Financial modeling is the most universally relied-upon and least examined part of the mid-market M&A workflow. Every deal has a model. Almost no firm has measured where the time inside it actually goes. This report breaks down the modeling effort on a representative mid-market mandate, separates the structural work from the judgment work, and identifies where the workflow is beginning to shift.
The figures here are mid-range estimates drawn from how mid-market advisory and sponsor teams describe their modeling process. They are not a survey; treat them as a calibrated map of a workflow, not a statistical claim. The purpose is to make a familiar process legible enough to reason about.
The headline: most modeling time is not modeling
The first thing the breakdown reveals is that the activity called “building the model” is mostly not the part people picture. The picture is an analyst constructing logic — projections, debt schedules, a returns waterfall. The reality is that the construction of logic is a minority of the hours. The majority is spent getting data into a state where the logic can be built at all.
| Activity | Share of modeling time | Type |
|---|---|---|
| Locating and indexing the right financial documents | ~12% | Structural |
| Extracting historicals from PDFs and inconsistent Excel | ~22% | Structural |
| Mapping and reconciling line items across periods | ~14% | Structural |
| Validating and tying out statements | ~12% | Structural |
| Normalization and add-back analysis | ~13% | Judgment |
| Building projections and the returns engine | ~15% | Judgment |
| Sensitivities, scenarios, and output presentation | ~12% | Judgment |
The split is roughly 60% structural, 40% judgment. Six in ten modeling hours go to extraction, mapping, and validation — moving numbers from documents into a workable state and confirming they’re right. Only four in ten go to the analytical work that a client is actually paying the firm to perform.
Why the structural half is so large
Three properties of mid-market deal data explain why the front end consumes so much time.
Source heterogeneity. Mid-market targets don’t hand over clean data. Historical financials arrive as a mix of management-accounts PDFs, exported Excel with broken formatting, scanned audited statements, and the occasional figure that exists only in an email. Each format fails differently and resists straightforward extraction.
Label drift. Line-item labels change across periods. A cost line called one thing in FY22 is renamed in FY23 and split in two in FY24. Reconciling these onto a consistent chart of accounts is slow, manual, and error-prone, and it has to be done before any historical build is trustworthy.
The validation tax. Because the inputs are heterogeneous and hand-keyed, they have to be checked: does the P&L tie to the balance sheet, do the management accounts reconcile to the audited figures, does every month foot? This checking is real work, and it scales with how messy the source data is.
None of this is analysis. All of it is mandatory. It is the structural floor every model is built on top of, and it is where the majority of the elapsed time lives.
The maintenance problem nobody budgets for
The breakdown above measures the initial build. It understates total modeling effort, because it omits the cost that accrues after the model is first built: keeping it current.
A live deal’s data room is not static. Sellers upload revised accounts, additional monthlies, restated figures, QoE reclassifications. Each event should propagate into the model. In practice, propagation is manual and unowned — someone has to notice the new document, find the affected cells, re-key, and re-validate. On a busy deal this maintenance is inconsistent, and the model quietly drifts from the data room until a buyer’s diligence team surfaces the gap.
Teams rarely budget for maintenance because it has no fixed schedule. But across a live process it can rival the initial build in total hours, and it carries more risk, because a stale model that looks current is more dangerous than an obviously unfinished one.
What the structural/judgment split implies
Once you see the workflow as 60% structural and 40% judgment, two things follow.
The structural half is the same on every deal. Locating documents, extracting historicals, mapping labels, validating — the content changes but the task is identical across mandates. This is the part that is mechanical, repetitive, and standardizable. It is also, not coincidentally, the part most exposed to error from fatigue and time pressure.
The judgment half is where the firm’s value actually sits. Normalization decisions, projection assumptions, scenario design, the analytical narrative a number supports — this is the work clients pay for, and it is irreducibly human. It is also the work that gets compressed when the structural half overruns, which it usually does.
The implication is not “automate the model.” It is that the structural floor and the judgment work are different in kind, and the workflow improves most when the structural floor stops eating the time the judgment work needs.
Where the workflow is shifting
The change underway in the better-run mid-market teams is not faster modeling in the old sense. It’s a reordering of when the analyst engages.
Historically the analyst engages at the start of the structural work and is exhausted by the time the judgment work begins. The shift is toward the analyst engaging at the end of the structural work — inheriting a structured, source-traced historical build produced directly from the data room, and spending their hours on normalization, assumptions, and analysis instead of extraction.
Three capabilities make that shift viable:
- Extraction that handles heterogeneous sources without a human keying every figure
- Source-tracing so the inherited build can be audited in seconds rather than trusted on faith
- Change-awareness so that when the data room moves, the affected figures are flagged rather than silently stale
Where these are in place, the 60/40 split inverts in effect: the structural floor is produced rather than hand-built, and the analyst’s hours concentrate on the 40% that was always the point.
The bottom line
The state of mid-market modeling today is a workflow that spends most of its time on the least valuable, most error-prone, most repetitive half of the work, and compresses the valuable half into what’s left. The teams pulling ahead are not modeling faster cell by cell. They are removing the structural floor as a manual task, so that the model becomes the starting point for analysis rather than the thing that consumes the time analysis needed.
Further reading
- Quality of Earnings methodology references on normalization and the validation burden
- The FAST and SMART modeling standards on structure and error reduction
- Macabacus and Wall Street Prep on mid-market model construction
- Prior NaS_OS resources on data-room diligence and source-traceability frameworks