A small advertising agency can cover its main AI jobs with six tools: Claude for copy variants, Adobe Firefly for concept mock-ups, Canva for resizing, AgencyAnalytics for client reports, Perplexity for research and Fathom for brief capture. Together they cost roughly $725 a month for a nine-person agency, before checking each tool's rights and disclosure terms.
That list is one defensible answer, not the only one, and the prices come with a caveat: plan names and allowances in this market change several times a year, so every figure below needs confirming on the vendor's own page on the day you buy. The part that outlasts any price is the method: one tool per job, a test on real work for each, and a rights check before anything reaches a client.
How the six jobs were chosen
Agency work splits into jobs that repeat across every client and jobs that are craft. AI pays off fastest on the first group, where the output is easy to check against something concrete: a character limit, a brand guide, a spreadsheet. It pays off least where the value is the idea itself. So each job below is repetitive, checkable and tied to hours you can count. The tools are picked for fit and verified pricing, and I name alternatives only where they change the decision.
| Job | Tool | Listed price (check before buying) |
|---|---|---|
| Copy variants | Claude (Team plan) | $20 per seat a month annual, $25 monthly; 2 to 150 seats |
| Concept mock-ups | Adobe Firefly | Standard $9.99 (2,000 credits); Pro $19.99 (4,000 credits) |
| Resizing and versions | Canva (Business) | $250 a year per person; Pro is $144 a year |
| Client reports | AgencyAnalytics (Core) | $20 per client a month on annual billing |
| Research and competitor scans | Perplexity Pro | About $20 a month (an annual rate near $17 a month has also been shown) |
| Brief capture from calls | Fathom | Free plan; Team $15 per user annual, $19 monthly, two-user minimum |
1. Copy variants: Claude with a brand-voice Project
Claude's Team plan lists at $20 per seat a month billed annually ($25 monthly) for 2 to 150 seats, and Claude Projects let you store a client's voice guide, banned phrases and past approved copy so each request starts from the same context. A general assistant on a business plan from OpenAI does the same job, and ChatGPT Business lists at the same price, so choose on which one your writers prefer after a one-week trial. Business plans from both vendors do not train on business content by default, which matters when you paste in a client's unreleased campaign.
Headline variants for an event venue, counted
An illustration. An event venue wants responsive search ad headlines for a winter wedding fair. Headlines in that format have a tight character limit (Google's responsive search ad headlines have long been capped at 30 characters; confirm the current limit in the ad editor). The prompt:
Client: event venue, barn-and-ballroom wedding site.
Offer: winter wedding fair, Saturday, free entry, venue tours.
Voice: warm, plain, no clichés ("dream", "magical", "perfect day").
Write 12 headlines, each 30 characters or fewer including spaces.
Vary the angle: 3 date/urgency, 3 offer, 3 venue, 3 reassurance.
After each, show the character count.
A realistic sample of what comes back, with the checks a person must run:
Book Your Winter Wedding Fair (29)
Tour the Venue This Saturday (28)
Weddings From Barn to Ballroom (30)
Free Planning Chat With Us (26)
Your Dream Winter Wedding Starts Here (37)
What you fix: the last line is over the limit and breaks the "no clichés" instruction in the same breath. Models often miscount characters, so paste the final set into a character counter or the ad platform itself rather than trusting the numbers it printed. Next, read for claims: "Free Planning Chat" is fine only if the venue really offers one. Then sort by angle and test two or three per angle rather than twelve at once.
For keeping many clients' voices apart, how agencies keep brand voice consistent in AI-written content covers the Project structure. Watch-outs: variants converge on the same safe phrasing, so ask for "three angles the competitors would not take". Do not let a model invent testimonials, prices or availability. The client owns the claims in its ads, and you own the checking.

The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowOne weak variant, rewritten from the brief
A first-draft line from a model, for a catering company's corporate lunch offer: "Take your office lunches up a level with our delicious, high-quality catering." It says nothing a competitor could not say. A writer would ask the model for the specific detail in the client's brief: "Hot lunch for 20, delivered by 12:15, no minimum order." The second line is checkable, and checkable is the point. When a variant could sit on any competitor's page, send it back with "use only facts from the brief" and add the facts yourself.
2. Concept mock-ups: Adobe Firefly, for pitches before finals
Adobe's plans page lists Firefly Standard at $9.99 a month with 2,000 generative credits, Pro at $19.99 with 4,000, Pro Plus at $49.99 with 10,000 and Premium at $199.99 with 50,000, and team versions of Pro, Pro Plus and Premium, which cost more per licence than the individual plans. Two points on the rights side matter to an agency. Adobe states that outputs from its own Firefly models can be used commercially, and it says you are responsible for judging whether partner-model outputs suit your project, even though partner models from other companies appear in the same app. And Adobe attaches Content Credentials automatically to Firefly outputs, which records that the image is AI-generated. IP indemnification is offered to eligible enterprise customers for select outputs, not described for individual plans, so a small agency should not assume it is covered.
Three poster directions for a catering company
An illustration. A catering company wants three autumn menu campaign directions before commissioning a photographer. Spend an hour producing a one-page board per direction:
- Write a one-line brief for each direction ("warm kitchen, copper pans, steam, low light").
- Generate six images per direction using Adobe's own Firefly model, not a partner model, so the rights position is the clearest one.
- Pick the best two, drop them into the pitch deck with the label "concept, not final".
- Keep the prompt and the model name in the project folder.
The quick sum: three directions, six images each, 18 images. Credit use varies by model and feature, and standard Firefly features such as Generate Image on paid plans are described as unlimited, so the credit pool matters mostly for video and partner models. Check your plan's rules before budgeting credits.
When a mock-up promises food the kitchen cannot make
The concept image of a plated autumn dish has beautifully glazed vegetables the kitchen cannot produce. The client loves it and the final photograph disappoints. The rule is simple: mock-ups guide mood, and any final ad that shows the product uses real photography of the real product. Label every mock-up on the page as a concept and never let it drift into a live ad because the deadline slipped.
3. Resizing and versioning: Canva Business
The job that eats junior hours is turning one approved design into every format a campaign needs. Canva's pricing page lists AI tools including resize, translate and background removal on Pro at $144 a year per person, and Business at $250 a year per person adds team admin, approvals and up to 100 Brand Kits, so each client can have their own colours, fonts and logos locked in. Business also lists a larger monthly AI allowance, with add-ons at extra cost, so check what your volume needs.
One coworking banner, six formats
An illustration. A coworking space client approves a 1200 x 628 link banner. The agency needs it as 1080 x 1080 and 1080 x 1920 for social, 300 x 250, 728 x 90 and 160 x 600 for display, plus a 1200 x 630 email header. By hand a junior spends about 40 minutes per format family; with Canva's resize and a saved Brand Kit it takes perhaps 10 minutes to generate and 15 to fix: 25 minutes. Over 20 campaigns a month, that is about five hours saved, which at an internal $50 an hour is $250 against Business seats of about $21 each a month.
The fix list is where the value hides. Resize crops text, shrinks logos below legal minimums and puts faces under platform interface elements. Keep a checklist per format: text inside the safe zone, logo at least the client's minimum size, the call-to-action not hidden under platform buttons, file weight within platform limits. Use the approval feature so a client-side person signs off the set before anyone uploads it.
4. Client reports: AgencyAnalytics
AgencyAnalytics' pricing page lists a Core plan at $20 per client a month on annual billing, with volume pricing from 25 clients, a white-labelled client portal, and more than 90 marketing integrations. Its AI features on that page include insights and analysis, benchmarks and forecasting, and access through MCP to ChatGPT and Claude. A feature comparison also names Ask AI, AI Summary, anomaly detection, metric insights and trend forecasting without tying each to a plan, so ask which are in the plan you are quoted. An optional AI Tracker add-on lists at $20.83 a month per 250 credits.
Monthly reports for twelve clients, timed
An illustration. An agency reports to 12 clients each month and each report takes three hours: pulling numbers, writing commentary, formatting. With dashboards built once and an AI summary as the first draft of commentary, say it takes one hour. Time saved is 24 hours a month, about $1,200 at $50 an hour. The cost is 12 clients x $20 = $240 a month. Those hours are an assumption you must measure: time three real reports before and after.
A sample AI-written commentary, and what to fix:
Leads from paid search rose 40% month on month, driven by
the new campaign. Cost per lead fell to $18. Organic traffic
was stable.
The check: leads rose from five to seven, which is 40% and not a trend. The new campaign launched on the 24th, so it did not drive the month. And "stable" hides a drop in one landing page. Rewrite the sentence with the counts, the dates and a caution about the small sample, then add the next action. Client reports are the place AI makes a small number look like a story, so never send a summary you have not matched to the figures. Build the dashboard template once per service line, not once per client: a paid-search template, a social template and a local-visibility template, each with the same five or six metrics in the same order. Then the AI summary has a consistent shape to work from, and you can compare clients with each other. Put the client's goal in the first line of every report, because a summary written without the goal tends to praise whatever went up. A longer treatment is in how marketing agencies use AI to automate client reporting.
5. Research and competitor scans: Perplexity Pro
Perplexity answers with cited sources, which makes it quicker than a general chatbot for a first scan of a market, and the citations let you check. Pro lists at about $20 a month, though one page has shown an annual rate near $17 a month, so confirm which applies. A privacy detail matters for agencies: the AI data retention setting was on by default across Free, Pro and Max plans when last checked, so switch it off or keep confidential client information out of prompts.
A competitor scan for a translation agency
An illustration. A translation agency wants to know how three competitors describe their turnaround times. A prompt such as "list the turnaround claims on these three sites, with the exact wording and the page address" returns a table in a few minutes. The checks: open every cited page and confirm the wording and the date, because an answer built from an old cached page will quote a claim the competitor dropped. Anything that will appear in an ad, such as a comparison with a named competitor, needs substantiation you can show the client, and a screenshot saved with the date is a minimum. Treat the scan as a starting list of pages to read, not as the finding.
A research prompt template that keeps answers checkable:
Task: list the claims these three competitors make about
[topic] on their own sites. For each, give the exact wording,
the page address, and the date shown on the page if any.
Do not summarise. Do not include anything you cannot cite.
Mark any claim that appears on only one page.
Asking for exact wording, address and date turns the answer into a list of things to open and read. Without those three, a research tool hands you a confident paragraph and nothing to verify it against.
6. Brief capture from client calls: Fathom
Fathom's free plan lists unlimited recordings and transcripts, the Team plan costs $15 per user a month on annual billing ($19 monthly) with a two-user minimum, and Premium for an individual is $16 annual ($20 monthly). A kickoff call is the highest-value recording in an agency: it is where the brief is actually given, and where the account manager's notes usually miss something. I covered retention, consent and plan limits for note takers in the accounting-practice comparison; the same rules apply here.
A filled-in brief from a kickoff call
Client: event venue. Call: kickoff, 40 min.
Objective: fill 12 winter wedding-fair slots by 15 January.
Audience: engaged couples within travelling distance; 25-40.
Budget: client confirmed a paid-search budget; figure to be sent.
Must include: free entry, venue tours, sample menu tasting.
Must avoid: the word "affordable"; any mention of last year's
date clash.
Approvals: owner signs off; allow 2 working days.
Open questions: tasting numbers; photography rights from the
last event.
What a reviewer fixes: "budget confirmed" has no figure, so the summary becomes an open question, not a fact. "Within travelling distance" is the tool's gloss; the client said "an hour's drive" and the brief should keep their words. Send the brief back to the client within a day and ask them to confirm it in writing. That one email prevents most scope arguments.
The rights and disclosure checks, tool by tool
| Tool | Check before client work | Where it goes |
|---|---|---|
| Claude | Business plan, not a personal account; the client contract allows AI-assisted copy; no invented claims | Contract clause; review step |
| Adobe Firefly | Use Adobe's own models for client-facing work; read partner-model terms; keep Content Credentials; note no indemnity on small plans | Project folder note per image |
| Canva | Licence terms for templates, stock and AI output as used in client deliverables; approvals switched on for client sets | Agency asset policy |
| AgencyAnalytics | Data-processing terms; who at the client can see which dashboards; AI summaries reviewed before sending | Client onboarding |
| Perplexity | Retention setting off or no confidential material; sources opened and dated | Research log |
| Fathom | Spoken consent at the start; retention period; no sharing to attendees by default | Kickoff script |
Put the client-facing part into the contract rather than into a promise made on a call. AI clauses for agency contracts: disclosure, ownership, approvals gives wording for the contract itself. This is not legal advice: ask a qualified adviser about ownership and liability terms in your own contracts.
Wording for a statement of work
AI-assisted work. We use AI tools in drafting, concept
visuals, resizing, reporting and research. A named person
reviews all client-facing output before delivery. We use
business-plan tools whose providers state they do not train
on our business content. We will tell you before using AI-
generated imagery in published work, and we will not use
your confidential material in any tool not listed here.
Ownership of deliverables is as set out in clause [x].
A review checklist for each type of output
One review step per output type, written down, is what separates an agency that uses AI from one that gets caught out by it. Pin this to the wall or paste it into your project template.
- Copy: every claim traced to the brief; character counts checked in the platform; banned words searched; no invented prices, dates or testimonials.
- Images: model recorded; rights checked for that model; labelled as concept; nothing that misrepresents a real product; any text in the image proofread letter by letter.
- Resized assets: each format opened at actual size; text inside safe zones; logo above the client's minimum; file weight within platform limits.
- Reports: each sentence matched to a figure; dates of changes checked against dates of results; small samples flagged as small.
- Research: every source opened and dated; competitor claims saved with a screenshot before they go anywhere near an ad.
- Briefs: the client's own words kept; open questions listed; sent back for written confirmation.
Two people should know about each review: the person who made the output and the person who signs it off. If those are the same person on a rushed Friday, the review is the first thing to go, so name a second pair of eyes for anything the client will see.
What the six tools cost together
An illustration for a nine-person agency at the listed annual-billing prices, as of November 2026:
- Claude Team, 9 seats x $20 = $180 a month
- Adobe Firefly Pro, 3 individual licences x $19.99 = about $60 a month (team licences cost more each)
- Canva Business, 6 people x $250 a year = $125 a month
- AgencyAnalytics Core, 12 clients x $20 = $240 a month
- Perplexity Pro, 3 seats x $20 = $60 a month
- Fathom Team, 4 users x $15 = $60 a month
That totals about $725 a month, or $8,700 a year. Suppose the tools save 15 hours on copy variants, 8 on mock-ups, 10 on resizing, 24 on reports, 6 on research and 6 on briefs: 69 hours a month, about $3,450 at $50 an hour. These hour figures are my assumptions, so replace them with your own timed numbers. If a tool does not repay its cost in your numbers within a quarter, cancel it at the next renewal. The question of whether to pass savings to clients is its own decision, covered in whether to charge clients less when AI speeds up your work.
A thirty-day rollout in order of payback
- Week 1: reporting. Build dashboards for two clients and time the old and new routes.
- Week 2: resizing. Set up Brand Kits for three clients and run one real campaign through.
- Week 3: copy variants. Create a Project for two clients and run a head-to-head test of the AI's variants against your writers' own on a live ad set.
- Week 4: briefs and research. Record kickoff calls with consent, and run one competitor scan. Hold mock-ups until the rights wording is in your contracts.
Review at the end of the month: hours saved, errors caught, anything a client queried. Keep a log of errors, because the log is what tells you where review needs to be stricter.
Three ways it goes wrong
- Client confidential material in a personal account. A designer pastes an unreleased product name into a free chatbot. Buy business seats and write the one-page rule.
- Mock-up becomes final. A concept image slips into a live ad. Mark every mock-up and keep the file trail.
- Unchecked AI commentary. A report tells a client a trend that is two data points. Match every sentence to the figures.
For the wider workflow from brief to approved draft, see an AI content workflow for agencies, and for the hub of everything AI-related on the site, the AI guide. If you want help choosing which of these jobs to automate first, that is part of my AI implementation consultation.
Questions agency owners ask before buying
Do we have to tell clients we used AI?
Put it in the contract or statement of work rather than deciding case by case. Say which tools you use, what a person always reviews, and who owns the outputs. Some clients ban AI on their work and some require disclosure on published pieces, so ask at the kickoff and record the answer.
Can we use AI images in finished client ads?
Only after checking the tool's terms for the model that made them. Adobe says outputs from its own Firefly models can be used commercially, and tells you to judge partner-model outputs yourself. Where an ad shows a real product, use real photography so the picture does not mislead.
Is one general assistant enough instead of six tools?
For a very small agency, yes: one business-plan assistant covers copy, research summaries and report drafts. The specialist tools earn their cost on volume jobs, such as resizing across formats and reporting across a dozen clients, where templates and integrations save hours.
How do we stop staff using personal AI accounts on client work?
Buy the approved tools on business plans, tell people which tool does which job, and write one page saying client material goes only into those. A ban without a good official tool just moves the use out of sight.
What should the first purchase be?
Start with the job that repeats most often and is easiest to check, which for most agencies is client reporting or resizing. Copy variants come next. Leave image mock-ups until your rights wording and review habits are in place.
Further reads
- How to Pilot AI in Shadow Mode Before Customers See It — Run a tool alongside your normal work before clients see it.
- How Small Agencies Use AI to Write New-Business Pitches — Using AI to write pitches without losing the agency's voice.
- AI Rollout Plan for a Ten-Person Marketing Agency — A rollout plan for a ten-person agency.
- How Small Translation Agencies Build an AI-Assisted Workflow — How a related agency type builds an AI-assisted workflow.
- Questions to Ask Before Buying AI That Touches Client Data — What to ask any vendor before client data goes in.
- Shadow AI: Is Your Team Using AI Without Telling You? — Finding and handling unapproved AI use in your team.
- How to Mention AI Use in Client Contracts and Proposals — How to mention AI use in proposals as well as contracts.
- AI Ad Copy Generator for PPC Agencies: Responsive Search Ads at Scale — A PPC agency workflow for AI-written responsive search ads: a client brief, formula checks, policy and brand gates, pinning and a test plan.
- AI Media Planning for Small Agencies: Channel Mix and Budget Splits — A prompt for a first-draft media plan, a worked $30,000 split, and five checks that catch bad sums, false sources and channels that cannot spend.
- Ad Agency AI Operating System: A Lightweight Version in Notion — Build a small Notion system for an agency: brand-voice pages, brief templates, agent prompts and approval gates, with a worked brief.
- Best AI for Content Analysis: Audit a Client's Article Library — A five-stage audit of a client's article library: export the URLs, cluster by topic, list gaps and overlaps, then rule on merge, refresh or prune.
- Best AI for Copywriting Ads: Variants, Testing and Client Sign-Off — A small-agency workflow for ad copy: variants by angle, platform character limits, an approval log the client signs, and a testing plan that reads cleanly.
- Create an Advertisement With AI: Brief, Copy, Visuals and Checks — One worked advertisement from brief to final artwork, with a claims check, an image-rights check and platform sizes before anything goes live.
- How to Advertise on ChatGPT: Approval, Budget and Reporting for Agencies — Run ChatGPT ads for a client with sign-off, a capped test budget, tracking and a weekly report, based on OpenAI's own Ads Manager documentation.
- Performance Max Reporting: Explain Asset Results to Clients With AI — Turn a Performance Max asset export into a client explanation that says what the ratings mean, what they don't, and what to change next.
- Ads in AI Overviews: A Pre-Spend Checklist for Ad Agencies and Clients — How Google ads in AI Overviews are bought and reported, plus a client pre-spend checklist, eligibility checks and a sensible test budget.
- Using AI in Advertising: What a Small Agency Can Hand to a Tool — A task-by-task map of what a small ad agency can hand to AI, where each approval gate sits, and the jobs that stay human.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Adobe Firefly plans page; Canva pricing page; AgencyAnalytics pricing page; Fathom pricing page; Anthropic, OpenAI and Perplexity pricing and privacy pages.