or the cost-to-serve.

None of those claims describes the actual group. They're hypotheses based on a recurring operating pattern. I would want case logs, workflow evidence and the people doing the job to confirm or challenge each one.

The important part is that we can arrive with a useful starting point and get to the evidence faster.

A value pool is more useful than a use-case list

Pattern 01

Failure demand

Reopened cases, missing evidence and repeat handling

Where does work go backwards?

Pattern 02

Queues and handoffs

Long elapsed time, short active handling time

Who or what is the case waiting for?

Pattern 03

Fragmentation

Duplicate steps, local variants and overlapping tools

Where do we pay for the same outcome twice?

Pattern 04

Specialist bottlenecks

Exceptions concentrated with a few experts

Which cases consume scarce judgement?

Process-pattern hypotheses for initial investigation. Categories overlap; do not add the same underlying effort twice.

A value pool is a concentration of avoidable cost, delayed value, wasted capacity, quality failures or risk within a business operation.

It gives the COO, Finance and delivery teams a common thing to assess. How large is the problem? What causes it? How much can realistically be addressed? What would an intervention cost? Who would benefit if it worked?

The patterns I keep looking for fall into four families:

1. Failure demand and rework. Cases reopened, documents chased, corrections, repeat contact, duplicate checks. The signal is work being done more than once for the same outcome.

2. Queues and handoffs. A task takes seven minutes but waits four days for a decision. The signal is lead time growing while actual work content stays small. Making a task faster will rarely fix a queue created by ownership or approval rules.

3. Fragmentation and duplication. Different business units maintain their own spreadsheets, integrations, procedures or overlapping software. The signal is teams paying repeatedly to complete what is effectively the same job.

4. Specialist bottlenecks and exceptions. The easy cases are handled well, while the difficult ones queue with expensive, scarce experts. The signal is a small share of cases absorbing a disproportionate share of effort.

These pools can overlap. A rejected case might also sit in a queue and be handled by a senior reviewer. Counting each as a separate saving would overstate the opportunity. That double-counting mistake is surprisingly easy to make when three teams submit their own AI business cases.

Put a number against the pattern

Take an illustrative operation handling 100,000 cases a year. Suppose 30% need one additional handling pass, taking eight minutes on average.

That means 30,000 repeat handling events, 4,000 hours of extra work and around £140,000 a year of capacity cost at an assumed £35 per hour.

Illustrative calculation · Not client data or realised savings

100,000

Cases per year

30%

Require another handling pass

8 min

Extra effort per affected case

£35/h

Assumed fully loaded labour cost

Candidate value pool

4,000 h · £140,000

(100,000 × 30% × 8) ÷ 60 × £35/h

If 50% of repeat effort proves avoidable

2,000 h · £70,000

Potential annual capacity value, before change costs

A capacity valuation is not a cash saving. Benefit realisation depends on changes to staffing, overtime, demand or output.

The £140,000 is a candidate value pool. The company has already paid for the effort. We still need to know why those cases came back and what proportion of the work could actually be avoided.

Suppose sampling suggests half the repeat handling could be removed by fixing two upstream causes. The working opportunity becomes 2,000 hours, or £70,000 of annual capacity value before the cost of making and running the change.

Now comes the awkward but necessary question: what happens to those 2,000 hours?

If the firm can avoid overtime or the next recruitment round, some of the value may turn into cash. If the same team simply has more room to serve growing demand, that is capacity and operating leverage. Both can be worthwhile. They belong in different columns of the business case.

I would also want to check that we are measuring unique cases and real extra effort, rather than counting multiple status changes as several incidents. Otherwise the number can look convincing and still be wrong.

This is why I don't get too excited when someone tells me their AI pilot saves 90 seconds per transaction. I want the denominator, the exception rate, the end-to-end effect and the business decision that will capture the benefit.

The same £70,000 pool can lead to very different solutions

Suppose the evidence shows that most of the extra handling comes from incomplete information at intake. There are several plausible ways to respond.

We could remove a request for data nobody uses. We could simplify the form and put the checks where the information first arrives. We could standardise one rule across acquired firms. We could improve the workflow so that missing evidence is resolved before work reaches an expensive specialist. We could automate a deterministic check. Or we could use AI to interpret genuinely varied documents where that adds value.

Those are different interventions with different costs, control requirements and failure modes. The most impressive demonstration is not automatically the best investment.

I'd compare them against the same target: how much repeat handling each one is likely to prevent, what it costs to operate, how quickly it can go live, what could go wrong and how confident we are in the evidence.

Sometimes AI wins. Sometimes the strongest result comes from changing the question a customer is asked at the beginning of a process.

How I'd turn the first day into a better investment decision

I would start with one bounded workflow, not the whole enterprise. Perhaps broker onboarding, claims evidence handling, tax document intake, a payment repair queue or an account-opening exception process.

My first pass would be straightforward:

  • Trace the work. Take a handful of real cases from arrival to completion, including those that failed or were sent back. Compare what happened with the official procedure.
  • Find the concentration. Check where the extra touches, waiting and specialist intervention sit. Segment the cases by the most common failure reason. A process map should show where evidence is missing, rather than quietly guessing.
  • Size the candidate pool. Use annual volume, affected share and incremental effort to produce a sensible range. Record assumptions and ask Finance what counts as cash, avoidance or released capacity.
  • Choose the next proof. Identify which suspected cause is both material and uncertain. Test that before committing delivery money to a particular technology.

AI can help assemble evidence, spot patterns, suggest questions and prepare a credible first-pass map. I'm interested in using it to cut the time expert practitioners spend getting from scattered evidence to an informed decision. The expert still needs to check the sources and own the recommendation.

That is also why I'm building specialist software alongside Leania's consulting work. The aim is to make the discovery and assessment work faster and more repeatable, while retaining the evidence and judgement a client should expect. Those capabilities are still being tested and developed; the delivery method matters more than a flashy tool name.

Why I care about the portfolio decision

In one earlier transformation programme, I led the process re-engineering and automation workstream as we worked through more than 200 submitted ideas. We narrowed them to 44 assessed opportunities, representing around 30,000 hours of potential annual capacity. The first three implemented improvements released a reported 10,000 annualised hours.

Those figures describe different stages. A pipeline is not a saving, and releasing time is not automatically reducing cost. But the work taught me something important: a good improvement programme has to make it easier to say no to attractive but low-value ideas, not simply process the backlog faster.

Here's a useful test to run this week. Choose a high-volume workflow that's attracting AI proposals. Ask for one month's completed cases, including the failures. How many needed another touch? Where did they wait? Which two causes account for the largest share of the extra effort? And if you removed it, what would the business actually do with the freed capacity?

You may find a better investment case before anyone writes another prompt or buys another licence.

If one workflow in your operation seems to be absorbing too much cost, delay or specialist effort, I'm happy to compare where I'd start looking.

Where to start

One engagement, two stages, one decision between them

21 working days to a Pilot Blueprint

  1. Stage 1 · Working days 1–10Establish, then Diagnose & Select
  2. Stage 2 · Working days 11–21Design, then Specify & Challenge
The decision at day 10

Stage 1 runs working days 1–10 and ends in a decision. If the evidence doesn't justify continuing, the engagement ends there: you keep the evidence and pay only for Stage 1. If it does, Stage 2 takes the selected intervention to a Pilot Blueprint by working day 21.

£7,500 to start. Stage 2 (£9,500) only if the evidence justifies it.

£17,000 — Total if both stages run, for one bounded workflow. Larger scopes are quoted.

See how the Evidence Sprint works →

Have this problem in a live workflow? Discuss the workflow.

Discuss the workflow

← All insights