The AI Agency Sales Process: From Cold Outreach to Signed Contract

Most AI agencies fail at sales, not delivery. The founders who know how to train models spend their time tweaking datasets and never build a pipeline. The result is feast-or-famine: one great referral client, then three quiet months.

This is the complete sales process for an AI agency selling custom model work.

The Sales Cycle Overview

Stage	Goal	Typical Duration
Outreach	Get a discovery call	2-4 weeks per batch
Discovery call	Qualify and uncover the problem	30-60 minutes
Proposal	Present solution and price	3-7 days after discovery
Follow-up	Address objections, build confidence	1-3 weeks
Contract	Close and collect deposit	1-5 days after proposal

Total average cycle: 4-8 weeks from cold contact to signed contract. Warm referrals close in 1-3 weeks.

Stage 1: Outreach

Who to Target

The best-fit buyer for custom AI model work:

Has an existing AI spend problem (OpenAI invoice growing, support ticket backlog)
Has structured data (customer interactions, documents, product catalogs)
Is in a niche where you have vertical credibility
Is decision-maker-accessible (founder, CTO, or VP at a sub-200-person company)

Avoid: enterprises with long procurement cycles, companies with no existing AI exposure (education required), industries with unclear data access.

Outreach Channels (Ranked by ROI)

1. Warm network referrals — Ask every past client, colleague, and professional contact if they know anyone with AI cost or quality problems. One sentence: "I'm helping [vertical] companies reduce their AI API costs by 70%+ with custom models. Do you know anyone who might be dealing with this?" Conversion rate: 30-50%.

2. LinkedIn cold outreach — Target by title (CTO, Head of Product, Founder) + industry. Connection request + short note referencing a specific pain point in their industry. First message is never a pitch — it is a question or observation. Conversion to call: 3-8% with well-targeted lists.

3. Industry communities — Slack workspaces, Discord servers, subreddits where your target buyer hangs out. Contribute genuinely for 2-4 weeks before any mention of your services. Post case studies anonymously ("here's how a legal firm cut AI costs 80%"). Conversion rate from warm community member: 10-20%.

4. Content-driven inbound — Blog posts that rank for "[vertical] AI fine-tuning" or "reduce [tool] API costs." Slower to start but self-qualifying. The buyer who finds you via search has already diagnosed the problem.

5. Cold email — Lower conversion than LinkedIn but scalable. Subject line: a specific metric from their industry. Body: two sentences on the pain, one sentence on your solution, one ask (call). Keep it under 100 words.

The Outreach Message That Works

Subject: [Company] — AI API costs

Hi [name],

Noticed [company] is using [OpenAI/Claude/Anthropic API] for [use case you know or can infer]. For [vertical] companies doing this at scale, API costs usually hit a wall around $3,000-5,000/month.

We help [vertical] businesses replace that spend with a fine-tuned local model — same or better accuracy, flat monthly cost instead of per-token.

Worth a 20-minute call to see if this applies to you?

The key: one specific pain point, not a feature list.

Stage 2: Discovery Call

The discovery call is the most important part of the sales process. If you do it wrong, the proposal will miss. If you do it right, you have almost sold the project before writing a word.

Structure (45-60 minutes)

10 minutes: Understand the current state

What are you currently using AI for?
What is your monthly AI spend (API costs)?
How is the model performing — what's working, what is failing?

15 minutes: Uncover the pain

What does a wrong AI response cost you? (Support ticket, manual rework, customer churn)
Is accuracy good enough, or are you seeing a specific failure mode?
What would you do differently if AI costs were not a constraint?

10 minutes: Understand the data

What data do you have that relates to this use case?
Is it structured? (Customer interactions, labeled outputs, documents)
How much of it? (Volume matters for training feasibility)

10 minutes: Understand the decision process

Who else is involved in this decision?
What does your timeline look like?
What is your budget range for solving this?

5 minutes: Outline your approach

High-level: fine-tuning on their data, local deployment, flat cost model
What you need from them to move forward
Next step: proposal within X days

Qualifying Questions You Must Answer Before Leaving

Is there a real problem? (Specific cost or accuracy pain — not hypothetical interest)
Is there accessible data? (At least hundreds of examples to work with)
Is there a budget? (They should be able to say a number, even a range)
Is this person the decision maker? (Or who is?)

If you cannot answer yes to all four, you may have an educational call — not a qualified prospect.

Stage 3: Proposal

Send the proposal within 3-5 days of the discovery call. Sooner is better — the project feels real while the conversation is fresh.

Use the 7-section structure: executive summary, problem definition, proposed solution, methodology, timeline, investment, why us. The executive summary leads with their cost number, not your technology.

Proposal email subject: [Company] — Custom AI Model Proposal

Email body:

Hi [name],

Attached is the proposal based on our call. The summary: we will build a fine-tuned model for [use case] that reduces your AI spend from $[X]/month to under $[Y]/month, targeting ≥[accuracy]% accuracy on your [task].

Total investment: $[price]. Timeline: [X] weeks.

Happy to walk through it on a call — I have [Tuesday at 2pm / Thursday at 3pm] open. Or reply with any questions.

[Name]

Do not write a long email. The proposal document carries the content.

Stage 4: Follow-Up and Objection Handling

Most deals are not closed in the first response. Follow up 3-4 times before letting a prospect go quiet.

Follow-up sequence:

Day 3 after proposal: "Did you have a chance to review? Happy to answer questions."
Day 7: Send a relevant piece of content (case study, article about their vertical)
Day 14: "Still thinking this through? Happy to get on a quick call to address any concerns."
Day 21: "Following up one more time — if the timing is off, no problem. Just let me know."

Common Objections and Responses

"This seems expensive." Go back to the ROI math from the proposal. If they are spending $3,500/month on API costs and you will reduce it to $200, your $12,000 project pays back in 3.3 months. Frame the investment as buying years of savings upfront.

"We do not have time to collect training data." Reframe: you handle the data pipeline. Their team exports the raw data (interaction logs, documents, tickets); you do the cleaning, formatting, and curation. Ask what format their data currently lives in.

"Can we start smaller?" Yes — a pilot project. Offer a smaller scope (single use case, 4-week engagement) at a proportionally lower price. This lowers the perceived risk and almost always converts to a full project or retainer once results are visible.

"We are evaluating other vendors." Ask what they are comparing on. If it is price, show the total cost of ownership (your project cost vs. 12 months of API spend). If it is capability, offer a demo using their data (or similar public data) showing your accuracy numbers.

"We need to get legal/IT/compliance approval." This is a real objection. Ask what the approval process looks like and what materials they need. Often you can prepare a security/data architecture overview that shortens the internal review.

Stage 5: Contract and Close

Once the prospect says yes (or signals they are ready), move fast. Send the contract within 24 hours. Momentum evaporates if you take a week to get paperwork ready.

Contract essentials:

Scope definition (exactly what model, for exactly what use case)
Deliverables list (model file, deployment setup, documentation, training session)
Timeline and milestone schedule
Payment terms (typically 50% upfront, 50% on delivery — or milestone-based)
IP ownership (client owns the trained model; you retain methodology)
Data handling and confidentiality (their data does not leave their environment or yours, never used for other clients)

Payment: Invoice via Stripe at contract signing for the upfront portion. Automated invoice for the remaining payment at the final milestone. Do not deliver the final model file before payment clears.

Onboarding trigger: Once deposit arrives, create the client project in Ertas, send the data collection guide, and schedule the kickoff call.

Ship AI that runs on your users' devices.

Ertas early bird pricing starts at $14.50/mo — locked in for life. Plans for builders and agencies.

View early bird pricing or join the waitlist →