Margin Architecture
Synthetic worked example — not client data
Worked Example · Managed ServicesThe report said 39% margins. The ledger said 1.9% EBITDA.
A $14.2M managed-services provider is heading into renewal season with a plan to hire four more engineers. The platform’s client margin report shows a 39% blended gross margin with every major client comfortably profitable. The general ledger shows 1.9% EBITDA. Nobody can explain the gap, and finance and operations have each stopped believing the other’s number.
Both teams could reproduce their number. Only the rebuilt model reconciled the complete signal.
Illustrative scope: two entities, fifteen major clients, $14.2M of net revenue, and six raw evidence exports in two currencies. Illustrative fees comprise a $14k Financial Signal Diagnostic and a $38k Financial Signal Fix. Illustrative Diagnostic fees reflect the decision scope, entity and evidence-stream count, evidence readiness, and reconstruction effort. Actual fixed fees are scoped after discovery and evidence-access review. Modeled opportunity amounts are illustrative scenario assumptions informed by the diagnostic findings. They are not realized or guaranteed results.
Reconciling a $697k revenue discrepancy and a $1.834M gross-margin overstatement.
This is the artifact the engagement turns on. Every step is a defect found in the evidence, not an adjustment of opinion. The two systems disagreed on revenue by $697k; correcting that alongside four further defects walks reported gross margin of $5,807k down to a true $3,973k — a $1.834M overstatement in total, and a blended margin of 28.0% rather than 39%.
None of this was fraud. A re-exported March batch was counted twice. Five credit memos sat outside the report’s filter. Labour was costed at a $50 standard rate against billable hours only, while the real loaded cost — US payroll plus Colombian payroll converted at actual monthly rates with social charges — had to be carried across every hour worked. Dedicated client infrastructure was never added to the passthrough configuration. And a subcontractor cost had been counted twice in the other direction, which is the one step that moves in the company’s favour.
Six raw evidence exports.
Cases built on clean data start where real engagements finish. This one starts where they actually begin — with exports nobody has reconciled, in two languages and two currencies.
| Evidence export | Rows | What was wrong with it |
|---|---|---|
| Platform invoices | 1,091 | A re-exported March batch double-counted at $412k; five credit memos totalling $285k excluded from the margin report; four or more name variants per major client; mixed date formats; roughly 14% of amounts stored as text. |
| Timesheets | 1,822 | Client field mixing names, variants and internal codes; 26 correction pairs; 130 bench rows with no client; negative-hour corrections. |
| General ledger (US) | 244 | Split debit and credit columns, comma-formatted text amounts, and a class field empty on every single row — the ledger had no client dimension at all, which is why the platform report existed in the first place. |
| Payroll (Colombia) | 264 | Spanish headers, pesos with dot separators, mixed period formats, and social charges at 45.9% that no one had loaded into the delivery rate. |
| Cloud vendor billing | 396 | Thirty-eight subscription-tag variants, and $465k of dedicated client infrastructure absent from the passthrough configuration. |
| Customer master | 18 | Two active clients missing, one duplicate identifier, one churned client still listed. |
Control totals, not opinions. Platform revenue of $14,897k against ledger revenue of $14,200k is a $697k gap that resolves exactly: $412k of duplicates plus $285k of credit memos. Delivery labour resolves the same way — $3,630k of US payroll plus $1,257k of Colombian payroll is $4,887k of real loaded cost, against $4,120k the report assumed. Every step ties or it does not ship.
What the corrections did to each client.
With revenue de-duplicated, labour costed at its real loaded rate across every hour worked, and dedicated infrastructure attributed, two of the three largest clients turn out to be losing money.
Two reporting levels, stated exactly. Subcontractor cost sits above the client line in the ledger and is attributed to no individual client. The client bars therefore weight to a different figure than the company blended rate, and the difference is precisely the subcontractor pool:
Client-weighted reported margin is 43.8%. Deducting the reported subcontractor pool of $705k gives the company blended reported margin of 39.0%. After correction, client-weighted margin is 32.3%; deducting the true subcontractor pool of $610k gives the company blended corrected margin of 28.0%. The $95k difference between the two pools is the duplicated subcontractor cost reversed in the bridge above.
The interactive figure shown is the company blended rate. Even after the rebuild one cost pool remains unattributed — the difference is that it is now measured rather than invisible.
Two of the top three are underwater.
Ordered by how far the platform report sat from the truth. Atlas and Pinewood consumed 2,050 hours beyond their contract allowances between them, flagged non-billable, so the margin report never saw the work at all.
Meridian is genuinely strong at 38.4% — and is 28% of revenue. Its risk is concentration, not margin. That distinction was unavailable while every client appeared profitable.
What the work meant.
The diagnostic took four weeks and cost $14,000. Most of that was not analysis — it was building a defensible pipeline from six unreconciled exports, with an evidence log, mapping tables, and control totals at every step. The Financial Signal Fix — conformed client facts, a true delivery-cost rate carrying actual currency conversion and social charges, a change-order regime for over-allowance hours, cloud tag governance, and duplicate and credit controls written into the close checklist — cost $38,000. Total illustrative investment: $52,000.
The corrective actions model to roughly $715,000 in the first year: $310k from repricing Atlas, Pinewood and Oakline at renewal with over-allowance and infrastructure evidence in hand, $240k from a change-order regime that bills work already being done, $150k from cloud tag governance and rightsizing dedicated infrastructure, and $15k from duplicate and credit controls. That is a 13.8× ratio of modeled opportunity to fees.
What leadership can act onReprice Atlas, Pinewood and Oakline at renewal
Over-allowance hours and dedicated infrastructure are now documented per client, which turns a difficult conversation into an evidenced one. Modeled at $310k.
Introduce a change-order regime for over-allowance hours
2,900 hours a year were consumed past contract allowances and flagged non-billable. The work was already happening; it was simply never priced. Modeled at $240k.
Govern cloud tags and rightsize dedicated infrastructure
Thirty-eight tag variants concealed $465k of client-specific infrastructure inside general overhead. Modeled at $150k.
Put duplicate and credit-memo controls in the close checklist
The $697k gap was two preventable process defects. Controls stop them recurring silently. Modeled at $15k.
Treat Meridian as a concentration question, not a margin one
At 38.4% true margin and 28% of revenue, the exposure is dependency, not profitability.
Sequence the hiring plan behind the repricing round
Four engineers were to be hired against 39% margins. At 28%, the plan is premature by exactly the repricing it should follow.
Until the margin report, the ledger and the payroll are one number, every renewal and every hire is a guess.
Modeled avoided downside.
- ~$180k — deferring two of the four planned engineer hires until the repricing round lands.
- ~$120k — nearshore pricing exposure, from a quoting calculator that omitted actual currency conversion and social-charge loading.
Neither amount enters the $715k headline or the 13.8× ratio. Both depend on decisions that were not executed — hires not made, quotes not issued. They are shown separately, with an explicit counterfactual label, rather than folded into a figure that would look better and mean less.
Managed-services gross margins average roughly 52%, with 50–60% the standard band and top performers above 70%; average MSP EBITDA sits near 18.4%. A resale-heavy revenue mix explains part of a lower blended figure — but the industry context is what makes the gap legible: the report said 39%, the ledger said 1.9% EBITDA, and the benchmark said neither number was being managed.
Sources: Gradient MSP gross margin playbook; MedhaCloud managed services market statistics. Accessed 2026.
Release controls
- Synthetic worked example — not client data. Kestrel Managed IT Services, its entities, its clients and its people are invented.
- Invented company, real method. The analytical method and the arithmetic are real; the company is not.
- Modeled opportunity, not realized or guaranteed results.
- Impact estimates include illustrative scenario assumptions informed by the diagnostic findings.
- Fees shown reflect this illustrative scope only.
- The Financial Signal Diagnostic and the Financial Signal Fix are separate phases.
Start with the two systems that disagree.
When two internal numbers contradict each other and both teams can defend theirs, that argument is the entry point — not an obstacle to the work.
The other worked examples begin with clean data. This one does not, because most engagements do not. Reconstructing a defensible signal from unreconciled evidence is the work — the analysis only becomes possible afterwards.