If you are a CXO buying AI work for the first time, the market presents itself as a wall of similar-looking firms. Some call themselves consultancies. Some call themselves agencies. Many use both words interchangeably. The pitches all start the same way. The pricing is wildly different. The deliverables are nothing alike.
Buying the wrong type is one of the most expensive mistakes in this category. Consulting firms and agencies solve different problems. Knowing which one you need is half the procurement decision.
The honest side-by-side
Consulting firms produce decisions, plans, and systems. The deliverable is intellectual. You hire a consultant because you do not know what to do, or you know what to do but cannot do it yourself with the team you have. The deliverable lands as a strategy, a forecast model, a built system, a process redesign.
Agencies produce execution. The deliverable is operational. You hire an agency because you know what you want to do, but you do not have the operating capacity to do it. The deliverable lands as a campaign, a content stream, a managed service.
The decision is upstream of the procurement
The first question to ask yourself, before reading a single pitch, is whether you know what to do. If you do not, no agency will tell you, because that is not what agencies are built for. The agency will execute against the brief you give them. If the brief is wrong, the execution will be wrong.
If you do know what to do, the next question is whether the work is ongoing or one-time. Ongoing work, where the same kind of output gets produced every week or every month, is what agencies are built for. One-time builds, where the deliverable is shipped and then operated by your team, fit a consulting firm.
What this looks like in AI specifically
The AI variant of this distinction tracks the broader pattern, with some twists.
AI consulting is right for: picking the right AI use cases for your business. Designing the data pipeline and integration architecture. Building a custom forecasting model, a custom agent for your operations, a custom analytics layer for your team. Training your team to operate what we have built. None of these have an obvious off-the-shelf product, all of them benefit from senior judgement at the design stage.
AI agency is right for: producing AI-generated content on a weekly cadence. Running paid acquisition with AI-optimised creatives. Managing a customer chat AI day-to-day, with humans in the loop. Maintaining and tuning an AI agent that an agency-style team can operate at lower cost than an in-house equivalent.
You probably need both, in sequence. The consulting work designs and builds the system. The agency work runs it. Many failures we see come from trying to do the consulting work with an agency model (they do not have the seniority for design) or trying to do the agency work with a consulting model (they do not have the cost structure for ongoing execution).
The pricing reveals the model
One way to read the market is to look at how the firm prices.
Consulting pricing is usually time and materials, project-based, or fixed fee per defined scope. The conversation centres on the deliverable, the team composition, and the timeline. Senior partners are visible in the pricing because they do the work.
Agency pricing is usually a monthly retainer or a per-campaign fee, with the team mostly invisible behind a single account manager. The conversation centres on volume, channels, and frequency. Junior labour is the cost base.
If a firm calling itself a consultancy prices like an agency (large monthly retainer with hidden team), it is probably an agency in consulting clothing. If a firm calling itself an agency prices like a consultancy (per-project fixed fee, senior-led), it is probably a consultancy that chose the wrong word.
The hybrid that is real and the hybrid that is fake
The real hybrid is a consulting firm that retains operating responsibility for what it built, for a period. We designed your forecasting system, we operate it with you for six months, we hand over once your team is competent. The pricing is hybrid (project for the build, retainer for the operate phase), the team is hybrid (senior for design, mid for operate), and the work flows naturally between the two modes.
The fake hybrid is a firm that pitches both because they want both kinds of revenue, with the same team. The team that is excellent at execution is rarely excellent at design. The team that is excellent at design rarely has the patience or cost structure for operations. Beware firms that say they do both well.
What to do this week
For each AI project on your roadmap, answer four questions before you brief anyone:
- Do you know exactly what to do, including the workflow, the integration, and the metric? If no, you need a consultant first.
- Is this a one-time build or an ongoing operation? Build means consultant. Ongoing means agency.
- Who will own the deliverable once it ships? If the answer is "we will, internally", consultant. If "we want someone else to keep running it", agency.
- Are you paying by deliverable or by time? Deliverable means consultant. By time, especially monthly, means agency.
If the answers point both ways, you probably have two projects, not one. Separate them and brief them differently.
The transparency rule
Ask the firm directly. Are you a consultancy or an agency. Listen to the answer. The firms that know what they are will tell you in one sentence. The firms that hedge are usually the ones to avoid, because the lack of clarity in the answer reflects the lack of clarity in the offer.
If you want to test this on us, ask. We are a consultancy. We design and build AI systems. We do not run ongoing campaigns. If you need a campaign, we will point you to a good agency. If you need a system, talk to us.