The Platform PM
Playbook

Integrating a platform after an acquisition

What to do with the products after you buy a platform company: sort every product and component into kill, keep independent, migrate or shared services, sequence the work, keep customers and revenue, and clear the licences, banks and registrations that slow regulated deals.

Last updated October 2026

The big picture

Your company has bought another platform company, or is about to. The deal team owns the price and counsel owns the filings. You own what happens to the products: which outcome each acquired product and platform component gets, what order the work happens in, how customers move without leaving, and who decides what.

Every component gets one of four outcomes.

1. Kill

What happens
Sunset it; move its customers to you or let them go
Choose it when
It duplicates yours and earns less than it costs
Usual cost
Lost customers; notice periods

2. Keep independent

What happens
Its own product, stack and often team
Choose it when
Its value is a market, team or trust you lack
Usual cost
Two of everything; drift

3. Migrate to your product

What happens
Its customers or functions move onto your product and stack
Choose it when
Similar products; a cost or scale thesis
Usual cost
Engineering years; churn per move

4. Shared platform services

What happens
Both stay, on common identity, billing, data, messaging, compliance
Choose it when
You need both products and one customer view
Usual cost
Years; it stalls
Four outcomes for every acquired product or component: kill it, keep it independent, migrate it onto your product, or put both products on shared services. Shared services look cheap on a slide and take years, and they often stall.

Most acquisitions miss at least one goal they set, usually the revenue one. The popular "70-90% fail" line doesn't hold up (Words that mean something else here), but a big 2004 meta-analysis did find that buying companies doesn't, on average, make acquirers perform better.

What I'd run this on:

  1. Timing: decide early, and migrate at the pace risk allows. Integrating tasks and systems fast helps; integrating people fast hurts. Whether speed pays at all depends on how related the two businesses are, and the "first 100 days" is mostly symbolic. What delay really costs is uncertainty, and an early, honest preview of the changes reduces it. The studies on speed are solid. The point about decisions rests on cases: the biggest losses I found came from wrong calls rather than late ones (FIS-Worldpay, Ericsson-Vonage), so day 100 is my deadline for deciding and never for integrating.
  2. The entity: it's the legal entity that's regulated, and the architecture behind it mostly isn't. I'm confident about the law; what it means for product is my own reading. A stock deal leaves licences, registrations, bank contracts and certifications with the same entity. The cost comes when you move customers, funds, numbers or data between entities, change who controls one, or change where a regulated service runs. If you buy a payouts company licensed in 40 states, its licences keep working while its customers stay on its entity, but once you move them onto yours your entity needs those licences too. Even a plan can slow approval: in the US money transmission model law, as Iowa enacted it, the fast track requires that the licensee "will not implement any material changes to its business plan as a result of the acquisition".
  3. Shared platform work: it looks cheap on a slide, then takes years and stalls. The cases all point the same way, though there's no large study. LinkedIn's Azure move, announced three years after the deal, was shelved in December 2023. GitHub stayed in its own data centers for about seven years, then started a 24-month move because it was running out of capacity. Trello accounts became Atlassian accounts about three years after that deal, and Fitbit's Google-account deadline slipped twice. The exception is Instagram, which moved into Facebook's data centers in 2013-2014, a young product landing on a mature platform.
  4. Customers: they judge the integration by what changes for them. Across more than 2,000 deals, customers were less satisfied a year on and firm value was about 2.4% lower than at comparable firms; leaders who kept their eye on customers lost less, a 2022 study found. I'm sure of the direction, much less so of how much churn to expect or when. There's a limit too: when switching is hard, forcing change can raise revenue while it builds regulatory trouble, as Broadcom's VMware bundle did.

Later sections add that sales and people collisions do the earliest revenue damage (phase 2) and that independence is usually a phase before something else (Keep independent).

In practice: decide every component's outcome, and the people and structure around it, by day 100, then move customers only as fast as entities, contracts and reliability allow.

Regulated platforms add their own drag:

  • Payments and cross-border payouts: US state money transmitter licences are "not transferable or assignable". The model law, adopted in about 33 states, needs prior approval of a new controller, and so do New York and California. The sponsor bank's contract decides what you may change.
  • Telco and CPaaS: 10DLC brands (the registration behind business texting in the US) are keyed to a tax ID. Since September 18, 2025 each provider has signed calls with its own STIR/SHAKEN token, and a stale Robocall Mitigation Database filing can get your traffic refused.
  • Security and identity: a SOC 2 report or PCI DSS attestation covers a defined scope, and your new product isn't in it. FedRAMP wants 30 business days' notice of a "transformative" infrastructure change. And merging logins means every enterprise customer reconfigures its single sign-on.
  • All of them: putting acquired customers' data on your entities usually means a sub-processor notice to every customer, typically 30 days, and they can terminate.

01Signing to closing

Before close

Every acquired product and component has a draft outcome on paper, every entity-bound asset and contract term is inventoried, and nothing you did steered the target.

Until closing you can plan, but you can't steer the target. You're still competitors: "Right up until consummation, the merger parties are still independent businesses," as the FTC's competition bureau put it in 2018. The HSR Act bars the buyer from taking operational control before the 30-day waiting period ends, and the Sherman Act bars coordinating prices, customers or bids until closing, whether you filed or not. Enforcers have gone after the corporate version of everyday product moves:

Computer Associates and Platinum (DOJ, 2002)

What the buyer did
Pre-approved target discounts above 20%; coordinated bids
Result
$638,000; consent decree
Product equivalent
Approving their deals or prices

Three crude-oil producers (FTC, January 2025)

What the buyer did
Approved spending above $250,000 and routine hires; halted planned work
Result
$5.6M, a record HSR penalty
Product equivalent
"Pause that roadmap item"

Altice and PT Portugal (EU, 2018)

What the buyer did
Vetoes over appointments, pricing and contracts
Result
€124.5M; largely upheld in 2023
Product equivalent
Veto rights over launches

Facebook and Giphy (UK CMA, 2021)

What the buyer did
Breached a hold-separate order after closing
Result
£50.5M
Product equivalent
Integration frozen after close

Illumina and Grail (EU)

What the buyer did
Closed while reviews were pending
Result
€432M fine fell away (2024); deal unwound
Product equivalent
Integrating before clearance

The FTC treats prices, discounts, costs, strategic plans, future products and customer-level data as competitively sensitive. Share only what diligence needs, through a clean team counsel has vetted, with aggregated data and masked customer names. If both companies sell email APIs to retailers, the clean team can tell you about a third of the target's revenue comes from retail, but only they see which retailers, and at what price. I'm not a lawyer, but as I read it, roadmaps, unreleased features, per-customer pricing and usage, win-loss data and pipeline are clean-team-only whenever the companies compete or could. So plan on paper, analyze overlap in aggregate, draft Day-1 messages and prepare filings. Don't approve the target's deals, pause its roadmap, direct its staff, sell jointly, tell its customers a product will die, or connect systems and data. Interim covenants, joint planning and retention agreements are grey areas: ask deal counsel.

Diligence should serve the four outcomes as well as the price. I'd build these inventories:

  • a regulated-asset register keyed by legal entity: licences, bank programs and BINs (the card number ranges a bank issues under), network registrations, PCI attestations, telco identifiers, SOC 2, ISO and FedRAMP scopes, data processing agreements and residency promises, each with an owner, renewal date and change-of-control trigger;
  • a contract inventory with revenue counted per clause type;
  • a platform inventory: APIs, SDK versions, webhooks, traffic and revenue by segment, and transition services.

Check foreign investment rules early. CFIUS conditions on data access or location can rule out shared services, and in the UK a deal that needed a National Security and Investment Act notice and didn't get cleared is void. And file the plan you'll actually run: if your filings promise no material changes while your deck assumes a migration, you lose the money transmission fast track and some credibility.

Write the Day-1 promise: what won't change for customers (APIs, pricing, support, data location), for how long, and who may change it. When Okta rebranded Auth0 in 2023 it promised no changes to URLs, APIs or code libraries. I'd copy that, for at least 12 months. Integrating fast after close takes longer preparation before it, as GE Capital's acquisitions showed.

Checklist

  • Get counsel's written gun-jumping protocol: what may be shared, who sits on the clean team, what stays with the target
  • Keep people with pricing or roadmap roles off the clean team, and approve nothing for the target: no discounts, contracts, hires, launches or pauses
  • Draft an outcome for every acquired product and major component, with your confidence and open questions
  • Build the regulated-asset register by legal entity, with an owner, renewal date and change-of-control trigger for each item
  • Count revenue exposed to change-of-control, assignment, sub-processor, residency and SLA terms in customer contracts
  • Map the filings that gate closing (state licences, FCA, PSD2, RPAA, CFIUS or NSIA, FCC 214) and check they describe the plan you'll run
  • List the people the thesis depends on and agree retention terms and future roles with HR and counsel
  • Write the Day-1 promise (what won't change for customers, for how long, who can change it) and draft Day-1 messages, sending none before close

02Days 1-100, with gates at day 30 and day 100

The first 100 days

Customers, people and revenue are stable, and every acquired product and component has a decided and announced outcome, an owner and a date.

Stabilize by day 30, decide by day 100. I keep "100" because a deadline forces decisions; integration takes far longer.

By day 30: stabilize

Start with people, because the best ones leave soonest. In about 4,000 acquisitions of US tech startups, a third of acquired workers left in the first year, nearly three times the rate of comparable hires, and top performers left most, a 2024 study found. Senior, visible executives go soonest, their leaving hurts performance, and giving them real top-team roles helped. The gap is widest in senior, technical, business development and sales roles, and it shrinks when founding teams stay together and units stay separate. So announce the acquired team's top two layers in the first weeks; GE Capital did it within days. Money holds people until a cliff date and stops working after it. HR and compensation leaders surveyed in 2024 rated career opportunities the best non-cash tool.

Sales and people collisions do the earliest revenue damage, I think, though that rests on one strong case plus the attrition studies. Okta merged its sales forces while buyers were still unsure which product to lead with; sales attrition rose, and about 16 months after close its long-term targets went under review (example). Picture both companies selling login products to mid-size software firms. Without a rule, two of your reps can pitch the same CTO two different products in the same week. Write down who sells what to whom before you merge anything.

Give customers a name: each acquired product gets one visible owner. Top accounts hear who it is and what won't change, in calls where you mostly listen. Freeze customer-visible changes except security fixes.

File what the change of control triggers: FinCEN re-registration within 180 days of a transfer above 10%; FINTRAC and RPAA updates within 30 days; the Robocall Mitigation Database within 10 business days; Form 499 within a week; FCC numbering and international 214 notices within 30 days. Then tell the auditors and meet each sponsor bank.

By day 100: decide and announce

By day 100 every component should have an outcome, an owner, a cost and dates. Use how to choose an outcome, what moves with the deal and the cost model, and write each call up as a decision record, so reversing one later means marking it "superseded" instead of reopening the argument.

Decide internally, and announce only once the migration path exists. Twilio announced the end of Programmable Video in December 2023, extended it after protests and reversed it in October 2024. Meta announced Parse's shutdown with an open-source server, a migration tool and a year's notice. Tell top accounts privately first.

Plan the dis-synergies too. Lost customers are the revenue hit deal models usually leave out, as McKinsey pointed out in 2004. Track key-people retention, renewal intent, market share and release velocity alongside synergies; FIS beat its synergy targets, then wrote down $17.6bn (example).

Be explicit about what isn't decided yet. My cadence: a Day-1 all-hands on what is and isn't decided, a weekly written update, and a dated calendar of open decisions.

Checklist

  • Day 30: announce leaders and reporting lines for the acquired team's top two layers
  • Day 30: confirm retention terms and roles for the people the thesis depends on, and review departures weekly
  • Day 30: name one owner per acquired product and tell top accounts who it is and what won't change
  • Day 30: freeze customer-visible changes (pricing, API deprecations, support model) except security fixes
  • Day 30: write "who sells what to whom" rules before any sales merger
  • Day 30: file the post-close regulatory notices and tell auditors and certification bodies
  • Day 100: record an outcome, owner, cost, revenue at risk and dates for every component in a decision record
  • Day 100: budget two stacks, with dis-synergies and churn as plan lines next to synergies
  • Day 100: write the trigger and end state for every unit you keep independent, and publish a dated calendar of open decisions

03Months 4-12

Build the bridge

The migration path exists and has been rehearsed, new business goes to the target platform, and the public notice is out early enough to give customers at least 12 months.

Expect to budget for architecture you'll later throw away. Martin Fowler's strangler fig grows new components around the old system until it can be retired, and you pay for "transitional architecture" along the way. Two rules from Cartwright, Horn and Lewis's writing on legacy displacement:

  • Legacy mimic: your platform impersonates the old one toward whatever can't change yet (a partner bank's files, a carrier interface, customers' code). A facade that speaks the acquired API, webhooks included, lets customers cut over without changing code. A freight customer's system, for instance, posts to the acquired company's shipments endpoint and listens for its status webhooks; the facade accepts the same calls and sends the same payloads, while your platform does the work behind it. Stripe's "version change modules" show how to keep old behavior contained.
  • Not feature parity, which turns every undocumented behavior and workaround into a requirement. Ask instead: parity for which customer jobs, held by what share of revenue?

Run the new path in parallel before you switch. GitHub ran a rewritten merge path beside the old one, returned the old result, logged mismatches and ramped from 1% to 100% of traffic over four days. It found bugs in the old system too. Stripe moves live data by dual-writing and backfilling, then moving reads, then writes, then deleting the old.

Shared groundwork goes in roughly this order, which I've taken from the cases rather than any study: internal services (employee identity, finance reporting, security monitoring), then observability and on-call, customer identity, billing and entitlements, messaging and payment rails, and compute last. Identity can move in bulk or lazily at each user's first login. Either way inactive users need a final sweep, and each enterprise customer reconfigures its own SSO. For billing I found no public timeline.

Move new business first. From September 1, 2019 Microsoft sent new Skype for Business Online customers to Teams, while existing ones kept the old service until July 31, 2021. In payments, new customers go to the new processor while existing cards stay on the old one until they're imported. Stop selling the old product when you give notice.

Give at least 12 months' notice, which is the floor big platforms commit to for generally available APIs:

PlatformPublished commitment
Google Cloud12 months' notice before discontinuing a service or breaking a customer-facing API
SalesforceEach API version at least 3 years; 1 year's notice before retirement
ShopifyEach quarterly version at least 12 months, with 9 months' overlap
Meta GraphEach version at least 2 years
GitHub RESTPrevious version at least 24 months after a new one

Working back, a shutdown at month 24 needs the notice out by month 6 to 9. Mark endpoints with the Deprecation header (RFC 9745, March 2025) and the date with Sunset (RFC 8594). For deep integrations, regulated customers and registrations I'd give 18 to 24 months.

With regulated pieces the order matters. Move customers between entities only after the destination entity holds the licences, bank program, registrations and certifications. Start third-party work early. 10DLC campaigns move between providers in minutes without re-vetting if the connectivity chain is kept ("clear chain" wipes the vetting). An old card processor takes "a few days or several weeks" to release card data, and declines can spike after the move.

Design the rollback before the cut-over. Keep the old system warm, make writes reversible, name the point of no return (the first settlement file, carrier route change or SSO switch) and rehearse at production scale. TSB tested only one of its two data centers (example).

Watch the retention cliff as well. Retention deals usually run 12 to 18 months, so they end mid-migration (that's from an older survey I've only seen summarized). Stagger them for people on the critical path, and say which feature work is paused.

Checklist

  • Build a compatibility facade for the acquired API and webhooks, with Deprecation and Sunset headers, and run the new path in parallel on production traffic with logged mismatches
  • Start shared identity and billing groundwork: the SSO migration path, the entitlements model, invoice and tax mapping
  • Onboard new customers to the target platform and end sales of the old product
  • Publish the public notice with dates at least 12 months before shutdown, longer for regulated customers and deep integrations
  • Ship migration tooling: self-serve for the long tail, assisted for larger accounts, and a progress view customers can see
  • Before customers change entity or shared services touch regulated flows, have the destination's licences, bank program, registrations and certifications ready, file outsourcing and sub-processor notices, and get the sponsor bank's sign-off
  • Run a pilot wave with friendly customers and rehearse rollback at production scale
  • Stagger retention cliffs for people on the migration's critical path

04Months 12-36

Migrate in waves and decommission

Customers have moved, or left on purpose; the old stack is off, reconciled and archived; its licences and registrations are surrendered; and every unit you kept independent has been reviewed.

Segment customers, then move them in waves. My default: top accounts get a named plan and engineer, mid-size accounts move in waves led by customer success, and the long tail gets a tool, reminders and a date. Different deadlines by tier are fine if you say so up front: Google stopped standard Universal Analytics on July 1, 2023 and gave paying 360 customers another year. Every wave has entry and exit criteria, a rollback path, a customer message and stop rules agreed in advance.

Paying for the customer's move keeps showing up in the cases. My guess is that these offers beat the alternatives, but I haven't seen that measured.

  • Fund their dual-run: Atlassian waived server renewals for up to 12 months after a cloud purchase and gives Opsgenie customers 120 days of parallel access; Palo Alto Networks offered free migration services to IBM QRadar SaaS customers.
  • Make it an upgrade (Skype users signed in to Teams Free with their Skype credentials and contacts), step price protection down rather than off (Atlassian's loyalty discount went from 40% to 20%), and give an exit to those who won't move (Parse open-sourced its server).
  • Avoid killing before the replacement exists (Twilio Video), very short notice (Heroku's free tier got about 95 days) and price shocks bundled with the move (VMware).

The deal itself rarely needs customer consent, but the migration often does. Under Delaware law a reverse triangular merger isn't an assignment of the target's contracts (Meso Scale v. Roche, 2013). But moving a customer onto your product usually means a new contracting entity, terms and sub-processors. That's a renewal-like moment, when the customer can renegotiate or leave. Take a hospital group whose contract names the acquired company's entity and lists its sub-processors. Moving it to your platform changes both, so its procurement team has to sign, and will likely reopen price. Under the EU Data Act, customers can switch on two months' notice, and switching charges are banned from January 12, 2027. Run consent campaigns before you announce dates.

Expect the tail to outlast the plan. Twilio moved Notify's end of life twice, to December 31, 2025, and Fitbit's deadline moved twice too. Publish one date and budget the old stack for 6 to 12 months beyond it. Never shorten a published date: pulling CentOS 8's end of life forward from 2029 to 2021 caused lasting backlash. Let an account go when migrating and serving it costs more than its lifetime margin, and give it an export and a referral.

Reconcile before you switch anything off. In payments, reconcile three ways every day of the dual-run (old ledger, new ledger, bank or scheme statements), never net mismatches off, and switch off only after zero unexplained breaks for agreed periods. For a payroll platform paying out $20M a day, a $3,000 gap between the old and new ledgers on Tuesday has to be explained before Wednesday's run. After the middleware company Synapse went bankrupt in April 2024, its ledger and its partner banks' records turned out to be up to $95M apart, and end users were frozen for weeks or months.

"Migration complete" really means three milestones: customers moved; old system off; contracts, data, registrations and certifications closed. Archive before deleting: PCI DSS keeps 12 months of audit logs, three of them immediately available (requirement 10.5.1). Give leavers at least 30 days to get their data out, then surrender the licences, registrations, leases, numbers and bank programs you no longer need.

Measure progress in traffic and revenue moved rather than in accounts. GitHub's reports on its Azure move count workload by workload.

Review what you kept independent against the trigger you wrote on day 100: continue, shared services, migrate or sell. Then write the post-mortem. In Zollo and Singh's study of 228 US bank deals, writing integration know-how down improved later deals, while plain experience didn't.

Checklist

  • Run waves by segment, each with entry and exit criteria, a rollback path and a customer message: named plans for top accounts, waves led by customer success for mid-size accounts, a tool and a date for the long tail
  • Enforce stop rules: pause a wave when mismatches, reconciliation mismatches, authorization or delivery rates, or top-account escalations cross agreed thresholds
  • Run consent campaigns for contracts that need assignment or new terms before announcing their dates
  • Budget one extension and segment it by contract value if needed; never shorten a published date
  • Review the scorecard weekly: traffic and revenue migrated, accounts by stage, deprecated calls, breaks, old-stack cost
  • Switch off old access only after zero unexplained reconciliation mismatches for agreed periods, archive to retention rules, and give leavers an export window
  • Surrender unneeded licences and registrations, retire BINs, release numbers and short codes, and close extra compliance scopes
  • Review every independent unit against its written trigger, then write the post-mortem and update your playbook

How to choose an outcome

Classify each product and component on its own. The classic typology, from Haspeslagh and Jemison, sorts whole deals into absorption, preservation, symbiosis and holding, which map roughly to migrate, keep independent and shared services. But how much you integrate and how much autonomy you leave are separate dials, and you can set them per component. Kill isn't in the typology, yet it's common: about half the consumer apps that five big tech companies bought from 2015 to 2019 were shut down, one study found.

What I ask of each component:

  1. Where does the value come from? Cost or scale from similar assets points to migrate or kill; capabilities, talent or a market you lack point to keep independent or shared services.
  2. How mature is it? Integrating the structure made targets that hadn't launched yet less likely to launch new products, one study found.
  3. Who would leave, and what would customers have to change?
  4. What do regulators, licences and contracts allow? Microsoft gave the European Commission five-year commitments on LinkedIn.
  5. Can your stack carry it? GitHub's Azure move was triggered by running out of capacity rather than by a synergy plan.
  6. Does it need a different capital model? FIS said its merchant business needed acquisition capital "we just cannot feed" inside FIS.

Kill

Choose it when the product duplicates yours, costs a large share of its revenue to run, or its customers want your product anyway. In my cost model, a $5M-a-year product costing $6M to run is better killed on almost any retention assumption.

What it costs: the customers who don't follow, notice periods, credits and an archive. Atlassian bought Opsgenie in 2018, stopped selling it on June 4, 2025 and will shut it down on April 5, 2027, deleting unmigrated data: about 25 months from end of sale to shutdown.

Early signs it's wrong: extension requests in the first weeks; escalations because the replacement can't do the customer's job yet.

Regulated platforms: bank program wind-downs (BIN retirement, cardholder notices), number port-outs, short-code and campaign decisions, and possibly a material business-plan change for the licence.

Acqui-hires need a customer plan too. In July 2025 Google's $2.4bn licence-and-hire deal took an AI coding start-up's CEO and some researchers to Google DeepMind; days later another company reportedly bought the rest, more than 350 enterprise customers included.

Keep independent

Choose it when the value is a market, team or trust you lack, the product is young, or the licensed entity is the asset. Keeping it separate also means fewer people leave.

What it costs: two of everything, and drift. Ericsson kept Vonage separate for a market that hadn't formed yet and wrote off about two thirds of the price.

Early signs it's wrong: no shared roadmap item after a year; restated milestones; impairment tests; shrinking headroom.

Independence tends to be a phase rather than an end state. I've seen that in four cases, which is enough for me to plan around but isn't proof. GitHub ran independently from 2018 to 2025, then joined Microsoft's CoreAI group without a new CEO and began moving to Azure. LinkedIn's CEO also took over Office in June 2025. Twilio ran Segment as a division, then a business unit, and from 2024 built its profiles into communications products. Okta ran Auth0 as a business unit, merged sales, then split its sellers again. So write down the trigger that ends independence (the founder leaving, capacity running out, a missed cross-sell milestone) and the end state.

Regulated platforms: this is the cheapest outcome. A stock deal leaves licences, registrations and certifications in place, though you still owe controller approvals and ownership notices.

Migrate to existing product infrastructure

Choose it when the products are similar, the thesis is cost or scale, and your platform can carry the load. Integration improved performance when the businesses were similar (Zollo and Singh's banks).

What it costs: the most engineering, a churn event every time a customer moves, and a long dual-run. In my model it's the only outcome that beats keep independent in steady state, and it pays back around year 7 on cost alone.

Early signs it's wrong: the parity list keeps growing and no customer has moved after six months; a contract or a dual-run bill, not readiness, sets the cut-over date (TSB).

Regulated platforms: customer-by-customer assignment or new contracts; new KYC, bank programs and BINs where entities differ; campaign moves and number ports; a PCI scope change; GDPR controller notices.

Move to shared platform services

Choose it when you need both products and one view of the customer: one login, one invoice, one data model for cross-sell. Twilio's Segment thesis needed exactly that, yet Segment ran as a division for about three years. After writing down $285.7M of its intangibles in late 2023, Twilio kept it under a new president.

What it costs: years. Bain puts full IT integration at two to three years; LinkedIn's Azure move was shelved; GitHub's is a 24-month plan. In my model, if the work takes 36 months instead of 18, this outcome trails keep independent for more than five years.

Early signs it's wrong: "pause" memos; platform work competing with the roadmap; the acquired team shipping less often.

Regulated platforms: outsourcing notices (UK PSRs regulation 25, the EBA guidelines and the DORA register all cover providers inside your own group), sub-processor notices to every customer, sponsor-bank approval of subcontractors, PCI, SOC 2 and ISO re-scoping, a FedRAMP transformative-change notice, and any CFIUS conditions on data location.

Who decides and how to record it

Here's how I'd split decision rights:

RoleOwns
Executive sponsorThe thesis; which businesses stay independent; kills with material revenue
Integration management officeOne plan, cadence, escalations; synergy and dis-synergy tracking with finance
Head of product (you)Recommends each outcome; customer roadmap; migration, sunset and deprecation plans
EngineeringArchitecture, cut-over method, dual-run budget, reliability gates
Legal, compliance, riskVeto on anything tied to a licence, registration, certification or contract term
Sales and customer successWho sells what to whom; account ownership; the renewal risk list
HRRetention list, leadership appointments, communication cadence

Why this shape: one study found integration managers create value and stop it leaking, and Bain argues that a decision that's 80% right beats a late one and that good acquirers focus on about 20 critical decisions.

Record each outcome as a decision record: context, outcome, owner, notice date, dual-run dates, decommission date, cost line, and kill criteria written in advance (for example: the replacement covers the jobs of accounts holding 90% of revenue, and no licence is tied only to the acquired entity).

What moves with the deal, and what needs approval

The entity holds the licence, so a stock deal moves almost nothing and an entity change moves almost everything. This is as of October 2026, and "estimate" marks my own inference. Every row varies by jurisdiction, deal structure and contract, so check yours with counsel.

Say you buy a business-texting provider. In a stock deal its 10DLC brands keep working, because its tax ID doesn't change; merge it into your entity and every brand has to be registered and vetted again.

US state money transmitter licence

Stock deal
Stays; control approval per state
Asset deal or entity merger
Doesn't transfer; new licences
Approval or notice
Prior approval in model-law states, NY, CA; control presumed at 10%
Lead time
~60 days per state once complete; 3-9 months nationwide (estimate)
For the four outcomes
Keep the entity; a kill or migrate plan can lose the fast track

UK or EU payment or e-money firm

Stock deal
Stays; the new controller needs approval (UK) or gives notice (EU)
Asset deal or entity merger
Not transferable (inference)
Approval or notice
UK: prior approval, and closing without it is a crime; EU: notice at 20, 30, 50%
Lead time
UK: up to 60 working days; EU: national
For the four outcomes
Shared services need outsourcing notices and a DORA register update

Canada RPAA registration

Stock deal
New application before control changes
Asset deal or entity merger
New application
Approval or notice
Prior
Lead time
Not found
For the four outcomes
Gates Canadian end users

Sponsor-bank program, network registrations

Stock deal
Contract decides; consent often needed
Asset deal or entity merger
Bank consent; new sponsor registers agents
Approval or notice
Contract, bank diligence, network rules
Lead time
Weeks to months (estimate)
For the four outcomes
BIN migration, re-underwriting; the bank may veto shared services

PCI DSS attestation

Stock deal
Stays for its scope
Asset deal or entity merger
New validation
Approval or notice
Card brand or acquirer decides
Lead time
Next assessment
For the four outcomes
Migrate or shared expands your card scope

10DLC brands and campaigns

Stock deal
Unchanged (same tax ID)
Asset deal or entity merger
New brand and vetting; customers' campaigns move between providers
Approval or notice
Registry and carrier review
Lead time
Minutes to 3 weeks
For the four outcomes
Throughput may reset; campaigns are the most portable layer

Numbers, short codes

Stock deal
Stay
Asset deal or entity merger
Port; lease transfer, carrier moves
Approval or notice
Industry process, authorization letters
Lead time
Days to weeks
For the four outcomes
Port-outs on a kill

FCC 214 authority

Stock deal
Domestic: day 31 if streamlined; international: prior approval
Asset deal or entity merger
Assignment approval
Approval or notice
FCC, plus security review if foreign
Lead time
31 days to months
For the four outcomes
Gating if the target is a carrier

SOC 2 report, ISO certificate

Stock deal
Stays; bridge letters ~3 months; tell the ISO body
Asset deal or entity merger
New scope or system description
Approval or notice
Auditor, certification body
Lead time
Next audit or observation period
For the four outcomes
Separate reports while independent

FedRAMP authorization

Stock deal
Stays
Asset deal or entity merger
Open question
Approval or notice
Transformative change: 30 business days ahead
Lead time
Months if reassessed
For the four outcomes
Infrastructure moves are slowest

Customer data (GDPR)

Stock deal
Controller unchanged; notice when your affiliates process
Asset deal or entity merger
New controller: notices, lawful basis
Approval or notice
Contract: typically 30 days' notice, then termination
Lead time
30 days or more
For the four outcomes
Every customer can object; pooling data changes the controller

Customer contracts

Stock deal
Change-of-control clauses only
Asset deal or entity merger
Assignment needs consent
Approval or notice
Contract
Lead time
Weeks to months
For the four outcomes
Consent campaigns before migrating

CFIUS, NSIA, EU FDI

Stock deal
Clearance before closing
Asset deal or entity merger
Same
Approval or notice
CFIUS mandatory filing 30+ days before completion; NSIA void if uncleared
Lead time
30 to 120+ days
For the four outcomes
Mitigation can forbid shared services

Some detail the table can't carry: under the money transmission model law, a change of control counts as approved 61 days after a complete application, or 30 days after a notice for a controller who's already approved and isn't changing the business plan materially. I found no guidance on whether an integration plan counts as such a change. Regulators approve people and money, not architecture, so the technology can usually consolidate behind a licensed entity under intragroup agreements, within outsourcing and safeguarding rules (the UK has required daily safeguarding reconciliations per firm since May 7, 2026). And many limits are just contracts: Block's agreement with Marqeta keeps its Cash App program served for a period if Marqeta changes control.

Take these to counsel, in writing:

  1. Which pre-closing product activities may we do jointly, only through a clean team, or not at all, and do our interim covenants give us consent rights over the target's decisions?
  2. State by state, is our change of control prior approval, notice or post-closing, and does our integration plan rule out the fast track?
  3. If the acquired payment firm moves onto our group platform, is that outsourcing, and what must we notify, and when?
  4. Does our sponsor bank need to consent to this change of control, or to moving processing or ledgers to our entities?
  5. If we pool both products' customer data in one service, do we become a new controller or sub-processor, and for which customers?

What each outcome costs

I'd price each outcome two ways, three-year net and steady state, and keep revenue synergies out of the cost case so they can't hide a weak one. For each component, add up the one-off costs (engineering, tooling, the customer program, incentives, legal work, re-registrations, decommissioning); the run cost of both stacks, then of steady state; revenue at risk (revenue by segment times the extra churn); and time, since every month of dual-run is a cost line.

A worked example follows, in $M. Every input is an assumption except the loaded engineer cost, which comes from the BLS median software developer wage (about $136,000 in May 2025), with wages at about 70% of private employers' compensation costs (December 2025): about $194,000, rounded to $200,000. The acquired product earns $40M a year, flat, from 1,250 customers (the top 50 hold 60%, 400 mid-size accounts 30%, 800 small ones 10%). Its stack costs $14.8M a year (infrastructure 3.0, 55 people 11.0, compliance and licences 0.8). Serving moved revenue on your platform costs 20% of that. Three years, undiscounted.

Keep independent (reference)

3-yr revenue
120.0
3-yr cost
44.9
3-yr net
75.1
Steady state a year
25.2
Overtakes keep independent
n/a

Kill: notice at month 6, off at month 18

3-yr revenue
95.4
3-yr cost
36.4
3-yr net
59.0
Steady state a year
18.9
Overtakes keep independent
Never, at these inputs

Migrate: waves in months 6-24, old stack off at month 30

3-yr revenue
114.2
3-yr cost
54.9
3-yr net
59.3
Steady state a year
28.9
Overtakes keep independent
Around year 7

Shared services: identity, billing, observability, notifications

3-yr revenue
118.8
3-yr cost
47.75
3-yr net
71.05
Steady state a year
26.5
Overtakes keep independent
Around year 6

Most of the result comes from the churn assumptions. Kill keeps 70% of top-account revenue, 50% of mid-size and 20% of the tail (23.6 of 40); migrate keeps 95%, 88% and 70% (36.2); shared services loses 2% to visible SSO and billing changes. Migrate's one-off of 20.9 covers 25 acquired engineers for two years, 10 of yours, tooling, a customer program and credits.

What I take from it, though these are still hunches:

  • On cost alone, keep independent wins over three years for a healthy $40M product; migrate only wins in steady state, and late. The case for migrate or shared services has to come from the thesis: 10% cross-sell on retained revenue from year 3 brings migrate's crossover to about year 5. That fits a survey of 232 horizontal deals, where market-related performance mattered more than cost savings.
  • What moves migrate most: top-account churn (each 10 points costs $2.4M of revenue a year) and how long the dual-run lasts (each extra six months costs about $4.9M).
  • Shared services is most sensitive to time: at 36 months instead of 18, the one-off roughly doubles and the savings arrive 18 months later.
  • Left out: roadmap opportunity cost, key people leaving, regulatory lead times, revenue synergies.

For scale: across 236 deals of $500M or more, the consultancy EY found integration cost 1% to 4% of deal value, with IT among the top three costs. I found no public figure for what a dual-run costs.

Examples

Figures are what the companies reported unless I say otherwise.

FIS and Worldpay: synergies beaten, then written down

FIS bought Worldpay in 2019 for about $43bn and beat its synergy targets by 2021. Meanwhile fintech rivals took share, and in 2023 FIS wrote down $17.6bn and later sold Worldpay off. The synergy scorecard measured the integration and said little about how the business was doing. Track share, renewals and release speed next to synergies (phase 2).

Ericsson and Vonage: kept separate for a market that didn't arrive

Ericsson paid $6.2bn for Vonage in 2022 to sell network APIs, and kept it independent while that market formed. It didn't form, and two impairments wrote off about two thirds of the price. If you keep a unit independent, set up front the milestones that trigger a kill, merger or sale (Keep independent).

Okta and Auth0: product kept, sales merged too early

Okta kept Auth0 as its own product but merged the two sales teams early. Sellers didn't know which product to lead with, attrition reportedly passed 20%, and Okta cut guidance in 2022. Merging sales is its own decision. Settle who sells what to whom first (phase 2).

Microsoft and LinkedIn: independent on purpose

Microsoft kept LinkedIn's brand, culture and CEO after the 2016 deal, and a planned move to Azure was reportedly shelved. Independence protected a network that runs on user trust. Even so, the leadership converged at the top in 2025 (Keep independent).

Microsoft and GitHub: seven years independent, then a move under pressure

GitHub ran on its own inside Microsoft from 2018, and only in 2025 started moving off its own data centers. In August 2026 a capacity failure took it down for almost eight hours, and its CTO said it had failed to scale in time. If you put off a platform decision, plan for when it comes due; at GitHub it came due as an incident (Move to shared platform services).

Microsoft and Skype: migrated onto Teams, then killed

Microsoft moved Skype's business customers to Teams from 2019, then closed consumer Skype in 2025 with under ten weeks' notice. It was survivable because users signed in to Teams Free with their Skype logins, contacts and chats. Build the clean path and set the sunset date together, or the tail runs for years (phase 3).

Broadcom and VMware: divest, bundle, force new terms

Broadcom bought VMware in 2023, sold the non-core units and moved customers onto one bundle with no perpetual licences. Revenue jumped, but some bills reportedly rose fivefold and a European cloud trade group complained to regulators. When switching is hard, forced change can pay now and cost you partners and regulators later (phase 4).

TSB and Sabadell: a big-bang weekend

Rising fees on its old owner's platform pushed TSB into one big cut-over. In April 2018 it moved about 5 million customers in a single weekend, having tested only one of its two data centers, an independent review found. It failed for months, and regulators fined TSB £48.65M. Budget the dual-run so readiness sets the date, and rehearse the rollback (phase 3).

What usually goes wrong

Thesis fails behind green synergy numbers

Early signs
Synergies met while share and win rates slip
Who pays
Shareholders; customers in a later split
Case
FIS and Worldpay

Sales collision

Early signs
Sellers unsure what to lead with; acquired pipeline drops
Who pays
Revenue for the next 2 to 4 quarters
Case
Okta and Auth0

Key people leave

Early signs
Top-layer and technical exits in year one
Who pays
The acquired roadmap
Case
Slack's CEO, about 18 months after close

Deferred platform work becomes an incident

Early signs
Shrinking capacity headroom
Who pays
Customers; trust
Case
GitHub, 2026

Big-bang cut-over set by a date

Early signs
Date fixed before readiness; rehearsals cut
Who pays
Customers, then fines and redress
Case
TSB, 2018

Forced change without a reason

Early signs
Short-notice price changes; partner complaints
Who pays
Long tail and partners first
Case
VMware

Independence without a value path

Early signs
No shared roadmap item after a year
Who pays
Shareholders; the acquired team
Case
Ericsson and Vonage

Integration kills what you bought

Early signs
Acquired release cadence drops after the reorganization
Who pays
Future revenue
Case
Puranam et al. 2006

Reconciliation gaps

Early signs
Breaks carried forward; manual adjustments growing
Who pays
End users, then regulators
Case
Synapse, 2024

Words that mean something else here

TermWhat you'd assumeWhat it means here
"70-90% fail"A research findingUncited at 90%; the 70% mixes share price, executives' opinions and revenue synergies
SynergyValue createdA planned number: FIS reported $750M run-rate before a $17.6bn write-down
IndependentSeparate foreverOwn brand and CEO, often shared infrastructure and HR; a phase
First 100 daysThe integration windowA convention; at best a decision deadline
MigratedDoneAccount created, traffic moved, old access revoked, data reconciled: four stages
DeprecatedIt stopsTo engineers "no new use", to customers "it stops": say end of sale and shutdown
Change of controlCrossing 50%10% under the money transmission model law; well below 50% for CFIUS; contracts set their own
ControllerOne meaningFCA: a shareholder with control. GDPR: whoever decides why and how data is processed

On the "70-90% fail" row: a 1999 KPMG study found most deals hadn't added shareholder value a year on, while most executives thought theirs had succeeded. McKinsey's 2004 study of 160 mergers found about 70% missed their revenue synergies, which gets misquoted as "70% fail".

Sources

Read on October 2, 2026 unless dated; "search result" means seen only as a search snippet or summary.

Academic research

Consultancy and survey figures

Books and practitioner writing (ideas only)

Regulators, law and standards

Law firms, auditors and advisers

Company documents, filings and reporting

Earlier research on this site

  • Telco rules come from the site's earlier telco research, and the UK safeguarding date from its cross-border payouts research

Cost inputs

Examples