In short: The transitional service agreement end date is the only fixed point in a divestiture, and every planning decision should be scheduled backwards from it. Three pieces of work take longer than the deal team assumes: splitting one hierarchy into two when a large share of items, locations and customers belong to both sides, re-establishing identifiers that cannot simply be copied because they carry the seller's GS1 company prefix or a regulatory registration, and running a shared plant that serves two owners with two forecasts and one capacity. The history split is the quiet one, because a divested business inherits demand series that were allocated rather than observed, and forecasting on them without saying so produces confident numbers with nothing behind them.
The announcement lands on a Tuesday and by Thursday somebody has asked the planning team a question with a real answer in it. The Poznan plant makes fourteen SKUs, six of them for the business being sold. The buyer does not want the plant. The transitional service agreement gives them supply from it for twenty-four months. In the meantime there is one item master, one production schedule, one set of customer records where a supermarket group buys both portfolios on one purchase order, and one planning system with one instance.
Twenty-four months sounds generous on a Thursday in month zero. It contains one system build, one master data programme, two peak seasons, a regulatory transfer with an external timetable you do not control, and a set of retail listing changes that can only be requested at fixed points in a customer's year.
Post-merger integration has the opposite problem and its own treatment (Z6). Separation is harder to schedule because the deadline is contractual and the work has an order that cannot be compressed.
The TSA date is the only fixed point
Everything else in the programme is an estimate. The transitional service agreement has a date, an exit fee if it is extended, and usually a step-up in the monthly charge for each extension period. That makes it the spine of the plan.
Work backwards from it and the sequencing falls out. The last thing to happen is the buyer running their own plan on their own data. Before that they need a planning system with two cycles of parallel running behind it. Before that they need master data they trust. Before that somebody has to have decided which records are theirs. Each of those has a real duration, and stacking them against a twenty-four month clock is usually the first honest conversation in the programme.
The incomplete contracts literature is a fair description of what a TSA is. Hart and Moore's 1990 paper in the Journal of Political Economy makes the argument that contracts cannot specify every contingency, so what matters is who holds residual control rights when something unanticipated happens. A TSA is exactly that kind of document. It says the seller will provide planning and supply services at a stated level, and it will not say what happens when a shared line is short in November and both businesses have a promotion running. Write the allocation rule into the agreement or it gets decided by whoever the plant manager reports to, which is the seller.
Emilie Feldman's 2020 book on divestitures argues that how a separation is structured and executed shapes the value realised as much as the decision of what to sell. That is the sober version of the point. The deal team sets a date, the operating detail decides whether the date is achievable.
Two hierarchies out of one item master
The instinct is to tag every record as retained or divested and run a split. That instinct survives about a week.
The classification has three outcomes in practice, and the third one is where the effort goes. Retained only, divested only, and shared. Shared is larger than anyone forecasts.
Work through the categories. Items: raw materials and packaging components used across both portfolios, which is most of them where the businesses share a manufacturing base. Locations: distribution centres holding both, plants running both, and a third-party warehouse operating under one contract with one client of record. Customers: a retailer buying both portfolios, where the customer master holds one payer, one credit limit, one set of payment terms and one EDI relationship. Suppliers: a resin supplier with one contract, one volume tier and a rebate that depends on total volume across both businesses, which means the day the businesses split, both sides fall to a worse price.
The shared bucket needs a disposition per record type rather than per record. For a shared raw material, both sides get their own record in their own system and the seller keeps supplying under the TSA. For a shared customer, the two businesses need separate trading relationships, which means the buyer has to be set up as a new vendor at that retailer, and vendor setup at a large grocer takes months and can only start once the buyer has a legal entity and a bank account. That dependency chain is worth drawing on one page early, because it runs through functions that do not attend the supply chain workstream meeting.
The count that makes this concrete: take your item master, join it to twelve months of production and shipment history, and classify by usage rather than by ownership. A component flagged as belonging to the retained business but consumed on a divested SKU in the last year is shared, whatever the master data says.
The identifiers you cannot copy
Some records can be duplicated into a new system on day one. Others carry an issuing authority, and those set the real timetable.
The clearest example is barcodes. Under the GS1 General Specifications, a GTIN contains the company prefix licensed to the brand owner. When a brand is divested, the numbers on the packs in the trade carry the seller's prefix. Two routes exist and both cost time. The seller licenses continued use of the prefixed numbers for a transition period, which means the seller's GS1 record still points at products they no longer own and the data pools have to be managed accordingly. Or the buyer renumbers, which triggers new artwork, new listings at every retailer, a period where both numbers are live in the trade, and a demand history that breaks at the renumbering date because the new GTIN has no past.
Global location numbers behave the same way for the sites and legal entities involved in EDI. A ship-from location that changes ownership needs a new identifier under the buyer's prefix, and every trading partner has to be updated before the first order routes correctly.
The same pattern shows up in registrations that sit with a regulator: product licences, food business registrations, hazardous materials filings, customs authorisations such as an AEO status that does not transfer with an asset. Each has an external timetable, none of them cares about the TSA date, and several of them gate the ability to ship at all.
Build a register early with three columns per identifier: who issues it, what the transfer or reissue lead time is, and what stops working if it is late. It is a dull artefact and it is the one that predicts whether the date holds.
The plant that serves both sides for two more years
A shared plant under a supply agreement is a planning problem with an unusual property. Two independent companies produce two forecasts against one finite capacity, and neither can see the other's numbers.
The mechanics that have to be written down before go-live are specific. A capacity reservation, expressed as hours on the constrained line per period rather than as a share of volume, since volume shares drift with mix. A forecast commitment window, meaning how far ahead the buyer must commit and at what tolerance, typically firm inside a short horizon and flexible beyond it. A changeover cost allocation, because a shared line's efficiency depends on sequencing and the sequence favours whoever has the longer runs. And the shortage rule.
The shortage rule is the one that gets skipped and it is the one that will be needed. Work an example. The line runs 400 hours a month. The retained business reserves 260, the divested business 140. In November both sides want 20% more than their reservation. Pro rata against reservation gives the retained business 260 and the divested 140, so both are short by the same proportion, which sounds fair and quietly favours the party whose reservation was set from a historical baseline that included the other's volume. Pro rata against forecast rewards whoever forecasts high, which is a rule that teaches both parties to inflate. A cleaner construction is a reservation that is firm for both sides up to the committed volume, with anything above it allocated by a stated rule such as contribution per line hour, and a documented right for either party to buy the other's unused hours at a set price with a stated notice period.
Write that rule while both businesses are still owned by the same company. Once the buyer's finance director is on the other side of a table, the same conversation takes six weeks.
The history split nobody schedules
The divested business needs demand history to plan with, and the history it inherits is worse than it looks.
Three defects turn up reliably. Shipment history at customer level may be recorded against a shared payer, so allocating it to the divested portfolio means splitting at line level and hoping the line detail survived. Volume that was constrained or allocated is censored, meaning the recorded shipment understates what the market wanted, and the divested business will treat that number as demand. And the split fragments the series: a national account that bought both portfolios on one order becomes two smaller series, and smaller series behave differently. Items that were smooth at the combined level become intermittent at the divested level, which changes the appropriate method to something in the Croston family rather than the exponential smoothing the old system was configured for.
The evaluation discipline is worth importing at the same time. Tashman's 2000 review in the International Journal of Forecasting sets out why out-of-sample testing has to use rolling origins rather than a single holdout, and a newly separated business has every reason to be strict about this, because it is choosing methods on short, freshly reconstructed series where a single lucky split will mislead.
The deliverable that helps is a documented history file handed to the buyer with the reconstruction rules attached: how shared customers were split, which periods were supply constrained and therefore censored, where a GTIN change breaks continuity, and how many months of clean history each item-location actually has. A buyer who receives that can plan honestly. A buyer who receives a clean-looking extract with none of it will build a forecast model on artefacts and discover the problem in their first quarter as an independent company.
Where this stops
Some of this is only worth doing at scale. A divestiture of one brand with forty SKUs and two customers does not need a register of issuing authorities or a capacity reservation model. It needs a spreadsheet and one person who knows the accounts, and treating it as a programme wastes months that the deal timetable does not have.
The bigger limit is that the seller controls the schedule and the buyer carries the risk. Separation work competes for the same planning people who are running the retained business, and the retained business has a current-year number to hit. In practice the separation gets the second half of everyone's attention, which is the real reason TSA extensions are so common. Naming that in the plan, and funding backfill for the retained roles explicitly, is more useful than a governance forum.
There is also a limit on what any of this can do about data that was never good. A divestiture surfaces every master data defect at once, because records that were tolerable while one company owned everything become contractual once two companies read them differently. The separation programme is where the defects are found. Fixing them properly is a longer piece of work than the TSA window allows, so the honest approach is to fix what gates the split and register the rest for the buyer to inherit knowingly.
Pull the item master, join it to twelve months of consumption, and produce the count of records that are genuinely shared between the two businesses. That number sets the shape of everything else.