An AI software vendor sells you access to a capability.
An AI consulting firm builds the infrastructure that makes the capability work for your company specifically.
These are different products. Confusing them produces two expensive mistakes: paying consulting firm rates for software that already exists, and paying software vendor prices for a tool that produces generic outputs without the foundation a consulting firm would have built.
The right answer for most $5M–$25M companies is both, in a specific order.
This article defines what each type of provider delivers, where each is appropriate, and what the combination looks like for a mid-market non-tech company.
What an AI software vendor actually delivers: the honest picture
The three types of AI software vendor
Type 1: Foundation model platforms (Claude, ChatGPT, Gemini)
Anthropic, OpenAI, and Google sell access to their AI models, via subscription (Claude Pro, ChatGPT Plus) or API.
What they deliver: a capable, general-purpose AI that can assist with writing, analysis, coding, and research.
What they are responsible for: the model’s capability, reliability, and safety behaviour. They are not responsible for producing company-specific outputs. The model is calibrated to the general population of users, not to any specific company’s voice, clients, or operating standards.
What is missing: everything that makes the model produce company-specific outputs, the context pack, the workflow documentation, the client archetypes, the decision rules. These must be built by the company or by a consulting firm on the company’s behalf.
Type 2: AI workflow and automation platforms (Make, Zapier AI, n8n)
Automation platforms sell workflow orchestration capability, the ability to connect tools, build triggers, route data, and include AI processing steps.
What they deliver: a no-code or low-code environment for building automated workflows that include AI.
What they are responsible for: the platform working reliably and the integrations functioning. They are not responsible for the quality of the AI outputs those workflows produce. That quality depends on the prompt architecture and context loaded, which the platform does not build.
Type 3: Purpose-built AI software (HubSpot AI, AI-powered accounting tools, AI customer service platforms)
These vendors sell AI capability embedded in a specific function or tool.
What they deliver: AI features for a specific use case, pre-integrated with the tool’s data and interface.
What they are responsible for: the AI feature working within the tool’s scope. They are not responsible for producing outputs that match the company’s voice, reflect its client archetypes, or apply its decision rules. Most purpose-built AI features produce reasonable generic outputs. Few produce outputs that feel written by someone who knows the company.
The software vendor’s delivery boundary
Every AI software vendor’s delivery ends at the same place: “the tool is available and works as described.”
The outputs the tool produces for any specific company depend on what the company loads into the tool. What the company loads into the tool is the consulting firm’s work.
What an AI consulting firm actually delivers: and the variation in what “consulting” means
Type 1: Advisory AI consulting firms
Advisory firms deliver strategy, recommendations, and roadmaps. They interview leadership, assess the current state, and produce a set of recommendations for what the company should build and in what order.
What they deliver: a document. Typically an AI maturity assessment, an AI strategy recommendation, a prioritised roadmap of AI initiatives, and a leadership presentation.
What they do not deliver: a context pack, trained team members, documented workflows, a running AI system, or a named AI system owner.
When advisory is the right choice: companies with strong internal AI capability that need strategic direction. Companies whose primary gap is “what should we do?” rather than “how do we build it?”
When advisory is the wrong choice: most $5M–$25M non-tech companies. The recommendations produce a plan that requires internal capability the company does not have to execute.
Type 2: Embedded AI consulting firms
Embedded firms build the AI system inside the company’s operations.
What they deliver: a running AI system. They write the context pack, document the workflows, train the team on role-specific workflows using real current work, install the improvement loop, and train the AI system owner before the engagement ends.
What they are responsible for: the AI system running at measurable quality, acceptance rate, adoption frequency, by the end of the engagement.
When embedded is the right choice: most $5M–$25M non-tech companies. The internal capability to implement is missing. The embedded firm provides it.
The most important evaluation question
Before signing with any AI consulting firm, ask:
“What will be running and measurable in 90 days?”
If the answer is a strategy document, the firm is advisory.
If the answer is a trained context pack, three running workflows with measured acceptance rates, and a trained AI system owner, the firm is embedded.
The combination: what to buy from whom and in what order
Most $5M–$25M non-tech companies need both a software vendor and a consulting firm. The question is what to buy from each and when.
Step 1: Buy the foundation model platform subscription (before engaging a consulting firm)
Claude Teams or ChatGPT Team: $25/user/month for the platform. This is the environment the consulting firm will build the context pack, workflow library, and shared workspace in.
Why before the consulting firm: the consulting firm does not sell you the platform, they build what lives inside it. Having the platform in place before the engagement means the first week of Phase 1 goes to context pack writing rather than tool selection and setup.
Cost: $250/month for a 10-user team. Setup time: 30 minutes.
Step 2: Engage an embedded AI consulting firm for Phase 1 and Phase 2
The consulting firm writes the context pack and workflow documentation (Phase 1), trains the team on role-specific workflows (Phase 2), and installs the improvement loop and AI system owner. The engagement runs 8–12 weeks.
What the consulting firm delivers that the software vendor does not:
- The context that makes the platform produce company-specific outputs
- The workflow documentation that makes the platform’s capability accessible to every team member
- The trained team that actually uses the system
Cost: $15,000–$30,000 for Phase 1 and Phase 2 in a mid-market embedded engagement. Timeline: 8–12 weeks.
Step 3: Buy the automation platform after Phase 1 is complete
Make, Zapier, or n8n: $20–$100/month for the automation layer. This connects the AI platform to the company’s operational tools and enables the Phase 3 automated workflows.
Why after Phase 1: the automation platform’s value depends on having documented workflows to automate. Buying it before the workflows are documented produces an automation environment with nothing to automate. After Phase 1, the three to five documented workflows have specific inputs, prompt structures, and output requirements, the automation platform can be configured directly from the documentation.
Cost: $50–$100/month. Configuration time: 2–4 hours per workflow.
The total investment at a glance
| Investment | Provider type | Cost | What it produces |
|---|---|---|---|
| Claude Teams (10 users) | Software vendor | $250/month | The shared AI environment |
| Phase 1 + Phase 2 engagement | Embedded consulting firm | $15,000–$30,000 | Context, workflows, trained team, improvement loop |
| Make or Zapier (automation) | Software vendor | $50–$100/month | The automation layer for Phase 3 workflows |
Evaluating an AI consulting firm: what to look for and what to avoid
Question 1: “What will be running and measurable in 90 days?”
Advisory answer: a strategy document, a roadmap, and a set of recommendations.
Embedded answer: a loaded context pack, three to five running workflows at a specific acceptance rate target, a trained team, and a named AI system owner.
The answer tells you which type of firm you are evaluating before any further conversation is needed.
Question 2: “Who writes the context pack: your team or ours?”
A consulting firm that writes the context pack without significant input from the company’s founder and ops lead will produce a generic foundation.
The context pack must reflect the company’s specific voice, clients, and operating conventions, knowledge that only the company has.
The right answer: the consulting firm structures the work and produces the draft. The company’s leadership provides the operational knowledge through a structured interview process. Both contributions are essential.
Question 3: “How do you measure the success of your engagement?”
Advisory answer: the quality of the recommendations or the client’s satisfaction with the deliverable.
Embedded answer: the acceptance rate on the deployed workflows, the team adoption rates in the tracking log, and whether the AI system owner is maintaining the system independently by engagement end.
A firm that cannot answer with specific metrics is not measuring operational change.
Question 4: “What condition does the system have to be in before you exit?”
Advisory answer: the engagement ends when the deliverables are presented.
Embedded answer: the engagement does not end until the AI system owner is maintaining the system independently, the workflows are at the acceptance rate target, and the team is using the system without the firm’s presence in the room.
The exit condition is the most revealing evaluation signal. A consulting firm that commits to a specific system state on exit is an embedded firm. One that commits to a date and a deliverable is an advisory firm.
Common questions on AI consulting firm vs software vendor
”What if a software vendor also claims to provide consulting?”
Evaluate by deliverable, not by label. Ask: “What will be running and measurably working in 90 days?”
If the answer is a strategy document or a generic implementation guide, the consulting claim is advisory.
If the answer names a specific context pack, specific workflows at a specific acceptance rate, and a trained system owner, the consulting claim may be embedded.
”Is there a consulting firm that also sells the software?”
Some advisory firms have partnerships with specific AI tool vendors and receive referral fees. This is not necessarily a problem, but it is worth asking: “Do you receive any compensation from the tools you recommend?”
A transparent firm answers directly. A firm that redirects to “we recommend the best tools for your situation” may be bundling a software margin into a consulting fee.
”What if the consulting firm recommends a tool I already have?”
This is ideal. A consulting firm that can work inside the tools you already have produces a faster Phase 1 (no tool selection or setup) and a lower total investment (no additional tool subscriptions required).
The foundation model subscription (Claude Teams or equivalent) is almost always the only new tool a Phase 1 engagement requires.
”How do I know if an AI consulting firm is advisory or embedded before signing?”
Ask all four evaluation questions above before signing. The pattern of answers tells you the engagement type before the contract is written.
A firm that answers Questions 1 and 4 with specifics, a named running system at a named acceptance rate as the exit condition, is embedded. A firm that answers with delivery dates and document lists is advisory.
”What should I ask for in a consulting contract to ensure I get an operational system?”
Three contract terms that signal an embedded engagement:
- Acceptance rate targets: the contract specifies a minimum acceptance rate (typically 80%) for each deployed workflow before the engagement concludes
- System state exit condition: the engagement does not end until named conditions are met, not until a named date arrives
- Training completion standard: the contract specifies that training sessions run on real current work, not on demo inputs, and end when the team member has produced an output they actually used
”Can I use a software vendor’s implementation team instead of an external consulting firm?”
Some foundation model providers (Anthropic, OpenAI) offer enterprise implementation support. This is worth evaluating, with the same four questions applied to the vendor’s implementation team that you would apply to an external consulting firm.
The distinction between “implementation support” (configuration help for the tool) and “embedded consulting” (building the operational foundation) applies to vendor implementation teams as much as to independent firms.
Looking for the embedded firm: the one that produces a running system, not a strategy document?
An AI software vendor and an AI consulting firm are different products that solve different problems. The software vendor sells the capability. The consulting firm builds the infrastructure that makes the capability work for the company’s specific operations.
Both are necessary. Neither substitutes for the other.
The company that gets the sequence right has a working AI system at the end of Phase 1. The company that confuses the two types of provider spends the same money and has either a capable tool nobody uses or a strategy document for a system nobody built.
Path one: use the four evaluation questions today. Apply them to the firm you are currently considering. The pattern of answers tells you whether you are evaluating an advisory or embedded firm before the proposal arrives.
Path two: bring in a partner. Phos AI Labs answers all four evaluation questions with specifics, because every engagement produces a running system, not a strategy document, and the exit condition is a named system state, not a named date. We have run 400+ AI engagements. Clients include Zapier, Coca-Cola, Medtronic, Dataiku, and American Express. Thirty minutes, no deck. Start here.