Digital value is emergent, co-created, and dynamic — determined largely outside the offer document, through adoption, relationship, and evolution over the platform's life. That premise changes the selection itself.
Traditional evaluation answers one question — which compliant offer scores best on the parameters we can know today? Digital selection must answer another: which partner and platform maximize the value we will realize over a decade, much of which is not visible in any offer? A different question demands different criteria, different evidence, and a different process.
You are selecting three parties, not one bid
A large digital program typically involves the platform vendor, whose software, data model, and roadmap you will live with; the implementation partner, whose people will deliver the program; and often the cloud provider beneath both. Bundling them into one turnkey award — natural in the EPC world — carries a known risk: if the system integrator selects the software, you tend to get the product the integrator is best at delivering, not necessarily the one that fits your operation. Bundling has genuine advantages, single-point accountability above all, but for strategic platforms the balance of practice favors selecting the platform first, on its merits, then running a separate competitive selection for the partner.
People and methodology are a large part of the product
In equipment supply, a vendor's track record predicts well, because quality is a property of a stable industrial system — plant, tooling, QA — largely independent of which individuals were on shift. The fiftieth turbine resembles the first. That intuition misleads buyers in digital, where the production system is a specific group of people applying a methodology to your specific context. You are engaging a team, not only a logo.
Two consequences follow for vendor selection. First, key-expert loss — key personnel risk, in contract terms — is a first-order project risk, largely independent of vendor brand. Proposals are fronted by the vendor's strongest architects; whether those individuals staff your program, and stay on it, is a separate question no corporate pedigree answers. MillerCoors alleged precisely this — understaffing and departure from the vendor's own methodology — and Hertz alleged that when key people were replaced, project knowledge left with them. The risk both describe is real, and it is manageable by contract.
Second, the methodology as practiced matters more than the portfolio. How a vendor actually runs delivery — iterative releases of working software, real user involvement, disciplined testing, honest escalation, scope governance, benefits tracking — is the machinery that converts money into value. Every major vendor has an impressive methodology on slides; whether it survives the first schedule crunch depends on the team and the governance imposed, not the brand.
Credentials deserve a hard look in this light. SAP, Oracle, HCL, and Accenture — established firms with vast portfolios — sat on the vendor side of some of the most public technology-program failures on record: Lidl, Birmingham City Council, MillerCoors, Hertz. Large vendors do most large implementations, so they will appear in most large failures as a matter of arithmetic, and implementer experience remains a positive factor in the ERP-success research; credentials are not meaningless. But the cases show they are not sufficient predictors, because digital delivery success is substantially team-specific (A-teams are scarce and allocation opaque), context-specific (the reference project's key ingredient — that organization — is not part of your offer), and exposed to scale pathologies such as utilization pressure and change-order incentives.
The corollary cuts both ways. A small or young vendor can be a strong choice — closer attention from its best people, a more modern product, greater flexibility, a real stake in your success — provided viability risk is honestly priced: financial due diligence, roadmap credibility, and contractual protections such as source-code escrow and continuity provisions. Credentials are an entry ticket, not a prediction, in either direction. The prediction comes from demonstrated fit of the product, the specific people, and the methodology in action, which is what the process below is designed to test.
Selection starts a relationship; it does not close a process
Traditional procurement is designed to conclude: award, deliver, accept, close warranty, file. In digital, go-live typically falls near the midpoint of lifetime spend and marks the start of value creation. Adoption, optimization, upgrades, renewals, and eventual exit all lie beyond it — negotiated with the same vendor, by then holding incumbent leverage. Selection therefore assesses not only "can they deliver?" but "can we work with these people for a decade?" — and it is why post-go-live terms — service levels (SLAs), renewal pricing, exit assistance — must be negotiated before award, while leverage still exists.
How vendor evaluation criteria differ for digital platforms
In digital vendor selection — an ERP, an operations platform, a strategic SaaS — compliance becomes graded and tested: how well does the standard product fit your scenarios, and what adaptation does each side require? Price becomes one input, assessed as multi-year total cost of ownership (TCO) including integration, adoption, escalation, and exit.
The evidence changes with the question. Written bid responses give way to working evidence: scripted demonstrations on your data, a paid proof of concept, investigative reference checks. Vendor standing is read forward — financial statements still matter, but the questions are trajectory: R&D investment, roadmap, release cadence, analyst positioning (an input, never the answer), acquisition risk. And methodology, the supplier's internal affair in a traditional tender, becomes an evaluated criterion in its own right.
Criteria that barely register in traditional evaluation move toward the center. Ecosystem: the availability of skilled implementation partners and talent — a real constraint for remote mining operations — plus third-party integrations and user community. Openness: APIs, data export rights, adherence to standards — the primary insurance against vendor lock-in. Data migration: tooling, method, and vendor experience in moving your data, one of the most common proximate causes of ERP failure and a headline risk in Birmingham's reimplementation. Delivery capability, assessed at the level of the specific team: named personnel, continuity protections, the staffing plan. And the relationship itself: a decade-long working arrangement, which makes cultural and operating-model fit a scored criterion rather than a nicety.
Scale the process to your stakes — and your leverage
This process carries real cost, so apply it in proportion to the stakes: a strategic operational platform warrants the full funnel below; a departmental SaaS tool does not. Tiering purchases by strategic importance and switching cost — and reserving the heavyweight process for the top tier — is itself part of the discipline.
The process also assumes a buyer with leverage, and not every buyer has it. A mid-size operation buying from a mega-vendor will not obtain named-personnel clauses or a scripted pilot merely by asking. Where leverage is short, substitute rather than surrender. Demand less deviation from standard — the less you customize, the less leverage you need. Buy where you are a significant customer rather than a rounding error, which often argues for mid-market vendors or strong regional partners over the largest logo. Time commitments to the vendor's quarter and year end. And concentrate the leverage you do have on the two or three data and exit clauses that matter most. What a small buyer cannot negotiate, it must select around: openness and standard fit become the substitutes for contractual power.
For public bodies, procurement law constrains dialogue, pilots, and parallel negotiation — but the designed channels for exactly this kind of purchase exist: competitive dialogue and competitive procedures with negotiation under EU/UK regimes and their equivalents. They are slower and costlier than a standard tender, and no procedure substitutes for a capable buyer — but they were built for purchases that cannot be fully specified in advance. The discipline below transfers; the mechanics must run through the applicable framework.
Nine practices for the vendor selection process
1. Run a funnel, not a single-round tender
Replace the one-shot "RFP out, score, award" event with stages that eliminate on different evidence: market scan, long list, RFI, shortlist of three to four, scenario evaluation, paid pilot with one or two finalists, negotiation, award. Slower on paper; typically far cheaper than remediation after a paper-based award.
2. Script the demonstrations yourself, on your own data
Standard demos are rehearsed on polished sample data. Hand shortlisted vendors your scenarios — the month-end close, the shift handover with a breakdown on the primary crusher — and a sanitized extract of real data, including its real quality problems. Where vendors resist, that is itself information.
3. Pay for a proof of concept designed to test the hard parts
A six-to-twelve-week paid pilot is among the highest-return investments in the process: paying improves your claim on strong vendor staff, keeps findings yours, and signals a serious buyer. Build it around the hardest scenarios — messy integrations, real data quality, migration of a representative data slice — because a proof of concept that cannot fail proves nothing. A pilot surfaces problems; it does not resolve them — Lidl's valuation conflict was visible early, and what was missing was governance willing to act. Pair the pilot with a gate that has genuine authority to stop.
4. Reference-check like an investigator
Insist on references of comparable scale and complexity, and visit them. Ask what bid documents never answer: What broke? How did the vendor behave when it did? What did year-two renewal pricing look like? A vendor's conduct in its worst project is more informative than in its best.
5. Score openness and exit as first-class criteria
Before signing, know how you would leave: contractual data export and transition assistance, complete and documented APIs, data in accessible formats, and — where vendor viability is a concern — source-code escrow with tested release conditions. Evaluate the exit while leverage exists, which is to say before award.
6. Keep competitive tension to the end
Take two finalists through the pilot and into parallel negotiation where the applicable procurement framework allows; conclude terms before announcing. Once a vendor knows it has won, leverage on the clauses that matter falls sharply.
7. Weight the scoring for the lifecycle
An indicative skeleton: demonstrated solution fit around 30%; multi-year cost of ownership 20–25%; vendor viability, roadmap, and ecosystem 15–20%; delivery team and approach around 15%; openness, security, and terms around 15%. The exact weights matter less than the principle: the majority of the score should rest on evidence gathered through demonstration, pilot, and investigation rather than on the offer document alone.
8. Evaluate the methodology in action, not on slides
Use the pilot as an audit of working culture: do they show working software or status decks; how do they involve your users, log problems, respond to ugly data? Ask the proposed delivery leads about their last failure. A vendor that cannot discuss failure candidly in a sales cycle is unlikely to report it candidly in delivery.
9. Contract for the team, not just the deliverables
Name key experts in the contract; take approval rights over substitutions; require overlap and documented knowledge transfer on any replacement; attach commercial consequences to unagreed changes. Unusual by equipment-supply habits, where staffing is the supplier's business — but in digital delivery, staffing is a substantial part of the product.
For AI-inflected purchases, add a category of terms that barely existed five years ago: rights over your data in model training, model-change and performance-drift provisions, IP and output indemnities, and evaluation methods suited to probabilistic systems. The discipline is the same; the clause set is new.
What this demands of the organization
Every practice above requires judgments — architecture fit, data models, methodology as practiced, team quality, roadmap credibility — that no generalist tender team can make alone. Who holds which decision rights, how the award becomes a joint gate rather than a single signature, and how the organization must be built to buy well are operating-model questions in their own right.
- Lidl/SAP: Handelsblatt (2018) on Lidl's eLWIS program.
- Birmingham/Oracle: Grant Thornton, report on Birmingham City Council's Oracle implementation (February 2025).
- MillerCoors LLC v. HCL Technologies Ltd., U.S. District Court, N.D. Illinois (filed 2017; settled 2018).
- Hertz Global Holdings v. Accenture LLP, U.S. District Court, S.D.N.Y. (filed 2019; settled).
- Sergii Dovgalenko, The Technology Procurement Handbook (Kogan Page, 2020)
- David W. Tollen, The Tech Contracts Handbook, 3rd ed. (ABA) — the standard reference on cloud and software agreements, exit and data terms in particular
- Gartner, SaaS vendor-management and IT Sourcing, Procurement and Vendor Management (SPVM) research
Technology executive — enterprise AI, automation & digital strategy for enterprises, NGOs, and UN agencies. 15+ years, primarily across Africa's emerging markets.