AI for Consultants and Small Agencies: Research, Proposals and Client Decks (Without the Deloitte Mistake)

Updated 2026-10-17 · Some links may be affiliate links. Disclosure

In 2025, Deloitte Australia repaid A$97,587 of a A$440,000 government contract after its report was found to contain AI-related errors, including a fabricated quote from a federal court judgment and references to academic papers that don’t exist (AFR; AP News). The report had to be reissued. If a Big Four firm can ship a hallucinated citation, a three-person consultancy can too. This guide sets up AI for the work consultants and agencies do every week, with the checks that keep those errors out.

Ranked picks

Why this order: A general assistant on a business plan does the most jobs for the least money, so start there. Cited research is the next biggest win, because consulting work lives or dies on sources. Decks, call notes and proposal software save real hours but are optional; plenty of solo consultants will be fine with ranks 1 and 2 plus the tools they already use. Start with one or two tools, not five.

Step 1: Settle confidentiality before anyone uploads client files

1. Read your client contracts and NDAs. Some restrict sharing client information with third parties or require notice before you use AI. Your contract terms come first.
2. Use business tiers, not personal accounts. OpenAI says it doesn’t train on ChatGPT Business workspace data by default (OpenAI). Perplexity says Enterprise Pro and Max don’t use customer data to train models (Perplexity).
3. Change Gamma’s default. On individual Gamma plans (Free, Plus, Pro, Ultra), your content can be used to train Gamma’s AI by default. Turn it off under Settings, then Data Controls, before uploading client material. Business and Enterprise workspaces are excluded and the setting is locked (Gamma Help).
4. Write it down. A one-page AI policy (approved tools, what can’t be uploaded, who reviews) is enough for a small firm. Our accounting firm AI policy guide is a template that carries over.

Step 2: Capture discovery calls automatically

Use an AI notetaker such as Fireflies on discovery and status calls so you get a transcript and summary without typing during the call (Fireflies). Tell the client you’re recording and get their OK at the start of the call; some places require every party’s consent. Then paste the summary into your CRM or project notes the same day. More detail in our meeting notetaker guide.

Step 3: Do background research with citations you can open

Ask Perplexity the questions you’d give a junior analyst: market size estimates, competitor moves, recent regulation. Its answers come with numbered source links (Perplexity). Open every source you plan to use and confirm it says what the summary claims. Prefer primary sources (the regulator, the company’s filing, the survey publisher) over blogs that quote them.

Step 4: Draft the proposal from your own best work

1. Save two or three of your past winning proposals (with client names removed) as a reference file in your ChatGPT Business workspace.
2. Paste in the cleaned-up discovery summary from Step 2 and ask for a draft in the same structure: problem, approach, phases, deliverables, timeline.
3. Write scope, pricing and assumptions yourself. That’s where proposals win or lose money, and it’s where a confident-sounding AI guess does the most damage.

Step 5: Turn the outline into a deck

Paste your final outline into Gamma and generate a first draft (paid plans generate up to 100 slides per prompt; Free allows 10) (Gamma Help). Then do the real work: cut slides, swap in your own charts, and apply your brand. Pro adds custom branding, password protection and viewer analytics, so you can see whether the client actually opened it (Gamma).

Step 6: Run the verification gate (non-negotiable)

Before anything AI-assisted goes to a client, one named person checks:

• Every number traces to a source you opened yourself
• Every quote is checked word for word against the original
• Every citation, case, report or paper exists and says what you claim
• Client names and figures are correct and from the right client

This is exactly the check that would have caught the fabricated quote and nonexistent references in the Deloitte report (AP News). After the episode, Deloitte told the team behind it to take additional training on responsible AI use and on reviewing AI output (AFR).

Step 7: Send, track and sign

If you send many priced proposals, PandaDoc’s Business plan adds pricing tables, a content library, branding, CRM integrations and approval workflows; Starter doesn’t include those proposal features (PandaDoc Help; PandaDoc). Only people who send documents need a seat; clients who view or sign don’t (PandaDoc). If you only send a few proposals a year, a PDF and your usual e-signature tool are fine.

Step 8: Be upfront and measure

Add a short line to your statement of work saying how you use AI and that a person reviews all deliverables. After 30 days, compare hours spent per proposal and per deck with your last few projects, and track your proposal win rate over the next quarter.

Quick checklist

• Client contracts and NDAs checked for AI or third-party limits
• Business-tier accounts set up; Gamma’s AI-training setting switched off on individual plans
• One-page AI policy written
• Recording consent asked at the start of every call
• Every research source opened and checked
• Scope and pricing written by a person
• Verification gate signed off before anything goes to a client
• Hours per proposal measured after 30 days

Tools in this guide

• ChatGPT Business: A firm-approved general assistant to replace staff using personal accounts.
• Fireflies.ai: Recapping client calls and advisory meetings.
• Gamma: Turning a finished outline into a first-draft client deck fast.
• Perplexity Enterprise Pro: Market scans, competitor research and background reading where every claim needs a source you can open.
• PandaDoc: Agencies and consultancies that send priced proposals and want to see when clients open them, then collect a signature.