SystemManufacturing Intelligence

Yield Loss Analysis

Yield loss analysis that takes each batch from theoretical yield to the tablets packed, opens the record behind each loss, and ranks the losses that keep coming back.

See one batch's lost tablets traced screen by screen ↓
Watch with transcript →
The problem

One batch passed its yield limit with the same rejects as thirteen others

Yield is calculated at each phase because the GMP rules require it: theoretical, actual and percentage of theoretical, verified by a second person. Then it is filed. A batch that lands inside its limits passes review, and its reconciliation sheet attributes most of what it lost to a step.

What nobody sees is that the largest step on that sheet, compression rejects, repeats on batch after batch run with the same punch set. Month-end spreadsheets arrive weeks late without the counter printouts behind them, and when a reconciliation goes over its limit, the pressure is to explain it rather than investigate it.

Built forProduction managersProcess engineersQA batch record reviewersPlant finance controllers

The Problem Today

A batch passes its yield limit, and the loss behind it returns on the next batch and the one after, unseen.

164
FDA FY2025 inspection citations for uninvestigated discrepancies published

Read one batch record at a time

  • eBR pages
  • Counters
  • IPC logs
  • Recon sheet
  • One loss, batch after batchA recurring loss looks like noise inside each single batch
  • Over the limit, explained awayA gap over its limit gets a story instead of an investigation

The Solution

  1. A waterfall for each batchTheoretical to packed, each step opening its batch page, counter or log
  2. Losses ranked across the yearLosses that keep coming back rank by tablets lost and by cost
  3. Over the limit, a deviationLeads from counts and weights go into a deviation QA investigates
  4. A person signs the fixThe owner acts; a second person confirms it after three batches

Built-in controls

  • Each figure opens the record it came from
  • Over the limit is a deviation, never a footnote
  • Nobody verifies or approves their own work

What You Get

Yield stops being a filed number and becomes a ledger of fixes that each line and plant works from.

  • An answer for the inspectorEach explained loss names the record that proves it
  • Clients see where their active wentA loss of the client's material comes with its records
  • A fix on one press reaches allA tooling-life check, once QA approves it, runs on each press
Features

Yield loss analysis features, from one batch's waterfall to the fix

Production managers, process engineers, QA reviewers and plant finance use them to see where tablets went and which losses to fix first. They follow one film-coated tablet batch.

demoSample data throughout: the plant, its client, the people, batches and costs.

  1. 01Yield waterfall

    Batch yield waterfall from theoretical yield to the tablets packed

    • Each phase's loss as its own bar. Dispensing, drying, blending, compression, coating, samples and packaging each show what they cost, so the largest loss stands out first.
    • Bulk yield beside the batch record limit. The percentage of theoretical yield sits next to its limit, so a reviewer sees the margin as well as pass or fail.
    • Unaccounted tablets kept as their own step. What the counts cannot place stays apart from losses with a record, so it is not folded into a bigger bar.
  2. 02Record behind a bar

    Reject counter printout opened from the compression bar, sample pulls netted

    • The press printout, with the lines read highlighted. A reviewer checks the figure against the press's own record instead of a retyped number.
    • Sample pulls netted out of rejects. Press checks routed through the reject gate stop being counted twice, a correction the challenger caught and QA approved as a rule.
  3. 03Recurring losses

    Recurring yield losses ranked across the year's batches by tablets or dollars

    • One switch from a batch to the year. A loss that looked like one batch's bad day shows as a pattern across batches, punch sets and shifts.
    • Ranked by tablets lost or by cost. The production manager fixes the loss that takes the most product first, with the signed cost basis behind each figure.
    • Groups compared as percent of theoretical. Batch size does not decide which press, set or shift looks worse, and the number of batches is always shown.
  4. 04Loss trace

    Root cause traced across batches to a worn tablet punch set, step by step

    • Agents trace while the manager watches. Each step names what was read, from counter printouts to the tooling record and the open work order.
    • The other set on the same press as baseline. Comparing two punch sets on one press separates tooling wear from the press, the granule and the shift.
  5. 05Decision

    Overdue tooling inspection raised as a deviation, with batches sent to QA

    • Challenger points settled before signing. Questions such as press or set, and whether the overdue inspection explains the loss, are answered with records first.
    • A GMP finding kept apart from the cost. Running a set past its inspection point becomes a deviation in its own right, not a line in a savings estimate.
    • Batches go to QA with no agent verdict. The product-impact view stays with quality, and the client hears from QA as the quality agreement says.
  6. 06Unaccounted

    Batch reconciliation over its limit opens a deviation, with the count unedited

    • The reconciliation sheet as it was closed. Each line shows its source and the closed sheet stays as it was, so the record an inspector reads is the one QA verified.
    • Leads proposed, the deviation raised by a person. Counts and weights point to where tablets may have gone, and the production manager opens the investigation the rules require.
  7. 07QA review

    Yield calculation verified by a second person, with impact reviews for QA

    • Each yield check on a working-day clock. QA verifies the percentage of theoretical yield the rules require, and sees which checks are late or blocked.
    • Makers do not verify their own yields. The person who made a record cannot verify its yield, and the record must be opened before signing.
  8. 08Value

    What each fix is worth, played forward with the assumptions in view

    • Options played forward before deciding. Pulling a set now, waiting for the planned change, or slowing the press each show what would still be lost.
    • Only the signed cost basis prices a loss. Finance proposes the cost per tablet and operational excellence confirms it, so plant finance and production work from one number.
  9. 09Loss ledger

    Yield loss ledger: each loss with its owner, next action and due date

    • Lanes from new to closed. Each recurring loss has an owner and a next step, so no loss sits in a report with nobody on it.
    • A fix counts after three batches. A second person confirms the fix only once three batches on the corrected setup show it held.

Four jobs around one yield ledger, from the press floor to plant finance

Production owns the losses, engineering finds them, QA verifies the yields, finance prices what they cost.

Production managers

Own output and the loss ledger, and approve the action on each loss

  • The loss to fix first, with its record
  • Deviations raised from evidence, not memory
  • A fix confirmed by batches that held
Process engineers

Build waterfalls, trace recurring losses and compile the annual yield section

  • Waterfalls ready when records close
  • Traces across presses, punch sets and shifts
  • No month-end spreadsheet to rebuild
QA batch record reviewers

Verify each yield calculation and sign explanations and rules

  • The record behind each figure in one click
  • Checks blocked by a hold shown as blocked
  • Proposed rules held until QA approves
Plant finance controllers

Own the cost basis and price each loss and each option

  • One signed cost basis for each loss
  • Assumptions shown and editable
  • Losses priced the morning records close

A double count, a worn punch set and a counter that reset at login

Four things the agents raised on one product's year, each one a filed yield would have hidden.

Sample tablets counted twice

The counter showed 30,000 rejects, but 1,200 were press-check samples sent through the reject gate; the corrected 28,800 became an approved rule.

One punch set wearing out

One set's 14 batches averaged 2.09% rejects against 0.80% for the other set on the same press: 216,000 tablets.

An inspection point passed

Five batches were compressed after that set reached its inspection point, with its work order open 37 days; each went to QA.

A counter reset at a shift login

About 5,500 rejects went uncounted at a login, so the deviation reaches two earlier batches with the same reset.

demoSample data, from the working solution.

More screens

The batch list, the trail, the limits and the ledger across plants

Behind the waterfall: each batch's status, each step's record, the limits from the master record, and the losses of other plants.

Each batch, one click from its waterfallThe year's batches with punch set, shift, bulk yield, rejects, unaccounted tablets and status, filtered by set or disposition.
Agents' and people's steps in one trailEach step by an agent or a person with its reason and signature meaning, kept with the batch records and exported as a file.
Limits read from the master batch recordYield and reconciliation limits come from the master batch record and change only by change control; learned rules wait for QA.
The same ledger across plants and clientsRecurring yield losses from four plants and three clients, counted one way: percent of theoretical, units and the signed cost basis.

demoSample data, from the working solution.

Seven agents count, trace and price, and none of them verifies a yield

Results move by name, from one agent to the next and then to a person.

Yield accountant

Builds each batch's waterfall from the batch record, the press counter, the in-process log, the sample list and the reconciliation sheet.

  • Changes no recorded count
  • Verifies none of its own calculations
Unaccounted-variance hunter

Looks for where unaccounted tablets could have gone, through the counts and the weights.

  • Closes no discrepancy over the limit
  • Accepts no weight without a record
Loss ranker

Groups losses across batches, punch sets, shifts and equipment as percent of theoretical.

  • Compares no groups without normalizing them
Line and tooling tracer

Links a recurring loss to tooling, equipment, material or shift through the press log, tooling records, work orders and the historian.

  • Names no cause without a record
  • Opens no work order itself
Inspector's eye

Asks of each explained loss, and of the agents' own draft, which record proves it.

  • Lets no unanswered point through
  • Decides nothing
Value modeler

Prices each loss and each option with the signed cost basis, its assumptions shown.

  • Invents no cost
  • Uses no unsigned basis
Review drafter

Drafts the yield section of the annual product review and the client's yield summary.

  • Signs and sends nothing

The plant records it reads, without writing back to any of them

Electronic batch recordEach phase's quantities and yields, read when the records close.
Press and line countersReject counter printouts by station, including sample-gate pulls.
In-process logs and LIMSSamples pulled for press checks and QC, so a sample is not counted as a reject.
Reconciliation sheetsThe bulk reconciliation as closed, with the unaccounted limit from the master record.
Tooling records and work ordersPunch set inspections, compression counts and open work orders from maintenance.
The signed cost basisCost per tablet proposed by finance and confirmed by operational excellence.

What plant teams ask about yield loss analysis

What is yield loss analysis?

Yield loss analysis finds where a batch's tablets went between theoretical and actual yield, and which of those losses keep recurring. Here agents build each batch's waterfall from its records, rank recurring losses across batches and trace their causes, while people approve actions and verify yields.

How is batch yield calculated in pharmaceutical production?

Actual yield is compared with theoretical yield at the end of each appropriate phase and recorded as a percentage of theoretical, then verified by a second person, or by one person when validated equipment calculates it. The agents' calculation is treated as such a system, validated before use, and QA verifies each yield.

What is batch reconciliation in the pharmaceutical industry?

Batch reconciliation balances the good tablets and each counted loss against theoretical yield; what remains is unaccounted and must sit within the batch record's limit. Here a reconciliation over its limit always becomes a deviation, and the closed sheet stays as it was.

Can the agents change a count or close a discrepancy?

No. They read the batch record, counters and logs without writing to them. They propose leads for unaccounted tablets; a person raises the deviation, the engineer attaches the evidence, and QA accepts it or sends it back.

How does it put a cost on a lost tablet?

With the cost basis finance signed: conversion cost plus the value of the active agreed with the client, per tablet. The assumptions behind each option are shown and editable, and no agent invents a cost.

Does it compare lines and plants fairly?

Yes. Losses are counted one way, as percent of theoretical with units and the signed cost basis, so presses, punch sets, shifts and plants compare on equal terms. A fix found on one press can become a check on each press once QA approves it.

When does a yield improvement count as done?

When the loss stays down on the corrected setup. Each recurring loss on the ledger has an owner, a next action and a due date; the owner acts, and a second person confirms the fix only after three batches on the corrected setup show it held.

Bring a batch that
passed its yield limit.

Send one closed batch record with its counter printouts and reconciliation sheet, and we will build its waterfall and rank what repeats.