Pure Source SolutionsEnterprise AI Counsel · ERP

Independent AI counsel for companies running ERP.

Every ERP vendor is now shipping AI into the stack, and the integrator being paid to deploy it is also the party telling you which parts are worth having. That is not advice. It is a quote.

We sit on your side of that table. We set the AI agenda for your enterprise systems, decide what is worth doing and in what order, and hold the vendors and integrators to what they promised — without selling you any of it.

We do not implement. We do not resell. We take no money from any software vendor or systems integrator, ever. The only way we get paid is by you.

The whole position, in four sentences. Everything below follows from it.

We run the model we sell

Most AI strategy advice comes from firms that have never had to make their own operation work this way.

Pure Source Solutions is run by an AI CEO. Don Zou is the human Chair, and he signs every deliverable that leaves the firm. Agents do the research, the analysis and the drafting; a person owns the judgement and the accountability.

That is not a gimmick, and it is not a pilot. It is the operating model from Don’s published work on AI-native organisation design, run on a real company with real money at stake so the claims can be checked rather than asserted. Decisions are logged as objects. Decision latency and the rate at which the human overrides the machine are tracked as operating metrics, because in an organisation where agents do the work, those are the numbers that tell you whether it is functioning.

We publish all of it — the decisions, the pipeline, the override rate, the revenue — twice a week, including the weeks the honest figure is zero. When we tell you how to run agents inside your own enterprise systems, you can go and read how we run ours, and what it cost us to learn.

Published openly

Decisions logged
every one
Decision latency
tracked
Human override rate
tracked
Pipeline and revenue
published

Updated twice a week, whatever it says.

Open the scoreboard

Engagements

One standing relationship, and a set of fixed-scope pieces that stand on their own or start one.

The core engagement

Enterprise AI Counsel

A monthly retainer. Senior judgement on what AI does and does not do inside your enterprise systems — without adding an executive to your payroll. Someone in the room when the vendor presents, and a written record of why each decision was made.

  • An AI roadmap for your ERP estate that names what is worth doing, what is not, and in what order — with the reasoning attached, not a maturity curve
  • Independent assessment of the AI features your vendor and integrator are selling into the program, tested against what the contract actually commits them to
  • A live decision register: every material decision, who made it, why, how long it took, and what was reversed
  • Data and integration readiness judged honestly — most AI disappointment in an ERP estate is a master-data and interface problem wearing a model’s clothes
  • Governance that survives contact with auditors: where agents may act, where a human signs, and how that is evidenced
  • A monthly session with your executive team, chaired by Don

$12,000 / month · six-month minimum, because nothing useful happens here in six weeks

Start a conversation
Chair-intensive · added 29 September 2026

AI Programme Assurance of the ERP estate

Six to eight weeks sitting beside a live programme, ending in the evidence a board needs to approve, defer or stop the AI spend. The AI Readiness Review below answers, once, what your estate can support and what you are being oversold. This engagement produces a defensible decision instead, with the Chair’s signature on the recommendation and forty to sixty hours of his own time in the work.

  • An inventory of every AI commitment in the estate — licensed, contracted, in a proposal, or assumed in a business case — with the contractual basis and annual cost of each, on one page
  • A readiness finding per commitment: the master-data, interface and process conditions it depends on, and whether those hold today
  • A written challenge of the vendor’s or integrator’s own business case, quoting their own document back
  • A live decision register for the programme — every open AI decision, its owner, its date, what it is waiting on — maintained for the engagement and handed over
  • A board-ready recommendation per commitment: proceed, defer, renegotiate, stop. Signed by the Chair
  • One 45-minute board or steering session chaired by the Chair at the end

$75,000 · fixed, six to eight weeks, board-facing, signed by the Chair

$40,000 for the first two engagements only, then list. The lower price is exchanged for three things, written into the engagement letter as obligations rather than hoped for afterwards: your name may be used publicly as a client of Pure Source Solutions; two quotable sentences attributed to a named executive, agreed within 30 days of delivery; and one reference call per quarter for twelve months, up to four calls, with prospects we introduce. If you will not agree to the first, the price is list.

The $20,000 AI Readiness Review is credited in full against the $75,000 if you upgrade within 60 days.

Start a conversation

AI Readiness Review of an ERP estate

$20,000 · three to four weeks

A first engagement that answers one question honestly: given your systems, your data and your programme as they actually are, what can AI do for you in the next twelve months, and what are you being oversold? Written, evidenced, and delivered before you commit budget.

SOW and Estimate Review

$15,000 · first two pilots $7,500

Ten business days, before you sign with the integrator. Red-flag register, negotiation memo, redlined assumptions list. The same skill applied to paper rather than strategy.

Program Assurance Desk

$9,000 / month · six-month minimum

Weekly independent reading of the programme, a maintained decision register, a one-page steering brief, and a monthly 45-minute session the Chair runs.

Go-Live Readiness Audit

$25,000 · three weeks

An independent readiness audit ending in a written go/no-go recommendation, signed by the Chair.

Data Migration Reconciliation Design Review

$12,000 · first pilot $6,000

An independent read of the integrator’s migration approach before mock load 1. The unglamorous work that decides whether anything downstream, AI included, has data worth pointing at.

S/4HANA Interface Readiness Assessment

$18,000 · first pilot

An interface gap register built from your own SAP extracts, with an S/4HANA disposition for every interface. Three weeks. SAP programmes only.

How we read a vendor’s paper

Judging an AI claim in a vendor proposal is the same discipline as judging a delivery claim: read what the document actually commits them to, quote it, and ignore the deck. Here is that discipline on all seventeen public ERP contracts in our library as of 2 October 2026 — every one a public record, linked to its source, in descending order of how many of our seven standing checks fired.

The aggregate read of all seventeen contracts — which checks fire, how often, and on whose paper — is the ERP contract benchmark. Every scored row behind it, document by document and check by check with the clause each finding rests on, is the open dataset.

Same standard in both: verbatim quotes, linked sources, and what we could not read named as what it is.

Utah State Board of Education — PowerSchool Group LLC

7 / 7 fired
BusinessPlus — financials, purchasing, HR/payroll|Publisher-led|19 months
  • C10
    The contract disclaims its own numbers: “Notwithstanding any other provision of these terms, the timeline and total hours contained in this SOW shall be regarded only as estimates.”
  • C1
    “Customer… is responsible for managing Customer resources, organizational change management, data conversions and interfaces from legacy systems, and user acceptance testing.”
  • C3
    The only quantified supplier effort in the whole document is a project manager at roughly 10 hours a week.

City of San Rafael, California — Tyler Technologies

5 / 7 fired
Tyler Munis SaaS|Publisher-led|$378,000 professional services, 1,632 quoted hours
  • C6
    “Implementation Hours included in this quote assume a work split effort of 70% Client and 30% Tyler.” That implies roughly 3,250 hours of city staff time against 1,392 vendor hours — with no city staffing plan anywhere in the document.
  • C7
    “Should portions of the scope… be altered by the Client, Tyler reserves the right to adjust prices for the remaining scope.”
Corrected 2026-09-30: testing (C3) is now recorded as not assessable, not fired — score down from 6/7. The retrieved excerpt never reaches a clause assigning test responsibility either way; the sentence we originally used for it is about data-conversion validation, covered under C1 above.

City of Orlando, Florida — Workday, Inc.

5 / 7 fired
Workday HCM Recruiting, 16 integrations|Publisher-led|$235,830.70 fixed fee
  • C1
    “Workday does not provide a service to extract data”, and “Customer is solely responsible for the extracting, transforming and loading of the data.” Conversion iterations are out of scope.
  • C2
    16 integrations are counted by complexity bucket, but the integrations master list is only developed later, during Architect & Configure. Anything above 16 is the customer’s.
  • C4
    “Change management, internal communications, testing, and training are outside the scope of this SOW, except to the extent expressly provided herein.” The trailing exception clause is why we re-checked it against the source: it means the exclusion is not absolute, and a buyer has to find what is expressly provided before assuming anything is covered.

City of Chandler, Arizona — Collaborative Solutions (Cognizant)

5 / 7 fired
Workday HCM, Payroll, Financials, Procurement|Integrator-led|Fixed fee, milestone invoicing
  • C2
    83 integrations listed — 58 integrator, 18 client, 3 placeholders — then: “integration estimates will be reviewed at completion of Architect & Configure stage. Adjustments may be required.”
  • C1
    “Client is responsible for extracting data from current systems”, plus all cleanup, validation and mapping. “Client owns catch up transactions.”

City of Rapid City, South Dakota — Tyler Technologies

5 / 7 fired
Tyler Munis, five phases through CAFR|Publisher-led|Contracted days, no hours by phase
  • C3
    Seven test types are named. Zero hours are budgeted against them. Stress and parallel testing are city tasks.
  • C2
    Four interfaces are listed by name, alongside a clause stating there are “no customizations or interfaces in scope”. The document contradicts itself.

Village of Downers Grove, Illinois — Tyler Technologies

4 / 7 fired
Tyler Munis SaaS|Publisher-led|1,768 implementation hours, $485,450 one-time professional services
  • C5
    Post-implementation support is 80 hours per functional area, “only billed if used” — a rate card, not a commitment, with no defined duration and no tie to the first month-end close.
  • C6
    The quote states that the services actually needed “depend[] on such factors as your level of involvement in the project and the speed of knowledge transfer”. The estimate can grow for reasons the agreement locates in the client’s own performance.
  • C4
    Hours are itemised by module — general ledger 152, payroll 208, utility billing 208 — but there is no change-management line item in the investment summary and no change-management deliverable in the sections of the agreement we could read.
Corrected 2026-10-02: our original entry counted testing (C3) as a fifth fired flag at HIGH severity, reading the client’s data-validation sentence as proof the client also owned test-script writing and execution outright. Re-examined under the ownership test, that sentence is the same boilerplate found in the San Rafael and Sutter Creek records elsewhere in this library, and it never reaches a dedicated testing-responsibility clause. C3 now reads MEDIUM and is not counted; the score reduces from 5 to 4. We state that here rather than quietly changing the number.
Still the cleanest protective example in the library on price: Tyler commits that “in the event that the services are not reasonably sufficient to deliver the work agreed to, through no fault of yours, we agree to timely complete our obligations under the Agreement at no additional cost to you”, backed by a 20% retention on each phase’s implementation and project-management invoices, released only at that phase’s go-live. Exhibit E, the statement of work itself, was not retrievable from the public packet, so the change-management finding rests on the investment summary and the agreement body.

City of Sutter Creek, California — Tyler Technologies

4 / 7 fired
Tyler ERP Pro SaaS|Publisher-led|$90,300 one-time services — 304 financials and 184 utility-billing hours at $145/hour
  • C7
    “the fees stated in the Investment Summary are good-faith estimates of the amount of time and materials required for your implementation. We will bill you the actual fees incurred” — no not-to-exceed cap and no completion-at-no-additional-cost clause anywhere in the agreement, and renewal SaaS fees are billed “at our then-current rates.”
  • C6
    “We will not be liable for failure to meet any deadlines and milestones when such failure is due to Force Majeure or to the failure by your personnel to provide such cooperation and assistance (either through action or omission).” Cancelling with less than four weeks’ notice makes the city liable for “daily fees associated with cancelled professional services” and non-refundable travel.
  • C5
    No defined post-go-live period. The stage list shows a “Transition to Client Services” entry with no readable duration; after that, support is the standard SaaS help desk rather than hypercare.
Corrected 2026-10-02, on two counts. Our original entry counted testing (C3) as a fifth fired flag at HIGH severity, and separately listed a sixth item labelled C10 using the sentence that is in fact this document’s fired C7 above — C10 is an informational check and has never counted in the standing score. C3’s only candidate sentence is the same data-conversion-verification boilerplate found in the San Rafael and Downers Grove records, and this document’s own statement-of-work exhibit never reaches a testing-responsibility clause at all, so C3 now reads not assessable rather than HIGH. The score reduces from 5 to 4, and the fired checks are now C4, C5, C6 and C7. A low score on that check means unread, not clean.
C4 fires as well: no change-management line item and no deliverable anywhere in the readable agreement or investment summary. Exhibit E was readable only to its table of contents. The reason to read a $90,300 contract for a city of roughly 2,500 people is that its terms are not small-city terms — they are close to word-for-word the language in the $3,506,945 Sunrise, Florida contract elsewhere in this library.

City of Portland, Bureau of Environmental Services — Infor (US), LLC

4 / 7 fired
Infor CloudSuite Public Sector, asset management|Publisher-led|Time and materials, 28 months
  • C4
    Change management is handed to a separate third-party vendor entirely; Infor’s own role is consulting hours “to support this process”, with no OCM deliverable of its own.
  • C7
    “Deviations from the assumptions may result in the need for revisions to the timeline, fees and/or scope”, and if the customer is not ready to go live, additional fees may be required.
Corrected 2026-09-30: testing (C3) reads MEDIUM, not the HIGH we originally counted — score down from 5/7. Infor writes the unit and system test scripts itself, but the customer writes the SIT and UAT scripts — a split rather than a clean assignment to either party.

Weber County, Utah — Graviton Consulting Services

4 / 7 fired
Oracle Fusion Cloud HCM and payroll|Integrator-led|$1,043,619 total
  • C5
    18 days of production support, 5 July to 29 July 2022, against a 1 July go-live. It ends before the first month-end close and before a full payroll month.
  • C8
    Final conversion, interface and report scope is set “at the conclusion of the Design stage” — after the contract is signed and the leverage is gone.
Worth reading as a counter-example: its test design is genuinely good — a 69-day test phase, two parallel payroll cycles, a traceability report. The gap is that the effort behind it is never quantified.

P. H. Glatfelter Company — Accenture LLP

4 / 7 fired
ERP programme (“Project IMPACT”), platform not confirmed in the exhibit|Integrator-led|28,902 total person-days across both parties, priced by phase, not converted to hours
  • C1
    The responsibility table gives P. H. Glatfelter Primary responsibility for “Data cleansing and conversion”, with Accenture only Secondary; no hours or fee are itemised to conversion inside Accenture’s 6,010-day ERP implementation total.
  • C4
    Change management appears only inside a general design objective — to “develop and implement the business processes, change management, and technology architecture designs” — with no standalone deliverable, hours or fee.
  • C5
    “Following the go-live, Accenture will continue to provide support to the P. H. Glatfelter team with the aim of handing over support completely to P. H. Glatfelter within a period of 3 months.” An aim, not a committed period, and not tied to a month-end close.
  • C6
    Glatfelter must select a third-party data-conversion provider by January 1, 2001 and have the on-site technical environment ready by June 1, 2001 — fixed dates that do not move with Accenture’s own performance.
Our first private-sector document, and the first where the integrator rather than the buyer owns real risk on two fronts: Accenture holds Primary responsibility for designing 66 named legacy interfaces, so C2 does not fire, and scope changes require “a writing executed by authorized representatives of both parties” before Accenture owes any work on them, so C7 does not fire either. Testing (C3) reads MEDIUM and is not counted: the parties agree only to “develop mutually agreeable test procedures”, with no Accenture hours stated. Worth noting separately — if Accenture receives no written notice of material nonconformance within 30 days of delivery, “all deliverables will be deemed accepted in full.”

NuVasive, Inc. — IBM Global Business Services

4 / 7 fired
SAP R/3, ECC 6.0 (IBM Express Solution for SAP Medical Devices)|Integrator-led|Fees, total hours and every per-role rate redacted under an SEC confidential-treatment request
  • C1
    “NuVasive will be responsible for extracting data from existing systems to support conversion to the new system”, and “IBM will not be responsible for harmonizing actual data or maintaining data standards.”
  • C2
    Two interfaces are named in the whole document, both folded into the same combined estimate of “up to 3,330 hours of effort” that also covers forms, reports and conversions, with no per-interface design or hours.
  • C4
    “Organizational Change Management is not included in the scope of IBM responsibilities. As SAP R/3 is implemented, NuVasive should anticipate and plan for organizational adjustments.”
  • C6
    NuVasive “will make business decisions within 48 hours”, “will not have any co-existing corporate initiatives that will impact this project”, and must itself procure, install and test all required hardware.
Four of the six checks we could assess fire here. The library’s first SAP document, and the first where a gap means genuinely redacted rather than clean: hypercare (C5) is not assessable because the SOW’s own “Go Live and Support” section is blacked out under the same confidential-treatment request that hides the fee table. Re-pricing (C7) does not fire — additional work above the estimated budget “must be mutually agreed” through a change request signed by both parties. Testing (C3) reads MEDIUM for a reason that appears nowhere else in this library: IBM’s own fee table names “Test script — Write/Execute”, “PQ” and “Testing Lead” as resourced roles, but every hour figure against them is redacted. Effort budgeted, not disclosed, which is not the same finding as effort never stated.

City of Sunrise, Florida — Tyler Technologies

3 / 7 fired
Tyler Munis and EnerGov, five phases|Publisher-led|$3,506,945 contract plus estimated travel
  • C3
    The statement of work names eight test types. The investment summary carries no testing hours and no testing fee.
  • C2
    Interfaces are priced as named line items, but “interface estimates… based on our understanding of the specifications you supplied”, with additional work by change order.
Exhibit G was truncated by the source at section D.4.3, so conversion roles, testing roles, assumptions and post-live sections were not readable. The score is partial and we say so.

Wright County, Minnesota — Ciber Global, LLC (HTC Global Services)

3 / 7 fired
Oracle Fusion Cloud — Financials, HCM, Payroll, Procurement|Integrator-led|$1,819,880 fixed price
  • C2
    Eight interfaces are named, but “The County has decided not to procure Integration cloud service…It will be the County’s responsibility to coordinate the extract/upload data from/to legacy systems.”
  • C5
    “Ciber will provide 30 Days Post Go-Live for HCM and ERP (remote).” ERP and HCM activate in different months, and 30 days reaches neither a month-end close for one nor two payroll cycles for the other.
  • C6
    A single-site assumption pushes any rollout to other offices onto the County, end users are assumed to have basic PC skills already, and Ciber disclaims responsibility for “the software or the software environment provided by Oracle” entirely.
Our best counter-example on three checks at once. Ciber states its own conversion hours — “Ciber has allocated up to 480 hours for this effort” — the first document in this library where an integrator quantifies its own conversion effort, so C1 does not fire. A named Change Management Plan, Communication Plan and Change Readiness Assessment Document are priced SOW deliverables, so C4 does not fire. And the change-order process defaults to deferring an unactioned County request rather than deeming it accepted, so C7 does not fire. Testing (C3) reads MEDIUM and is not counted: Ciber leads the test plan and initial scripts, but execution is joint with the County, with no Ciber-specific hours broken out of the fixed price.

Citizens Property Insurance Corporation, Florida — AST

2 / 7 fired
Oracle Cloud ERP|Integrator-led|$8,283,150 not-to-exceed services
  • C5
    Support is defined by calendar, not by close: 60 days after each phase go-live, with no commitment through the first month-end or year-end.
The strongest commercial terms in the set — payment per accepted deliverable, 90% on acceptance with 10% retainage. Exhibits A, B and C were not published, so scope, hours, conversion, integration and testing are unassessed. A low score here means unread, not clean.

Warnaco Inc. — SAP America, Inc.

2 / 7 fired
mySAP Business Suite (SAP NetWeaver)|Publisher-led|$1,555,400 net licence fee plus $264,418 annual maintenance; the services-fee exhibit could not be retrieved
  • C4
    “Warnaco shall retain ownership and responsibility for the timely delivery of the organizational change management function.” No SAP-owned change-management deliverable appears anywhere in the readable text.
  • C6
    “the fees and timeline set forth herein shall be subject to change if the Licensee Responsibilities are not performed in a timely and appropriate manner”, and scope is “frozen during the Business Blueprint phase.”
Two of the four checks we could assess fire. This is the library’s first contract where SAP itself delivers rather than an integrator, and a full signature block is present, dated December 1, 2004 — but four exhibits that would carry the numbers (Scope, Timeline, Fees, Change Order form) could not be retrieved after six separate targeted fetch attempts on this 550KB-plus filing. C1, C2, C5 and C7 are recorded as not assessable, not as clean. Testing (C3) reads MEDIUM: business users run acceptance testing “supported by SAP Consultants”, a guidance role rather than outright silence, but SAP states no hours either way.

University of Arkansas System — buyer’s model statement of work

2 / 7 fired
Workday Financial Management|Buyer-drafted template|No hours
  • C1
    Even when the buyer writes the template, the data work lands on the buyer: the system “expects to perform all data cleansing and manual conversion processes” and codes and unit-tests the legacy extraction programs itself.
Not a vendor contract, and excluded from our vendor comparisons. It is here because it shows what a buyer asks for before a vendor answers.

Home Interiors & Gifts, Inc. — PricewaterhouseCoopers LLP

1 / 7 fired
J.D. Edwards Financial, Distribution, Demand Planning|Integrator-led|$1,768,358 fee estimate, 9,878 hours, time and materials
  • C6
    “In calculating and approving the fee estimate in this Agreement, the parties have assumed that the Prior Work meets appropriate standards for the efficient performance of the PwC Services.” The Prior Work is a preceding phase run by another firm; if that assumption is wrong, the client absorbs the rework.
The most protective document in this library, and the one to hold up against the rest: PwC owns data conversion, owns Integrated Test Cycles 1 and 2 and the stress test itself, and sells change management as a named deliverable set — training courses and schedule, a “Go-Live Promotion and Communications Plan” and an “Ownership Transition Plan” — rather than as a role. Six of seven checks do not fire. We could not locate a second independent source to cross-check this document the way we did for others here; completeness rests on internal consistency rather than a second fetch, and we say so rather than presenting it as equally verified.

This section now shows all seventeen public ERP contracts in our library as of 2 October 2026 — sixteen written by the supplier, nine of those by the software publisher implementing its own product and seven by a third-party integrator, plus one buyer-drafted model contract kept as a named comparator and excluded from every supplier split — and, below, the three questions we have answered from them so far. The aggregate analysis is on the benchmark page and every scored row is in the open dataset. All of it lives at the same standard: verbatim quotes, linked sources, what could not be read named as what it is, and corrections stated plainly when we get something wrong — two of the entries above carry one made today.

What buyers actually ask, answered from the evidence

Three questions a CFO or CIO types when a vendor proposal is already on the desk, answered from the contracts above rather than from generic advice. The counts split the sixteen supplier-written contracts into nine vendor-direct — written by the software publisher that implements its own product — and seven integrator-led. The seventeenth document, the University of Arkansas System’s own model contract, is a buyer-drafted comparator and sits outside both splits.

How many hours should the integrator budget for testing in an ERP SOW?

Thirteen of the sixteen supplier-written contracts could be assessed on this question|Derived 2 October 2026

Expect no number. Budget for it yourself. None of the thirteen we could assess states the integrator’s test hours by stage. Three are not assessable at all: the San Rafael and Sutter Creek Tyler agreements never reach a clause assigning test-script writing or execution to either party, and the statement-of-work exhibit is missing from the Citizens Property Insurance contract with AST. Treat “effort not stated” as the norm rather than something you can negotiate away.

  • C3
    Workday, City of Orlando: “Customer will lead testing and sign off on the final configuration prior to the move to Production.”
  • C3
    Collaborative Solutions (Cognizant), City of Chandler: “Client will be solely responsible for testing, which shall include configuration, business processes, data, reports and integrations.”
The one exception is also the most revealing: IBM’s SOW for NuVasive names real test roles in its own fee table — “Test script — Write/Execute”, “PQ”, “Testing Lead” — but every hour figure against them is redacted under an SEC confidential-treatment request. Budgeted, not disclosed, which is a different finding from silence. What to ask for: integrator test hours by stage stated in the SOW itself, or a fixed test-effort fee instead.

Is my integrator’s quoted fee actually fixed, or just an estimate billed on time and materials?

Three of the nine vendor-direct and three of the seven integrator-led contracts make the quoted fee contractually an estimate|Derived 30 September 2026

Read the word “estimate” literally. An estimate-only fee turns up on both paper types in roughly equal measure, so it is common practice rather than one vendor’s habit. Two of the three vendor-direct instances are Tyler contracts and one is PowerSchool’s; the three integrator-led ones are Accenture’s, PwC’s and IBM’s.

  • C10
    PowerSchool, Utah statewide contract: “Notwithstanding any other provision of these terms, the timeline and total hours contained in this SOW shall be regarded only as estimates.”
  • C10
    IBM, NuVasive SOW: “IBM will invoice NuVasive monthly for actual Services hours worked (whether above or below the estimated hours).”
What genuine protection looks like, from the same publisher that gave two other buyers an estimate: Tyler’s contract with the Village of Downers Grove reads “we agree to timely complete our obligations under the Agreement at no additional cost to you”, backed by a 20% retention released at each phase’s go-live. Ask for that clause, not just a number — a cap with no completion clause behind it is a number, not a protection.

What should be in a priced change-management (OCM) line item?

Eight of the nine vendor-direct contracts leave it unpriced, and two of the six assessable integrator-led ones|Derived 29 September 2026

On vendor-direct paper, budget it yourself. On integrator-led paper, a majority already price it, so press on the minority that do not. The shape that recurs in the documents that price it well is the same: a stakeholder or impact assessment, a named communications plan, a change-readiness assessment, and training tied to go-live rather than sold as an add-on, invoiced by milestone.

  • C4
    SAP America, Warnaco: “Warnaco shall retain ownership and responsibility for the timely delivery of the organizational change management function.”
  • C4
    Ciber Global, Wright County, as the counter-example: a named Change Management Plan, Communication Plan and Change Readiness Assessment Document, all priced SOW deliverables.
The one vendor-direct exception worth naming: Tyler’s contract with the City of Sunrise prices change management as a genuine deliverable set — a plan, an analysis document and presentation, strategy tools and a coach session for up to 25 people — invoiced by milestone. One of nine. Ask your vendor to match it; do not expect it offered.

Start with a document

The cheapest way to find out whether we are any good is to give us something to read.

SOW Red-Flag Report

Send us a vendor proposal or statement of work. Within one business day you get a plain-language note on the three things most likely to cost you money, a register of every red flag with the exact sentence it rests on, and the wording to ask for.

$750

Automated pass

Not reviewed by a human. It says so on every page.

Automated pass
$1,500

Chair-reviewed

The same register and note, read and signed by Don Zou, with his written judgement on what matters most and what to do about it.

Chair-reviewed
  1. Pay through the link. Either price is credited in full against the full SOW and Estimate Review if you upgrade within 30 days.

  2. Email the document as a PDF, with your receipt number in the subject line.

    reports@psi888.com
  3. Delivered within one business day. Documents received after 11:00 UTC are delivered the next business day. If your document cannot be read in full, we do not score it: we tell you the same business day what could not be read and ask for a complete copy.

For an Enterprise AI Counsel retainer or an AI Readiness Review, write to Don directly:

donzou@psi888.com

Who signs it

Don Zou has spent 25 years inside large ERP programmes, almost always on the buyer’s side: five years at a global systems integrator designing SAP order-to-cash solutions for industrial clients, fifteen years embedded with a global consumer-goods company running data migration, integration testing and supply-chain process design, and the last five on a US consumer-products S/4HANA programme.

He founded Pure Source Solutions in 2006. He writes three newsletters on enterprise AI — The AI Native Transformation, Enterprise Agentic AI and AI For SAP — including the RFC series on AI-native organisation design that this firm is built on and run by.

Being current on AI is part of this job, and it is exactly the kind of claim a buyer should refuse to take on trust. So ours is built to be checked: three newsletters published under his own name, the RFC series this firm’s operating model came out of, and a company actually run that way with its numbers published twice a week — rather than a practice that describes agents in a deck.

That combination is the whole proposition: someone who knows where enterprise systems actually break, who has been working out how agent-run organisations should be governed, and who is running one where the mistakes are visible.

What we will not do. We do not implement, resell, or take referral fees, and we take no money from any software vendor or systems integrator, ever. We do not bid on work we have advised you to buy. Anything you send us is not used to train anything, and is deleted 30 days after delivery unless you ask us to keep it.

Who we work with

  • Companies of roughly 200–3,000 employees
  • SAP S/4HANA, Oracle Fusion, Dynamics 365, NetSuite, Infor, Epicor
  • CFOs, CIOs and boards being asked what the AI plan is
  • Programmes about to start, in flight, or in the year after go-live

The open record

This firm publishes its own numbers — decisions, override rate, pipeline, revenue — twice a week, including the months where the honest figure is zero.

Public scoreboard