LINOS LLC
SAP · ERP · AI

What we have learned delivering this work, written down.

No vendor summaries and no reprinted press releases. These are the arguments we find ourselves making on engagements, set out properly so you can disagree with them.

The 2027 deadline is a schedule problem, not a technology decision

SAP mainstream maintenance for ECC ends in 2027. By the time most boards ask for a conversion date, the honest answer is that no date is defensible yet, and that is the finding worth delivering.

Read it →
What it works through
  1. 01Why the date cannot come first
  2. 02The three things that actually set your timeline
  3. 03Brownfield, greenfield, selective: choosing on evidence
  4. 04What a defensible roadmap contains

The 2027 deadline is a schedule problem, not a technology decision

Sep 2, 2026 · 6 min
Contents
  1. 01Why the date cannot come first
  2. 02The three things that actually set your timeline
  3. 03Brownfield, greenfield, selective: choosing on evidence
  4. 04What a defensible roadmap contains

SAP mainstream maintenance for ECC ends in 2027. By the time most boards ask for a conversion date, the honest answer is that no date is defensible yet, and that is the finding worth delivering.

Every conversation about ECC and 2027 starts in the wrong place. A board asks when the conversion will happen, a date gets offered, and the program spends the next eighteen months discovering why that date was wrong. The date is an output, not an input.

Three things set a conversion timeline, and none of them are knowable without work. The first is custom code: two decades of modifications, usually with no current inventory. Pre-Check, ATC, and SCMON turn that into a sized remediation backlog, and until they have been run, any estimate is a guess. The second is data. Open items, master data quality, and the customer-vendor integration gate all sit on the critical path, and data problems surface late unless they are hunted early. The third is downtime tolerance, which is a business constraint rather than a technical one, and which drives the choice between conversion approaches more than any architectural preference.

Only once those three are known can brownfield, greenfield, and selective data transition be compared honestly. Brownfield carries the history and the technical debt. Greenfield buys a clean core at the cost of reimplementing what works. Selective transition sits between them and costs more to plan than either. The right answer depends on how much of the existing configuration is genuinely worth keeping, which is a question about your business, not about SAP.

A defensible roadmap contains a custom-code inventory with remediation sizing, a data quality assessment with a named owner per domain, a downtime target validated with the business, and a sequence tied to the funding calendar rather than to a fiscal year boundary. That package costs a fraction of the program and is the cheapest way to find out what the program actually is.

Most failed ERP programs are data programs in disguise

Aug 26, 2026 · 5 min
Contents
  1. 01The pattern
  2. 02Why data starts late
  3. 03Ownership before mapping
  4. 04What good looks like at each gate

The go-live slips, and the post-mortem blames scope or the integrator. Look at where the last three months actually went and it is almost always the load.

The pattern repeats across industries. Design lands, build goes broadly to plan, and then the schedule collapses in the last quarter. The stated cause is scope or testing. The actual cause, more often than not, is that the first realistic data load happened far too late and revealed quality problems nobody had budgeted to fix.

Data starts late for a structural reason: it is the one workstream with no natural owner. Finance owns the close, operations owns the process, IT owns the system, and nobody owns whether the material master is fit to migrate. So it waits until a load is technically possible, which is months after the problems could have been found.

The fix is unglamorous. Assign an owner per data domain before any mapping is written. Not a committee, a named person who can make a decision about whether a record is worth bringing. Profile the legacy estate in the first weeks, while there is still time for the answer to change the plan. Then run the load repeatedly, treating each rehearsal as a gate with a reconciliation rather than a technical exercise.

At design exit you should have a profiling report and a named steward per domain. At build exit, a repeatable load with a defect burn-down trend. At migration exit, balances reconciled within an agreed tolerance and signed by the person who owns the number. If those artifacts do not exist, the phase did not finish, whatever the plan says.

Clean core in practice: what it costs you, and where to break it

Aug 14, 2026 · 7 min
Contents
  1. 01What clean core actually promises
  2. 02The three places teams break it anyway
  3. 03Side-by-side is not free
  4. 04A decision rule worth using

Clean core is sound architectural advice that becomes unhelpful when treated as an absolute. The useful question is not whether to extend, but where the extension lives and what it costs at upgrade.

Clean core promises a cheaper future: fewer upgrade regressions, faster adoption of new releases, less regression testing. That promise is real. It is also a trade made against the present, because the processes that genuinely differentiate a business are frequently the ones standard does not cover.

Teams break clean core in three predictable places. Pricing and product configuration, where the logic is the commercial model and no standard covers it. Industry-specific compliance, where a regulation has a shape and the software does not. And integration-adjacent logic, where an interface needs enrichment that has to happen somewhere.

Moving that logic side-by-side onto BTP is the recommended answer, and it does work, but it is not free. You gain upgrade safety and lose transactional proximity: extra latency, a second runtime to operate, a second place errors can hide, and a team that now needs platform skills alongside functional ones. That cost is worth paying for genuine differentiation and is poor value for a small convenience.

The decision rule we use: if the logic would be a competitive loss to standardize, extend, and put it side-by-side with a named owner and a monitoring plan. If it exists because a process was never questioned, standardize it and absorb the change-management cost once. The expensive mistake is extending in-core for the second category, which is where most of the upgrade pain in older estates came from.

AI in a regulated ERP: an unexplainable decision is worse than a manual one

Jul 29, 2026 · 6 min
Contents
  1. 01The constraint nobody budgets for
  2. 02Human-in-the-loop as an architecture, not a disclaimer
  3. 03What an audit trail has to capture
  4. 04Where to start

In a GxP or DCAA environment, the constraint on automation is not accuracy. It is whether you can reconstruct, months later, why the system did what it did.

Accuracy is the number every AI pilot reports, and in a regulated environment it is not the binding constraint. The binding constraint is reconstruction: when an inspector or an auditor asks why a particular document was classified, a particular invoice posted, or a particular batch released, you need to answer with evidence rather than with a model version number.

That makes human-in-the-loop an architectural requirement rather than a line in a policy document. The checkpoint has to be a real decision point with a real decider, placed where reversal is still cheap, and recorded. A review step that everyone clicks through is not a control, and it will not read as one under scrutiny.

A usable audit trail captures the inputs as they were at decision time, the model and prompt version, the output with its confidence, the human who accepted or overrode it, and the timestamp. Inputs as they were matters most and is the part usually missed: if the source document has since been superseded, a trail pointing at the current version cannot reconstruct the decision.

Start where volume is highest and reversal is cheapest, which is usually document extraction feeding a queue a person already reviews. The automation earns its keep immediately, the control is already in the process, and you build the audit pattern on something low-stakes before applying it to anything that posts.

Your AMS contract is buying the wrong thing

Jul 15, 2026 · 5 min
Contents
  1. 01What response-time SLAs actually incentivize
  2. 02The recurring-incident test
  3. 03Buying capacity alongside coverage
  4. 04Terms worth insisting on

Most application management contracts are priced on ticket response. That pays a vendor to receive tickets, which is not the same as paying them to eliminate the causes.

Read a typical AMS contract and the commitments are almost entirely about responsiveness: acknowledge in fifteen minutes, respond in two hours, resolve a P2 in a day. Those are reasonable things to want. They also describe a service whose economics improve with ticket volume, and worsen when the underlying causes go away.

A quick diagnostic on any incumbent arrangement: pull the last two quarters of tickets and count how many are the same issue recurring. In estates we assess, a meaningful share of volume is a small number of unresolved root causes reopening under new ticket numbers. That is the cost of an incentive pointed at response rather than elimination.

The correction is to buy two things in the same contract. Coverage, priced on response and availability as usual. And capacity, a named monthly allocation for root-cause elimination and for the enhancements that never made project scope. Capacity is what converts a support relationship into an improving system, and it is the line most contracts omit.

Terms worth insisting on: a recurring-incident metric reported monthly with a downward target, root-cause documentation as a closure requirement for anything above a threshold, a fixed enhancement capacity that does not get eaten by incidents, and knowledge-transfer artifacts as a contractual deliverable rather than a courtesy. That last one is what keeps the relationship a choice.

SAP for mid-market manufacturers: choosing between GROW, RISE, and private cloud

Jun 30, 2026 · 8 min
Contents
  1. 01The three models, described honestly
  2. 02Where public cloud genuinely binds
  3. 03Sizing the extension question first
  4. 04A selection sequence that works

The deployment decision gets made on license economics and then constrains the process design for a decade. It deserves to be made the other way around.

Public cloud, private cloud, and on-premise are usually presented as a commercial choice with some technical footnotes. In practice the footnotes are the decision. Public cloud constrains what you can extend and how, and for a manufacturer with unusual costing, configuration, or quality requirements those constraints land directly on the process design.

Public cloud binds hardest in three places for mid-market manufacturers: custom product costing logic, shop-floor and automation integration that needs low-latency in-core behavior, and industry compliance requirements that standard does not model. None of those rule it out. All of them need to be checked before the commercial decision, not after.

So size the extension question first. List the processes that would be a competitive loss to standardize, and for each one determine whether standard covers it, whether a side-by-side extension covers it acceptably, or whether it needs in-core behavior. That list is short, and it decides the deployment model more reliably than any TCO model.

The sequence we recommend: baseline the genuinely differentiating processes, test each against standard in a system rather than in a slide, identify the ones needing extension and where that extension has to live, then take the deployment decision and only then negotiate commercially. Done in that order the license conversation is short, because the requirements are already settled.