If an AI consulting engagement is going to fail, the signals appear in the first 30 days.
Not dramatically, with no missed deadlines or visible quality problems.
Subtly: the context pack interview is 60 minutes instead of four hours, the first workflow document is built from observation rather than structured mapping, and the AI system owner has not been scheduled for their first shadowing session.
Each signal alone is survivable. Together, they predict the engagement outcome with uncomfortable accuracy.
This article describes what the first 30 days of a well-structured AI consulting engagement should produce: specific deliverables, specific company contributions required, and specific week-by-week progress markers.
The goal is a founder who can evaluate week one against what week one should look like, rather than discovering in month six that the first month set the wrong trajectory.
Week 1: Orientation and operational knowledge extraction
What the consulting firm does in week one
Days 1 and 2: engagement orientation
The consulting firm and the company confirm the engagement scope, the communication cadences, and the specific calendar blocks for Phase 1 work. The AI system owner is confirmed (or identified if not yet named).
The Phase 1 work schedule is established, not in general terms (“we will schedule interviews”) but with specific calendar commitments: the founder interview is on [date], the workflow mapping sessions are on [dates], the first training session is on [date].
The consulting firm also conducts a 60-minute baseline review with the founder. It covers what AI tools the team currently uses, what outputs they produce, what they find useful versus requiring heavy editing, and the team’s current relationship with AI.
Days 3 to 5: tool access and environment setup
The consulting firm configures the shared AI workspace. Claude Teams or the chosen platform is set up with appropriate user access, the project or shared space is created, and the structure for loading the context pack is established.
If the company does not yet have tool subscriptions in place, the consulting firm provides specific recommendations and the company purchases them directly.
Day five should end with the shared workspace accessible to the consulting firm, the founder, and the AI system owner.
What the company must provide in week one
- Founder or COO: 90 minutes for the baseline review conversation
- AI system owner (if named): 60 minutes to be briefed on the engagement structure and their role
- Tool access: Claude Teams or equivalent purchased and set up, or a clear timeline for when it will be
What should exist at the end of week one
- A shared workspace that is accessible and structured for Phase 1 content
- A calendar with all Phase 1 sessions booked, not tentative but confirmed
- A baseline understanding of current AI usage that will inform the context pack build
- The AI system owner briefed on their role in Phase 1
Signal that week one is off track: the context pack interview is not yet scheduled at the end of week one.
A consulting firm that has not committed to a specific date for the founder interview in the first week has either poor organisation or a planning approach that delays the founder-intensive work as long as possible.
This is the clearest early signal that Phase 1 will be built from surface information rather than operational knowledge.
Week 2: The founder interview and context pack first draft
The founder interview is not a kickoff meeting. It is a structured extraction of the operational knowledge that lives in the founder’s head and cannot be found in any document the company has published.
The six areas the founder interview must cover
Area 1: Voice and communication standards (45 minutes)
What does the company sound like at its best?
The consulting firm walks through three to five examples of the company’s best communications: a proposal the founder is proud of, a client email that got a positive response, an internal document that captures the company’s thinking well.
The consulting firm extracts specific vocabulary used and avoided, sentence structures, tone calibrations across contexts, and the one or two characteristics that make the company’s communication distinctively theirs.
Area 2: Client archetypes (60 minutes)
Who are the two to four types of clients the company serves? For each type, the consulting firm extracts the seven archetype components.
Those components are: role and context, operational situation, trigger to the conversation, primary concerns, communication preferences, success definition, and relationship sensitivities.
These cannot be accurate from website research alone.
Area 3: Decision rules (45 minutes)
What does the company do in specific recurring situations: discount request, client complaint, scope change, payment delay?
For each category, the consulting firm extracts the specific rule. Not “use good judgment” but “for retainer clients requesting a discount, the account lead can authorise up to X% without approval, and anything above requires the founder.”
Area 4: Competitive positioning (30 minutes)
How does the company describe its differentiation to a specific prospective client? What are the two or three things it does that the client could not get from a direct competitor?
What are the situations where the company declines to compete?
Area 5: Service and product standards (30 minutes)
What does good work look like at this company? For each service type, what are the quality markers that distinguish an excellent output from a merely acceptable one?
Area 6: Operating conventions (30 minutes)
What are the working conventions the team follows that are not documented anywhere? Meeting formats, approval processes, client communication cadences, delivery standards, the things every team member knows but new hires have to learn through osmosis.
Total interview time: approximately 4 hours. This is non-negotiable. A context pack built from less than this will show the gap in every AI output it produces.
Testing the first draft context pack
The first draft should be produced within 48 hours of the founder interview. The test is not a document review. It is a live output test.
Load the draft context pack into the shared workspace. Run three real current tasks against it:
- Draft a follow-up email to a specific client
- Produce a one-paragraph description of the company’s approach for a prospect presentation
- Draft a status update for an active project
Compare each output against what the team was producing before the context pack was loaded.
The test passes if the outputs are materially more specific: they sound like the company, reference the right clients, use the right vocabulary, and reflect the decision rules.
If they do not: identify the specific gap and update the relevant context pack section before proceeding.
What should exist at the end of week two
- The founder interview completed (minimum 4 hours)
- First draft context pack produced within 48 hours of the interview
- Context pack tested against three real current tasks
- At least one revision cycle completed based on the test results
- The AI system owner present for the interview and the first test session
Signal that week two is off track: the context pack draft is produced from a 60-minute kickoff meeting and document review rather than a structured 4-hour interview.
This is the most expensive early signal.
The context pack built from insufficient input will require significant revision in weeks four through eight when the team starts producing outputs that reflect the generic context rather than the company’s specific operational reality.
Weeks 3 and 4: Workflow inventory, mapping, and workspace loading
Workflow inventory (week 3, days 1 and 2)
The consulting firm conducts structured interviews with key team members to produce the workflow inventory, a list of every recurring AI-candidate task the team runs, with frequency and time estimates.
For a 10 to 20 person company, this takes four to six 20-minute conversations with team members in different functions.
By day two of week three, the consulting firm presents a ranked workflow list with the top five to eight candidates identified.
The founder and AI system owner confirm the top three to begin workflow mapping immediately.
Workflow mapping (week 3, days 3 to 5)
For the top three workflows, the consulting firm conducts the structured 20-minute mapping interview with the specific team member who runs each workflow.
Each interview produces a workflow specification document in the five-component format: trigger, inputs, decision points, human checkpoints, expected outputs.
By the end of week three, three completed workflow specification documents exist. Each is detailed enough that a new team member could run any workflow independently, and detailed enough for the consulting firm to write the AI prompt architecture from the document alone.
Shared workspace loading (week 4, days 1 to 3)
The context pack and the three workflow specification documents are loaded into the shared workspace. The consulting firm configures the workspace structure. The AI system owner is walked through the structure and confirms they can navigate it independently.
First independent workflow runs by the AI system owner (week 4, days 3 to 5)
The AI system owner runs each of the three documented workflows independently, without the consulting firm present.
They use real current work as the input. They log the outputs in the adoption tracking log and bring any below-standard outputs to the day-30 review meeting.
The day-30 deliverables checklist
At the end of day 30, six specific items should exist. Evaluate each one:
DAY-30 REVIEW CHECKLIST
------------------------
[ ] 1. Context pack: complete, tested, producing company-specific outputs on
the three test tasks
[ ] 2. Workflow specification documents: top three workflows documented in
five-component format, complete enough to build the AI prompt
architecture from
[ ] 3. Shared workspace: configured, context pack and workflow library loaded,
accessible to founder, AI system owner, and consulting firm
[ ] 4. Adoption tracking log: set up, with the AI system owner's first three
workflow test runs logged
[ ] 5. AI system owner: briefed, onboarded, and having run the first
independent workflow tests
[ ] 6. Day-30 review meeting: both parties assess the deliverables against
these criteriaWhat the day-30 review meeting should determine:
- Is the context pack producing company-specific outputs? (Test it live in the meeting if not already done)
- Are the three workflow specification documents complete enough to build from?
- Has the AI system owner run the workflows independently and logged the results?
- What is the acceptance rate on the AI system owner’s test runs? (Should be 70%+ even at this early stage with the context pack loaded)
If all six items are complete and the acceptance rate on the system owner’s test runs is 70%+: Phase 2 team training can begin in week five.
If any item is incomplete or the acceptance rate is below 70%: identify the specific gap and close it before Phase 2 training begins.
What to do if the first 30 days are not on track
If the founder interview has not happened by day 14
Send the consulting firm a specific message: the founder interview must be scheduled before any other Phase 1 work proceeds. Not because other work is not valuable, but because everything else depends on the interview’s content.
Reschedule the context pack draft delivery for one week after the interview is completed.
If the context pack draft lacks specificity
Run the three test tasks with the founder and the consulting firm together. For each output that is not company-specific enough, identify the specific element that is missing: vocabulary, client archetype detail, or decision rule.
Add that element to the context pack before the meeting ends. Require a revised draft within 24 hours.
If the workflow mapping interviews have not started by day 21
The Phase 2 training timeline is now compressed. Reschedule Phase 2 to start one week later than planned for every week of workflow mapping delay.
Do not begin training on undocumented workflows. An undocumented workflow cannot be trained consistently. Name the specific interviews that must happen in the next three business days.
If the AI system owner has not been involved
Include the AI system owner in every remaining Phase 1 activity: the context pack revision, the workflow mapping interviews, the workspace loading. Not as an observer but as a contributor.
The system owner who reaches Phase 2 without having participated in Phase 1 building will take longer to develop maintenance independence in Phase 3.
If the shared workspace is not configured by day 25
The first team training sessions cannot begin until the workspace is configured and tested. Treat the workspace configuration as an immediate priority. Schedule a 90-minute session with the consulting firm to complete it before day 30.
The workspace configuration itself takes 2 to 3 hours. The delay is usually scheduling, not complexity.
Common questions on the first 30 days
”What if the consulting firm says four hours for the founder interview is too much to ask?”
Four hours across two sessions, two hours each, is 240 minutes spread over a week. A consulting firm that resists this is either poorly organised or building the context pack from surface information.
The context pack built from a 60-minute kickoff and a document review will underperform for months before the gap is clearly diagnosed. The four-hour investment at the start prevents 40 hours of suboptimal outputs after.
”Should I review the context pack draft before it is loaded into the workspace?”
Review it as part of the live test, not as a document review. Reading the context pack document will not reveal the same quality gaps that running three real current tasks against it will.
The output test is the review. It takes 45 minutes and reveals specifically what needs to change. A document review takes the same time and reveals much less.
”What if the AI system owner is not available for the first 30 days?”
Do not begin the engagement without naming the AI system owner. This is the company-side failure cause that most commonly causes the handover to fail later in the engagement.
If the intended system owner is temporarily unavailable, delay the engagement start by the same period. Starting without them means they will be onboarded to a system they did not build, which extends the time to maintenance independence.
”How detailed should the workflow specification documents be at day 30?”
Detailed enough to build the AI prompt architecture from the document alone, without asking anyone additional questions.
A practical test: give the three workflow documents to someone who was not in the mapping interview and ask them to describe what input they would provide and what output they would expect.
If they can do this accurately, the documents are sufficiently detailed. If they need to ask clarifying questions, add the answers to the relevant section before proceeding to Phase 2.
”What happens if the context pack test results are poor, do we restart Phase 1?”
No. Identify the specific gap revealed by the test, update the specific section, and retest.
A poor first-draft test result is common and expected. The test is designed to reveal gaps before Phase 2 training begins, not to confirm that the first draft is perfect.
Restart Phase 1 only if the context pack is so generic that it produces outputs no more specific than unconfigured Claude.
That outcome indicates the founder interview did not happen at the depth required, and the interview should be repeated before any further work proceeds.
Want to know exactly what Phos AI Labs delivers in the first 30 days, and hold us to it?
The first 30 days of an AI consulting engagement produce five specific deliverables: the founder interview, the tested context pack, the workflow specification documents, the configured shared workspace, and the AI system owner’s first independent workflow runs.
These deliverables are the foundation everything subsequent is built on.
The engagement that produces all five by day 30 is positioned to run Phase 2 team training on a solid foundation in week five. The one that is still completing the founder interview in week four is setting up Phase 2 to train on an incomplete foundation.
Path one: use the day-30 checklist to evaluate your current engagement. If you are already in an engagement, assess which of the six items exist today. Any missing item is the first thing to address in the next working session.
Path two: bring in a partner. The five deliverables in this article are the contractual first-month output of every Phos AI Labs Phase 1 engagement. The day-30 review meeting is standard practice, with the founder evaluating the deliverables against the criteria above and both parties confirming Phase 2 readiness before training begins. We have run 400+ AI engagements. Clients include Zapier, Coca-Cola, Medtronic, Dataiku, and American Express. Thirty minutes, no deck. Start here.