THE ADD-BACK · episode 01

The Savings In The Model

Arc 1 · Pre-IC / confirmatory diligence
In the room: Deal team, investment committee
The question: Do I underwrite the automation savings in this model?

The position

SightingA target automated a back-office process and put $400,000 of run-rate savings into adjusted EBITDA. Three heads out, documented, termination letters on file.

The consensus readSupportable. The headcount schedule ties, the reduction is real, annualise it.

The mechanismThe reduction is real and the saving may not be. Automating an exception-rich process relocates work rather than removing it — out of a role with an owner and a cost centre, into a queue with neither. The heads left the schedule; the work went to a channel nobody codes to a department, so it never appears as an offsetting cost.

The exposureAt 9x, that add-back is $3.6M of enterprise value resting on a cost search nobody ran.

The testAsk for eight weeks of message volume in whatever channel the failures land in. Rising, the saving is overstated. Flat, underwrite it and move on.

The bridge — claimed to realised

The case below is a composite built to the shape of a real quote-to-cash automation. Cited research figures are real and sourced; the two are kept separate throughout. A composite presented as a deal is the fastest way to lose this room.

LineMovesRunning
Claimed — $400,000 run-rate, three FTEs, back-office automation, month 11$400,000
1. The headcount schedule is true and insufficient changes the story

A QoE provider tests an add-back on four criteria: realised or contractually committed, documented, correctly characterised as recurring, and not offset by incremental costs elsewhere. This one passes the first three cleanly. The fourth is where automation adjustments die, and it is the only one requiring the team to go looking rather than tie out. There is no schedule that shows you a cost you have not named.

−$0$400,000
2. The exception queue

Exceptions ran ~40/week at go-live and ~190 by month 11, in a channel nobody had plotted. Three people spend their mornings there; none sits in the process cost centre. Two are in customer service, one is a salaried account manager whose time is therefore invisible. Call it 1.5–2.0 FTE of absorbed load at loaded mid-market cost. The work that survives automation is the work that was never on the process map, because things that reliably work are never documented. The automation removed the corridor, not the question.

−$180,000 to −$260,000$140,000 to $220,000
3. The distribution split, and what it does to the KPI changes the story

The composite reported a 60% reduction in median cycle time. Also true — and mean cycle time went the other way. Automating part of a process creates two populations, a compressed easy majority and a stretched tail, while the operator keeps reporting one number. The median describes the population that got faster; the tail is where the customers who churn live. This moves no money and moves your confidence in every operational KPI in the CIM: management is not concealing anything, they simply do not know. Ask for the 90th percentile on any process metric supporting a value-creation assumption.

−$0$140,000 to $220,000
4. Where I was wrong — I put the boundary where the work looked different where I was wrong

Given a queue with no owner, my recommendation was the obvious one: classify exception-prone transactions at intake and route them to a named person with an SLA. Put the queue on the org chart. It failed structurally, not technically — the classifier had to fire before the system resolved the fact that determined whether a transaction was exception-prone, and that fact lived in a different system. It ran at ~61% precision, producing a second queue of misroutes. The general error: I put the boundary where the work looked different instead of where the information actually was.

What it cost: Operators stopped trusting the routing and quietly began checking everything, erasing the gain on the clean majority too. Six weeks in, the median had regressed most of the way to baseline. Full cost walked in Episode 05. For IC purposes the narrower point: when management presents a remediation plan for a known exception problem, this is usually the plan, and it usually does not work.

−$0$140,000 to $220,000
5. What survives residual

Note what the bridge did not conclude. The automation is not a failure — it genuinely compressed the majority of transactions and the operating leverage on that portion is real. What is overstated is the adjustment, by roughly 45–65% of its claimed value, and the overstatement is systematic rather than deceptive.

+$0$140,000 to $220,000
Realised$140,000 to $220,000

Overstated by roughly 45–65% of claimed value. Systematic, not deceptive — which is why it survives an honest management team and an accurate document set.

The downside

Value at risk · multiple 9x, illustrative

A $400,000 adjustment at an illustrative 9x is $3.6M of purchase price. If the offsetting-cost search lands where the composite lands, $180,000–$260,000 of it is unsupported. Substitute your own comps.

$1.6M to $2.3M of enterprise value — roughly 0.4 to 0.6 turns on the adjustment itself

When it surfaces. Pre-IC it is a price negotiation and costs you nothing but time. Discovered in the buyer's confirmatory diligence at exit it is a retrade after LOI, at the point of maximum asymmetry, against a seller who has already told their LPs the process is closing. The same finding is worth a discount going in and a discount coming out — and one of those is yours.

THE ADD-BACK · episode 01 · diligence pack

The Savings In The Model

The question: Do I underwrite the automation savings in this model?
Paste into a request list or a management agenda. Each question resolves to an artifact, not to a characterisation.

The diligence pack

  1. Export the full message history, with timestamps, of every channel, shared inbox or ticket queue that receives failed or exception transactions from the automated process, for the twelve months pre- and post-implementation.

    Artifact: Raw export. Not a summary.

  2. Provide a headcount schedule for the three eliminated roles, and separately, the current time allocation of every named individual appearing more than twenty times in that export.

    Artifact: Schedule plus named list.

    This is the offsetting-cost search, and it is the question that moves the number.

  3. Provide cycle time for the process at median, mean and 90th percentile, monthly, for twenty-four months.

    Artifact: Transaction-level data, not a dashboard screenshot.

  4. Identify every transaction type requiring a data element not present in the transaction's own record, and for each, describe how that element was obtained before implementation.

    Artifact: Written response from operations, not IT.

    The answer is frequently a person's name.

  5. Provide any remediation proposal, build estimate or vendor scope prepared in response to exception volume, and the board or management materials that reviewed it.

    Artifact: The documents themselves.

    If a remediation is already scoped, the exception problem is known internally and unpriced in the model.

Disqualifier

If the response to question 1 is that no such channel exists, that is not a clean answer. Ask where failures go instead. There is always somewhere.

Sources — every measurement with its sample

ClaimSourceSampleClass
Quote-to-cash automation case: $400,000 claimed, exceptions 40/wk rising to 190/wk by month 11, 1.5–2.0 FTE absorbed across adjacent functionsComposite, built to the shape of a real engagementn/a — compositecomposite
QoE four-criteria test for an add-back: realised or committed, documented, correctly characterised, not offset by incremental costsStandard quality-of-earnings practicen/a — practice standardreported
9x entry multiple used throughoutIllustrative mid-market compn/a — illustrative, arithmetic exposedillustrative

One standing caveat. Every number in this show is somebody else's measurement, and I'll tell you whose, with the sample. None of it is diligence on your deal. Do that yourself.

Episode theaddback-01-the-savings-in-the-modelVersion 1.0Dated 2026-08-03Canonical https://addback.enthropysystems.com/01-the-savings-in-the-model/
Enthropy Systems · addback.enthropysystems.com