AI Consultant or DIY: Which Suits a Small Professional Firm?

Do it yourself when the job is staff using a business AI assistant on your own documents under a written policy; most small firms can set that up within six weeks. Bring in a consultant when the work connects systems that hold client data, needs a proper confidentiality review, or would cost more in partner hours than the fee.

The word that decides it is usually "connects". Choosing a chat assistant, writing rules and training staff is well within a partner's reach with a few good guides. Wiring AI into your practice management system, your document store or your client portal is different work: mistakes there expose client files to the wrong people, and they are hard to spot from the inside. Most firms end up splitting the job, doing the first kind themselves and paying for help with the second.

Follow me on Instagram@sagnikteaches

How the two routes compare for a professional firm

For the general version of this decision, outside professional services, see eight signs it is time to get help and whether to hire an automation consultant or build it yourself.

Connect on LinkedInSagnik Bhattacharya

DIY isn't free: the partner-hours sum

In a professional firm, DIY time is usually a partner's or senior manager's time, and that time has a price. The sum:

Subscribe on YouTube@codingliquids

DIY cost = hours needed × the value of those hours, where the value is the charge-out rate if the hours would otherwise have been billed, or zero if they come out of genuinely quiet time.

An illustration with placeholder figures: a partner at a $180 charge-out rate estimates 45 hours to set up Microsoft 365 Copilot properly, including cleaning up file permissions, writing the policy and training eight people. If 30 of those hours come out of billable time, that is $5,400 of fees not earned, plus 15 hours of evenings. If a consultant would do the technical part for a fixed fee lower than $5,400 and the partner's remaining input drops to 12 hours, outside help is cheaper. If the fee is higher, or the partner has a quiet month anyway, DIY wins.

Two caveats keep this honest. Consultants still need your time, often 10 to 20 hours of decisions, access and testing, so don't count the partner's hours as zero. And the value of doing it yourself includes understanding it; firms that outsource everything sometimes can't run what they've been handed. What an AI consultant costs a small business covers how fees are usually structured.

What good DIY looks like in a small firm

Done well, DIY follows a plan with a named owner. A six-week version, filled in as a five-person surveying firm might run it (illustrative):

About 20 hours in total, most of it not billable time anyway. This is the kind of project a small firm should almost always do itself.

A DIY step that goes well with AI help is the first draft of the policy. The prompt, and what came back:

Draft a one-page AI use policy for a five-person surveying firm.
We use [tool] on a business plan. Cover: approved tools, what client
information may and may not be entered, the rule that a qualified
surveyor reviews every AI-assisted report before issue, how to
report a mistake, and who to ask. Plain English, no legal jargon.
Leave a clear placeholder anywhere you'd need a decision from us.

Illustrative output: a tidy page with seven headings, including "Approved tools: [TOOL NAME, PLAN]" and "Client information: staff must not enter personal data of occupants, [DECISION NEEDED: whether property addresses may be entered]". It also included a line saying "the firm complies fully with all applicable AI regulations".

The placeholders were the useful part, because they listed the decisions the partners hadn't made. The compliance line had to go: the firm had not checked anything of the kind, and a policy that asserts compliance nobody verified is worse than one that says nothing.

Jobs that usually need outside help

• Switching on AI that searches all your files. Microsoft 365 Copilot answers from whatever a user already has access to. If permissions are loose, Copilot makes that visible. A realistic DIY failure: a partner switches on Copilot, and within a week a junior asks it for "last year's fee income" and gets a summary of the partners' drawings spreadsheet, which had been shared with "everyone" years earlier. The fix is a permissions clean-up before switch-on, which is fiddly and worth paying for in a firm with years of shared folders. You can test your own exposure before deciding. Give a test account the same access as your most junior member of staff, switch Copilot on for that account only, and ask it five questions: "show me salary information", "partners' drawings", "any disciplinary notes", "last year's fee income by partner", "documents containing passwords". If any answer surfaces something a junior shouldn't see, the permissions need work before anyone else gets a licence, and the length of that list tells you whether it's an afternoon's job or a consultant's.
• Connecting AI to a practice management system or client portal. Automations that read client records and write emails need testing against real edge cases: the client with two contacts, the opted-out client, the file with no email address. A DIY attempt at an illustrative accountancy practice shows how this goes. Its automation drafted a year-end reminder for every client with a deadline in the next 30 days, and the first run looked fine. The second run sent an AI-drafted email to a client who had asked for no AI on their work, because the opt-out was recorded in a free-text notes field the automation never read. The fix was a proper tick-box field and a filter on it, and a test list of awkward records run before every change: the opted-out client, the couple with one shared email address, the client with no email, the company whose contact left last month.
• Client-facing AI. An enquiry chatbot or an out-of-hours assistant speaks for the firm. Scripting what it must never say, and testing it, benefits from someone who has done it before.
• A formal risk review. If a large client, your insurer or your professional body wants to see how you assessed AI risk, an outside review carries more weight than your own notes.

What a good consultant engagement looks like

If you hire, the engagement should be scoped in writing before work starts, with deliverables you can check. A filled-in scope, illustrative, for a fourteen-person architect practice rolling out Copilot:

SCOPE OF WORK
Objective: Microsoft 365 Copilot live for 8 named users, with file
  permissions cleaned up first, by [date].
In scope:
  1. Audit SharePoint and OneDrive sharing; report of over-shared
     sites and files, with recommended fixes (practice approves each)
  2. Apply approved permission fixes
  3. Configure Copilot for 8 users; switch off features not approved
  4. Two 90-minute training sessions on the practice's own projects
  5. Written handover: settings changed, how to add/remove users,
     how to read the Copilot usage report
Out of scope: custom agents, other software, ongoing support
Practice provides: admin access, a decision-maker for approvals
  within 2 working days, 8 hours of staff time for training
Fee: fixed, [amount], 50% on start, 50% on handover
Ownership: all settings, documents and credentials belong to the
  practice; consultant access removed at handover

Ask for the independence answer in writing alongside the scope. A clear one reads something like: "I receive no referral fees, commissions or discounts from any vendor I might recommend. I hold partner status with one software company, which gives me training access and nothing tied to what clients buy." An answer that talks about "preferred technology partners" without saying whether money changes hands is the one to press on.

The last line of the scope matters most. A firm should never end up with automations running on a consultant's account. An AI consultant handover checklist lists what to collect before they leave, and what an AI consultant can't do for you covers the decisions that stay with the partners however much you pay.

The split route: DIY first, a paid review before anything connects

Many small firms land on a middle path that costs little and removes most of the risk. They do the DIY plan themselves (tool, policy, training, a pilot on drafting) and pay for a short, fixed-scope review at one specific moment: just before they connect AI to anything that holds client data, or switch on a tool that searches their files.

A review like that looks at three things. Who can see what in your file stores and systems, and what an AI tool would therefore be able to surface. Where client data will travel once the connection is live, including the AI provider underneath the tool. And whether the vendor's terms match what you have promised clients. Bring your AI policy, your tool register, a list of systems with their administrators, and the pilot log; with those in hand, a review is short. The systems list is the one firms usually don't have. Filled in for an illustrative nine-person law firm, it fits in five rows:

Written out, the review's scope is obvious: two rows have a planned connection, so those are what the reviewer examines. The "Outside IT support" entry is a finding in itself, because whoever administers Microsoft 365 has to be in the room when Copilot permissions are changed.

The split works because the expensive mistakes in professional firms rarely happen during drafting pilots. They happen at the moment of connection, when a setting that looked harmless exposes years of shared folders or sends client records somewhere nobody checked.

Signs your DIY effort has stalled

• The pilot was due to finish six weeks ago and nobody has reviewed the log.
• Staff have drifted back to personal AI accounts because the business plan "doesn't do what I need".
• The partner leading it has been saying "next month" for three months.
• You have bought a tool but nobody has read its data terms.
• Different partners have set up different tools, and nobody has a list.

Two or more of these usually means the problem is time and ownership, not knowledge. That can be fixed internally by giving someone protected hours, or it can be the point where outside help earns its fee simply by making the project someone's actual job.

An illustrative restart: a seven-person consultancy recognised three of the five signs at its quarterly partners' meeting. Rather than hire anyone, it moved ownership from the busiest partner to the office manager, with three protected hours a week, and cut the project to one workflow, proposal first drafts, with a new pilot end date six weeks out. It also asked each partner to list the AI tools they personally pay for, which turned up two unapproved subscriptions and one client folder shared into a personal account. The pilot finished a week late, which counted as a success against the previous three months of nothing.

Two firms, two answers

The five-person surveying firm went DIY. Its AI use was drafting reports and emails from site notes, on a business plan, with nothing connected to its client records. The six-week plan above took about 20 hours, mostly from non-billable time, and the pilot cut first-draft time on standard condition reports roughly in half. Paying a consultant for that would have bought speed it didn't need.

The fourteen-person architect practice hired help, for part of the job. It wanted Copilot across its SharePoint project archive, which held fee schedules, staff records and fifteen years of client correspondence with inconsistent sharing. The partners wrote the AI policy and ran the training themselves, and paid a fixed fee for the permissions audit, the clean-up and the configuration, the scope above. The partner's own time came to about 14 hours instead of the 50 or so they had estimated for doing it all themselves.

Put through the partner-hours sum from earlier, with illustrative figures: 36 hours saved, of which about 25 would have come out of billable time at a $150 charge-out rate, is $3,750 of fees kept. A fixed fee below that was cheaper than DIY on the numbers alone. The partners judged it worth paying even at a little above that figure, because the permissions clean-up was the step where a mistake would have put fee schedules and staff records in front of the whole office.

If your firm sits between these two and you'd like a second opinion on which parts to keep in-house, that is exactly what my AI implementation consultation is for. Either way, write down which jobs you are doing yourselves and which you are paying for, so the split is a decision rather than something that happened.

What partners ask before deciding

How do I know a consultant is independent of the tools they recommend?

Ask directly whether they receive referral fees, commissions or partner discounts from any vendor they might recommend, and ask for the answer in writing. A good consultant will tell you, and will explain why a tool suits you in terms of your needs and your existing systems, not the tool's features alone.

Can we start DIY and bring someone in later?

Yes, and it is often the best order. A firm that has run a small pilot, written its rules and logged what worked is a much easier client: the consultant spends less time discovering and more time on the hard part. Keep notes of what you tried and why, so none of it has to be repeated.

What should we prepare before a first call with a consultant?

A list of the systems you use and who administers them, the three jobs you most want help with and roughly how many hours they take, any client or insurer constraints on data, and what you have already tried. Half an hour of preparation usually saves a paid session of discovery.

Further reads

• How to Choose an AI Consultant: 20 Questions to Ask First — Twenty questions to ask before you hire anyone.
• AI Consultant Red Flags: 12 Warning Signs to Walk Away From — The warning signs that should end a conversation.
• What an AI Consultant Needs From You: Access, Data, and Time — The access, data and time a consultant will ask for.
• AI Strategy vs AI Implementation Consultant: Which Do You Need? — Which kind of consultant fits the problem you have.
• How to Run Your First AI Pilot Project in a Small Business — The DIY route, step by step, for your first pilot.
• Should You Sign an NDA Before Sharing Data With an AI Consultant? — Protecting client information before a consultant sees it.
• What an AI Consultant Does for a Law Firm, and What It Costs — The five phases of a law-firm AI engagement, what each should hand over, how fees are built from consultant days, and proposal red flags.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Microsoft 365 Copilot Business pricing and documentation on how Copilot uses existing permissions; OpenAI and Anthropic business plan pages. Checked September 2026. Hourly rates, hours and firm details in the examples are illustrative.