Every AI Option for Your Business, On One Map — Salesforce and Beyond
Five jobs, every option, and the cost shapes vendors won't put side by side. The demystifying map of business AI — Salesforce, Claude, and beyond.
If the AI conversation around your business feels confusing, the confusion is not your failing. It's manufactured. Every vendor presents their product as "AI for your business" — Agentforce, Copilot, Data Cloud, credits, agents, prompts — and none of them show you the whole board. Your Salesforce rep isn't lying to you; they're selling from a catalog, and the catalog leaves out every option that doesn't carry their logo.
This is the whole board. Five jobs, every option worth knowing about — Salesforce's, Anthropic's, Microsoft's, and the ones nobody sells — with the cost shapes side by side. Read it once and you'll be the calmest person in your next vendor meeting.
The five jobs
Everything anyone will ever try to sell you fits one of these:
- Help a person while they work. Drafting an email, summarizing a record, filling fields at the click of a button. A human is present, reviewing everything.
- Do bounded work in the background. Score, classify, enrich, summarize, draft — at volume, on a schedule, with nobody watching. The workhorse job, and the one with the fastest payback.
- Act autonomously with customers. The agent answering questions on your website at 2am, resolving cases without a human. High stakes, real requirements.
- Let people talk to your data. "Which deals stalled last quarter and why?" — asked in plain English, answered from your actual CRM, documents, and email.
- Build and run your systems. AI doing the admin and development work itself — configuring, migrating, cleaning, documenting. The job vendors mention least, because there's no product to sell you. Only outcomes.
The map
Each job, the options across the whole market, and the shape of what you'd pay. Cost shapes matter more than exact prices — prices change quarterly; shapes don't.
| Job | Your options (all vendors) | Cost shape |
|---|---|---|
| 1. Help a person work | Claude or ChatGPT seats for staff; Microsoft Copilot; Salesforce prompt buttons (Prompt Builder) and Agentforce Coworker in the search bar; AI already inside tools you own (QuickBooks, Gmail, HubSpot…) | Per seat: ~$20–$30/user/mo for general AI seats; Salesforce's versions require their AI add-ons |
| 2. Background work at volume | Salesforce prompt templates called from Flow; an installed engine calling an AI model directly from your Salesforce; middleware (Make/Zapier) with an AI step; AppExchange AI apps; classic Einstein point features (lead scoring, activity capture) | Metered per action (~$0.10 native), OR a flat build + a utility-bill AI account (~$20–$100/mo total), OR per-user app pricing |
| 3. Autonomous customer-facing | Salesforce Agentforce (the real contender here); vertical AI products for support; custom-built agents for the ambitious | Per conversation (~$2) or add-on SKUs ($125–$550/user/mo); this job is genuinely expensive everywhere |
| 4. Talk to your data | Claude connected to Salesforce (MCP) under each person's own permissions; Agentforce Coworker inside Salesforce; Slack AI surfaces | A $20–$30 seat plus setup, or bundled with Salesforce AI add-ons |
| 5. Build & run your systems | An AI-native consultant operating on your org (the "headless" model); AI-assisted admins; increasingly, Salesforce's own natural-language setup tools | Services pricing — you buy outcomes, not licenses. No SKU exists for this job |
Notice what your rep's pitch covers: job 3 (the expensive one), the Salesforce column of jobs 1 and 4 — and almost never jobs 2 and 5, where small and mid-sized businesses usually get the fastest, largest returns. That's not malice; it's a catalog. Jobs 2 and 5 barely have SKUs.
The four cost shapes
Every AI price on the market is one of four shapes. Once you see them, every quote becomes comparable:
$/seat
Subscriptions — scale with headcount
$/action
Metered credits — scale with volume
Flat + utility
Build once, pay pennies to run
Bundled
Already in tools you own
The shapes fail in different ways. Per-seat pricing punishes growth in people. Metered pricing punishes growth in volume — and volume is the point of automating. Flat-build pricing costs more on day one and almost nothing after. Bundled AI is free until you need it to do something specific, at which point it isn't.
Anchors, all public: general AI seats run ~$20–$30/user/month. Salesforce's Agentforce add-ons run $125–$550/user/month, with metered usage at roughly $0.10 per standard action (credits sold at $500 per 100,000, minimum) and ~$2 per agent conversation. A flat-built engine — the kind we install and hand you to own — runs on your own AI account at $20–$100/month total, regardless of user count.
One job, three ways: a worked example
Say a business gets 2,000 inbound leads a month and wants each one scored, sorted, and paired with a draft follow-up — job 2, the background workhorse. Three legitimate ways to do it:
| Native prompts + Flow | Installed engine | Agentforce agent | |
|---|---|---|---|
| How it works | Salesforce Prompt Builder templates, called from Flow | Small engine in your org calls the AI model directly | Conversational agent with actions |
| Build cost | Similar either way — prompts, Flows, testing, training | Similar either way (a flat build you own, in our case) | Highest — agent design, testing, guardrails |
| Monthly usage at 2,000 leads | ~$200, metered, forever | ~$10–$30 on your own AI account (halved in overnight batch) | Wrong shape — priced per conversation, built for dialogue |
| At 10,000 leads | ~$1,000/mo | ~$50–$100/mo | — |
| Output | Text into a field | Structured values across many fields, atomically | Conversation transcripts + actions |
| You own it | Config in your org; usage meter is theirs | ✓ Code, config, and the AI account | Rented capability |
Below roughly 500 records a month, with Salesforce AI licensing already in place, the metered path costs under $50/month and the native option is a fine answer — the Trust Layer's data masking comes included, and there's no AI account to own. We build it that way when it fits. The engine earns its keep on volume, multi-field outputs, and cost predictability.
One more thing the worked example teaches: the build effort is the constant, not the variable. Whichever path you pick, someone competent still has to write the instructions, wire the automation, test it against real records, and train your team. What you're actually choosing between is the run-rate and the ceiling.
The job nobody sells: build and run your systems
Job 5 deserves its own moment, because it's the one with no brochure. The same AI that can score a lead can also audit your Salesforce org, untangle five years of accumulated configuration debt, migrate data, write the documentation nobody ever wrote, and keep the whole thing tuned — when operated by someone who knows what they're doing. Salesforce's own tooling is moving this direction (their newest tools let admin work happen through natural language), but you can't buy it as a product today. You buy it as outcomes from a practitioner who works this way.
This is where AI-native consulting quietly beats both the big-firm model (junior consultants, billable hours) and the do-nothing option — and it's why the ecosystem doesn't advertise it. There's no license revenue in it.
Five questions that cut through any AI pitch
Take these into every vendor meeting — including ones with us:
- What exactly do I own when this is done? Config in my org? Code? Nothing but a subscription?
- What's my bill at ten times the volume? Metered pricing hides its shape at pilot scale. Make them do this math in front of you.
- What do I have to buy first? Edition upgrades, add-on SKUs, data platforms — the prerequisites are routinely bigger than the pitch.
- What happens when I cancel? Does the capability stay, degrade, or vanish?
- Which of my jobs is this actually for? If the answer isn't one of the five jobs above, in plain English, the pitch is about their product, not your business.
You don't need to understand AI to buy it well. You need to know which job you're hiring it for, what shape the bill is, and what you own when the vendor leaves the room.
Eric Lovgren, lovgren.ai
Want the map read against your actual business?
Thirty minutes is usually enough to sort your top three jobs, name the right option for each — whichever column it lives in — and tell you plainly if the answer is "not yet."
No Pitch, Just Possibilities.