Buy the missing decision or capability
AI Consulting vs AI Agency
Compare advisory and agency proposals by the work they accept responsibility for, the evidence they deliver and the operational change you need.
“AI consultant” and “AI agency” are descriptions of a provider, not sufficiently precise descriptions of a contract. An agency may offer strong discovery and strategic advice. A consultancy may design, build and support working solutions. The buying decision should identify the missing work and ask which proposed team will accept responsibility for it.
If the business has not settled which problem to solve, advisory work should compare options and establish the decision boundary. If the business has a well-supported brief, a delivery team should demonstrate how it will produce, test and support the required result. A combined engagement can cover both, with distinct acceptance points.
Translate the proposal into actual obligations
| Proposal phrase | Ask what it means | Evidence to require |
|---|---|---|
| AI strategy workshop | Will the team challenge the problem and compare alternatives, or configure a chosen offering? | Decision outputs, assumptions and the treatment of rejected options |
| AI automation build | Which workflow, systems, users and actions are included? | Boundary, acceptance cases and excluded responsibilities |
| End-to-end service | Where does the service begin and end, including support? | Named owners at every handoff |
| Business outcome | What does the provider control, and what remains with the buyer? | Operational dependencies and a measurable contribution |
| Ongoing optimisation | Which changes, evidence reviews and response arrangements are included? | Defined service scope and a change process |
These questions are useful because polished proposals can hide different assumptions. One may include discovery and exclude production integration; another may include an attractive interface but assume the buyer has already resolved authority and data access. Compare the complete commitment before comparing price. The UK government’s AI procurement guidance recommends describing the problem and remaining open to alternatives; here that principle is applied as a buying method, not a Saudi procurement rule.
An events venue: is the missing work advice or delivery?
Consider an illustrative events venue that wants an AI enquiry assistant. A proposed agency package includes a conversational experience, lead capture and connection to a customer-management system. Those capabilities may be relevant. However, the venue has not agreed who can promise availability, how provisional bookings are represented or which packages can be quoted without manager review.
The first missing output is an operating decision: what may the assistant say and do? Advisory work should trace enquiry handling, identify authoritative availability information and compare an enquiry form, staff assistance and a customer-facing agent. It might recommend a narrower first release that collects requirements without confirming a booking.
If that brief is accepted, the delivery work becomes concrete: implement the chosen interaction, verify the information shown, test the handoff to staff and define support. The same provider may be qualified for both stages. Alternatively, the venue can take the evidence and brief to another team. The choice should depend on capability and commercial fit rather than the provider’s preferred label.
Use different acceptance records for advice and a working service
Accept advice when it enables the agreed decision and its reasoning is inspectable. Accept delivery when the agreed capability works under the relevant conditions. Confusing those tests can lead to paying for an attractive demonstration when the business requested an options analysis, or accepting a slide deck when it expected a supported service.
| Advisory acceptance | Delivery acceptance |
|---|---|
| The problem and affected users are evidenced | The specified users can complete the intended workflow |
| Credible options and exclusions are explained | Included functions and integrations meet agreed conditions |
| Authority, assumptions and unresolved questions are explicit | Approval, escalation and failure cases behave as agreed |
| The next investment decision has an owner and evidence requirement | Operating ownership, support and handover are accepted |
In the venue example, the advisory record should explain why the first version cannot confirm a booking. The delivery record should show that the deployed interaction respects that boundary and gives staff the information needed to respond. A disclaimer buried in a contract does not substitute for behaviour that follows the agreed scope.
A proof of concept can test an uncertain capability before production commitment. Specify what the test demonstrates and what remains unproven. The broader consulting-versus-implementation guide explains how advisory conclusions become delivery inputs.
Inspect incentives without assuming bad intent
A team that earns more from a subsequent build has a commercial interest in the recommendation. That does not make its advice wrong, but the buyer should be able to inspect alternatives and stop after discovery. Ask whether fees are contingent on later work, whether a preferred platform relationship influences the proposal and whether another team can use the outputs.
Publisher disclosure: Open for AI provides consulting and also develops solutions. Its recommendations should therefore be assessed with the same questions. Agree the advisory deliverables and any later delivery decision separately. Do not interpret “consulting” as a guarantee of vendor neutrality or assume an agency cannot challenge an unnecessary build.
When the problem is already well understood, buying another broad discovery exercise may add little. Ask for a targeted review of the remaining uncertainties. When the problem is unclear, accepting a bundled build price can prematurely fix the approach. The engagement-model guide helps structure those different needs.
Check who will do the work after the sales meeting
Meet the proposed people who will interpret the workflow, make technical decisions and manage delivery. Ask which responsibilities rely on subcontractors and how their work is reviewed. Request examples of relevant deliverables with confidential information removed. Evidence of one good interface does not establish competence in every operating environment.
For a Saudi organisation, verify the actual Arabic and English work required, the availability of the process owner and the expected collaboration arrangements. Do not infer local presence, sector expertise or delivery capacity from branding. The comparison with a freelance consultant explores a narrower individual engagement; systems integrators may be relevant when enterprise integration dominates the work.
Choose the handoff you can manage
Using separate advisors and a delivery agency can provide an additional challenge to the brief, but it also creates a handoff to manage. Name the buyer who resolves conflicting interpretations. Keep the evidence, scope and acceptance conditions available to both teams. A combined provider reduces some coordination but still needs transparent stage decisions.
If the business expects continual product learning, compare either external option with an in-house team. The appropriate choice is the arrangement that covers the work, preserves necessary knowledge and gives the buyer a usable result.
Through Open for AI’s agentic AI consulting, you can clarify a proposed workflow and the boundary between advice and delivery. Discuss an agency or consulting proposal with a redacted scope and the decisions it leaves unresolved. Agree which questions the engagement should answer before selecting the next service.
FAQ
Your questions, answered.
Does an AI agency only build software?
Do not assume that from the label. Ask whether the proposed team can challenge the brief, compare alternatives and explain the operating consequences as well as deliver software. Equally, a consultancy may include technical delivery. Evaluate the named people, scope and acceptance evidence. The distinction that matters is which work has been contracted and which decisions remain unresolved.
What should we ask when a proposal says “end to end”?
Ask the provider to identify the beginning, end and exclusions of the service. Clarify discovery, data preparation, integrations, testing, staff adoption, support and change handling. Name the party responsible at each handoff and distinguish included work from separately priced options. The phrase is useful only when it corresponds to obligations your organisation can verify and accept.
Should we pay for discovery before an agency build?
Pay for discovery when it resolves material uncertainty that affects scope, value or operating responsibility. If the problem and evidence are already clear, a focused review may be sufficient. Agree what decision the discovery will enable and whether you can use its outputs with another provider. Avoid buying an open-ended workshop series with no defined decision or acceptance criteria.
How do we assess an AI agency’s capacity when it serves several clients?
Ask which named people are committed to your engagement, their allocation, who can replace them and whether subcontractors perform material work. Check how competing client deadlines affect the promised milestones and who approves team changes. A polished agency portfolio does not establish the availability of the proposed team. Make staffing assumptions and substitution arrangements visible in the agreement.
Is a working demo enough to accept an AI agency project?
Only if the contracted deliverable is specifically a demonstration with stated limits. A production service needs evidence for the agreed users, workflow, exceptions, integrations and operating handover. A demo can show that an interaction is possible without proving that it behaves correctly with real variation or that somebody will support it after launch. Match acceptance to the commitment you purchased.
Who resolves disagreements between a consultant and a delivery agency?
Name an accountable buyer-side owner who can resolve scope and outcome questions, supported by the relevant specialists. Require both providers to use the same evidence and acceptance conditions. Distinguish a new requirement from a misunderstood brief and document the decision. Without that ownership, two competent providers can leave the organisation managing an unresolved boundary between their contracts.