The Vendor Shortlist Criteria That Filter Out Platform Sellers Before the First Discovery Call

Seven screening tests that expose platform sellers before you book a call.

· 8 min read
Cover illustration for “The Vendor Shortlist Criteria That Filter Out Platform Sellers Before the First Discovery Call”
AI-First Process Redesign · November 19, 2026 · 8 min read · 1,799 words

AI operations transformation for asset-heavy industries can start with one broken process, without buying a whole platform or first committing to a digital roadmap. The shortlist should reward evidence of rebuilt workflows, sector-specific operating knowledge, and live production deployments. Before booking a discovery call, require vendors to show what failed, what they changed, and what the operations team can now do.

Why operations leaders in asset-heavy industries are getting burned by platform sellers before the first meeting

Platform sellers can answer surface-level questions fluently, while operations leaders need something more demanding than a convincing account of available technology. In construction, logistics, manufacturing, and retail, the requirement is often to rebuild a painful process whose failures span people, data, and decisions. A broad demo does not show that a vendor can do this work. Yet shortlisting on modules, integrations, and supported use cases gives the strongest early position to vendors whose sales process is built around breadth.

Agentic AI projects can fail before they do useful work because orchestration design, task scoping, and failure-recovery logic were never worked out, even when the underlying models were strong. These are deployment questions. A vendor can present capable software without explaining how work will move through it when an input is missing or a task cannot be completed.

The Suffolk–MIT white paper on construction and AI identifies distinct opportunities across design automation, scheduling, supply chain and procurement, offsite manufacturing, permitting, and skilled labor. That range shows why a buyer must name the process that needs to change before judging a proposed solution. Broad case studies and proposals written to get through procurement can blur that distinction before operations has looked at the work.

What “platform seller” actually means and why the category is hard to spot

A platform seller is defined here by where ownership of the broken workflow ends. Enterprise AI includes platform vendors selling AI features inside licensed software and model providers selling foundation model access. Consultancies and systems integrators advise and implement, while product engineering or operations firms design and rebuild custom workflows on client data. Those roles serve different purposes. The shortlist should identify which vendor will own the specific process that needs to change.

The category becomes harder to spot when a platform seller claims to perform all of those roles. Broad reference lists, rehearsed procurement answers, and enterprise-grade security documentation can make that claim sound complete. Readiness can reflect experience selling across many verticals.

The strongest distinction is therefore a process-level before-and-after. A workflow owner can describe what broke, what the redesigned process looks like, and what the operations team can now do. Configuration alone does not establish that ownership. A feature set describes the product’s capabilities, whereas a rebuilt workflow describes a changed way of working.

How to read a vendor’s website and proposal for process depth before any contact

Start with the verbs in public case studies and any proposal already available. “Improved project delivery” names an outcome without explaining the intervention, while “reduced RFI cycle time by rebuilding the triage and routing workflow” identifies the process that changed. The second claim is more useful for screening.

Does it explain what the team does differently, or only what the software does? The customer also matters when judging the case, because work in facilities management alone does not show how well a vendor knows construction as a whole. A named customer should match the sector and operational scale of the process being assessed.

Service language provides another signal. Platform-led materials emphasize modules, integrations, model options, and supported use cases, while workflow-led materials identify where coordination breaks, data quality fails, or decisions depend on opinion. AI offered as a “roadmap item” should fail the production-evidence filter. Team profiles can then show whether anyone has held actual operations, procurement, scheduling, or field roles in the sector.

What sector specificity looks like in practice — and what passes for it

It is easy to copy a sector’s terms into sales materials. The test must go deeper than those words to find out what limits the work. In manufacturing, the question is not simply whether a vendor offers AI, but whether its system fits the plant, equipment, people, and decisions it supports. Document summaries and spreadsheet analysis do not establish plant-floor depth. A substantive answer describes the machine behaviors, maintenance histories, quality thresholds, and supplier constraints around which the system was built.

In construction, buyers also need to draw a clear line between scheduling and scheduling data. A vendor claiming scheduling expertise should explain whether its system has been tested on project schedules with the same activity count and complexity as the buyer’s. A small fit-out is not a megaproject. The white paper’s separate value levers also provide a useful boundary, because competence in one does not establish competence across all of “construction AI.”

In logistics and retail, the same test needs to cover a specific coordination or procurement workflow as well as any ERP integration. Without real-time tracking of material flow and supplier lead times, teams keep reacting to events, and adding AI to fragmented data does not fix the failure to coordinate. Customer logos do not answer that objection. Before a meeting, ask what process was rebuilt for the customer, what it looked like before, and whether the operations leader who ran it can discuss the change.

Why process audit evidence separates workflow owners from platform sellers faster than any reference call

A process audit exposes whether the vendor has investigated how work actually fails. “The data was messy” is too general to demonstrate that investigation. A useful account explains, for example, that RFI logs without due dates and owners cannot be aged and chased or linked to the programme activity they block. That is a concrete operational failure. It explains why assessments of delay risk depend on opinion.

The account should then connect the finding to a redesign. What did the team do, what did they do it in, where did it stall, and what can they do now? These questions distinguish a tool problem from a workflow problem. In failed construction AI deployments, the software can work even as the process around it fails, so teams log information but fail to coordinate work.

Before agreeing to a discovery call, ask the vendor to describe a process audit for a client in the same sector, including its findings, changes, and resulting team capabilities. An onboarding sequence is not an adequate substitute. A platform seller may explain configuration, while a workflow owner should explain who was observed, what broke, and what the redesign required.

How do you score vendors to fix one operations process without buying a platform?

Score each criterion 0, 1, or 2. A 0 means absent or evasive, a 1 means partial or promised, and a 2 means demonstrated with evidence. Confidence earns no extra credit. Every 2 needs material to back it up, so even a smooth answer scores 1 at best until the vendor shows evidence.

The minimum evidence package should include a target architecture and dependency map for the specific use case. It should also describe data flow and retention, including where prompts and completions are logged and who can access them. Security and AI-governance controls belong in that package. Clear-box transparency means those questions can be answered directly, rather than deferred because the vendor has not resolved them internally.

Implementation assumptions must identify who will do the work and what operational background those people bring. The package also needs the support model, full pricing and contract structure, and references or case evidence matching the sector, process type, and operational scale. Specificity should receive the greater weight. Sector knowledge, process audit evidence, and live deployments in the relevant use case class should count more than platform completeness.

Score against the workflow the business actually runs. A platform spreads its value across several use cases, while a specialist typically delivers a single use case better and faster. Neither model is better in every setting. But when the immediate goal is one painful process, platform breadth can become a cost paid for flexibility that the buyer will not use.

What the in-session test reveals that no pre-call filter can — and when to use it

A vendor that passes the evidence filter has earned a discovery call. The first session should include at least one live test outside the scripted demonstration. Ask the vendor to build a cross-workflow agent sequence with inputs from the actual process in place of a demo dataset. The response exposes how it handles specificity. A workflow owner will engage with the process, while a platform seller may redirect the discussion to a prepared path.

Ask who owns the LLM, whether the client can bring its own, where prompts and completions are logged, and who has access. Then ask what happens when the agent encounters a step it cannot complete. General assurances are not enough for this test. The answer needs to describe failure-recovery logic, notification responsibility, and how the system avoids proceeding when uncertain.

The people in the room provide another signal. A sales engineer who can build during the meeting demonstrates something different from a solutions consultant presenting slides, but the buyer must also ask whether that person will participate in implementation. A vendor that passes these tests can engage with what breaks and be held to a specific process outcome.

What the shortlist looks like when the criteria are applied — and what gets filtered out

The resulting shortlist is likely to be short. Vendors drop out when their case studies give results without saying how the process changed, when their claims to know the sector rely on integration partners instead of their own experience doing the work, or when production AI is still a promise for the future. Deferring governance answers to a later “technical call” creates another gap in the evidence. These exclusions follow from missing proof.

The remaining candidates can describe a rebuilt process and show sector experience within the team that will do the work. They begin with a process audit before recommending technology because automating a broken process can make it worse faster. Delivery timing becomes a further distinction. A commitment to a live first process within weeks is consistent with rebuilding from operational knowledge.

Procurement may still ask for three candidates when only two meet the criteria. The criteria should prevent padding the shortlist with unsupported claims of fit. Two qualified vendors and one platform seller for comparison are more useful than five platform sellers and no workflow owners. The same evidence standard also makes the difference visible before evaluation consumes six months and ends in a pilot that never reaches production.

Once the scripted demo is set aside, can this vendor show a rebuilt process, or only a configured product?

More in AI-First Process Redesign