Most firms picked their AI tool the way they picked their first practice management system: whoever got there first won. Someone signed up for ChatGPT, someone else switched on Copilot, and now the whole team uses whichever tab happens to be open for everything from client emails to month end analysis.
Here is the problem. AI models are not interchangeable. Each one is trained, tuned and packaged for different jobs, and the gap between the right tool and the wrong one shows up in accuracy, time saved and, increasingly, in your software bill.
So before you roll out another licence, it is worth asking a simple question. Are you using the correct AI model for the job in front of you? Let’s look at the main contenders, where each one shines, where each one stumbles, and the moments when you don’t need AI at all.
Gemini: the visual, multimodal all rounder
Google’s Gemini is built to handle text, images, audio and video in a single model, and it sits right inside Gmail, Docs, Sheets, Slides and Meet. If your firm already lives in Google Workspace, it is often included in your Business plan with no extra licence.
Where it shines. Gemini is strong at anything visual. Google has wrapped image and video creation tools around it, so it is a natural choice for social graphics, event visuals and quick presentation assets. Its very large context window also means it can read long documents, recordings and image heavy files in one go. For Workspace customers, prompts and responses are not used to train the model.
Where it falls short. The product line changes fast and the naming is confusing, so it is hard to know which version your team is actually using. Most UK businesses run on Microsoft 365, which means Gemini’s biggest advantage, its Workspace integration, simply does not apply. Long, nuanced written copy can also feel more generic than its rivals.
Best for: visual content, multimodal tasks and firms already on Google Workspace.
ChatGPT: the Swiss Army knife
ChatGPT is the most widely used AI tool in the world, and it has grown from a chatbot into a work platform with file analysis, image creation, voice and agent features. Business plans start from around 25 US dollars per user per month and do not use company data for training.
Where it shines. Breadth. ChatGPT will brainstorm a campaign, analyse an uploaded spreadsheet, generate an image and draft a client letter in the same conversation. Custom GPTs let you package a repeatable prompt for your team, and because almost everyone has tried it, adoption is easy.
Where it falls short. Being good at everything means it is rarely the best at any one thing. Its copy can slip into a recognisable “AI voice” that clients are starting to spot. It can state a wrong figure with total confidence, which is a real risk when numbers matter. The biggest issue for firms is shadow AI: staff pasting client data into personal accounts that sit outside your governance entirely.
Best for: ideation, general research, quick data exploration and teams taking their first steps with AI.
Claude: the writer and the deep thinker
Anthropic’s Claude has built its reputation on careful reasoning and natural writing. It can work through very long documents in one go, business data is not used for model training, and it is increasingly used for complex, multistep work and agents that run recurring workflows.
Where it shines. Copy. Claude tends to produce writing that sounds like a person rather than a press release, and it follows tone of voice and style rules closely, which makes it a favourite for blogs, proposals and client communications. It is also strong at reading a lengthy engagement letter, policy or set of accounts and reasoning through it step by step, and it is widely used for writing and checking code.
Where it falls short. Claude does not generate images natively, so it is not your design tool. Its consumer ecosystem is smaller than ChatGPT’s, and heavy users can hit usage limits on lower tier plans. It can also be more cautious than rivals on some requests, which occasionally means an extra prompt to get what you need.
Best for: long form copy, document review, nuanced analysis and building agents for multistep tasks.
Microsoft Copilot: the one already in your inbox
Microsoft 365 Copilot works directly inside Word, Excel, PowerPoint, Outlook and Teams. Its biggest lever is access to your own firm’s documents, chats and calendars through the Microsoft Graph, and it now offers a choice of underlying models rather than relying on one.
Where it shines. Context. Copilot can summarise last week’s Teams call, find the email where a client agreed a fee, or draft a reply using the attachment already in the thread. For firms on Microsoft 365, there is no new tool to learn and your existing permissions and security settings carry across.
Where it falls short. It is a paid add on licence per user per month, on top of Microsoft 365, and many firms are paying for seats that barely get used. It is only as good as your data hygiene: if your SharePoint is a maze, Copilot will happily surface the wrong version of a document. Its writing and Excel analysis are solid rather than exceptional, and it is not built for creative design work.
Best for: everyday admin inside Microsoft 365, such as email triage, meeting summaries and finding information across your own files.
Agents inside your tools: AI where the work already lives
Almost every platform your firm pays for now has its own AI built in. monday.com is a good example: its Sidekick assistant works inside your workspace, respects existing permissions and can connect to your other work apps, while monday agents act as workers that handle real tasks across departments.
Drowning in manual, repetitive work? Tell us the task and we’ll show you what to automate.
Where it shines. Context without the copy and paste. An agent inside your project or CRM platform can already see the board, the deadlines and the client history, so it can chase an overdue task or draft an update without anyone briefing it. They are usually set up with a no code builder, so operations teams can create them without IT.
Where it falls short. Each agent only sees its own tool. Your monday.com agent has no idea what is happening in your practice management software, your inbox or your bookkeeping platform. Costs are moving to credit based models, with monday.com now charging credits for Sidekick and AI workflow runs, which makes monthly spend harder to predict. And when every vendor adds an agent, you can end up with a dozen disconnected assistants and no one owning them.
Best for: task and project updates inside a single platform your team already uses every day.
Why the right model matters
Matching the tool to the task is not about chasing the newest release. It is about getting a better answer first time, and that pays off in four ways.
Better output, fewer rewrites. A model that suits the job gets you to a usable draft faster. Ask a design focused tool for a social graphic and a writing focused tool for the blog that goes with it, and you skip the rounds of “close, but not quite”.
Lower risk. Choosing tools deliberately means choosing where your client data goes. A firm that approves the right business tools, with training switched off and proper admin controls, removes the temptation for staff to reach for personal accounts.
Smarter spend. Many firms pay for overlapping licences that each get used for a fraction of their potential. Mapping tasks to tools shows you where a seat is earning its keep and where it is not.
Faster adoption. When people know which tool to open for which job, they stop experimenting and start getting value. Clear guidance beats a free for all.
As a starting point, here is how the tools line up against everyday firm tasks.
| Task | Best fit | Why |
|---|---|---|
| Blog posts, proposals, client letters | Claude | Natural tone, follows style rules closely |
| Social graphics and visuals | Gemini | Strong image and video tools |
| Brainstorming and general research | ChatGPT | Breadth and flexibility |
| Email triage and meeting summaries | Copilot | Sees your own Microsoft 365 data |
| Reviewing long contracts or policies | Claude or Gemini | Handle very long documents well |
| Clinical notes and letters | Heidi | Built for that one workflow |
| Chasing tasks inside a project board | In tool agent | Already has the context |
When you don’t need AI at all
Here is the part the AI hype tends to skip. Plenty of the work slowing your firm down does not need intelligence. It needs consistency. If you can write the rules down, automation will usually do the job better and cheaper than any AI model.
Think about the tasks your team repeats every day: matching bank transactions, keying invoice data into your ledger, moving client details from an onboarding form into your practice management system, generating the same monthly reports, chasing missing documents on a fixed schedule. These follow the same steps every time. They do not need a model to interpret anything.
Automation wins on result. A rules based process does exactly the same thing every time, with no creative interpretation and no risk of a confidently invented figure. Every step is logged, which makes it easy to audit, and that matters when accuracy and compliance are non negotiable.
Automation wins on cost. AI tools charge per seat, per message or per credit, and every run needs someone to check the output. A well built automation runs thousands of times in the background for a predictable cost, without anyone reviewing each result. Using AI for a task that never changes is like hiring a consultant to do your filing.
The best results combine both. The smartest workflows use automation for the predictable steps and call on AI only where judgement or language is genuinely needed, such as reading an unstructured email or drafting a personalised client update. That blend keeps costs down and puts AI exactly where it adds value.
The catch is ownership. Automation only delivers if someone builds it properly, monitors it and fixes it when a system changes. That is exactly why we built our Automation-as-a-Service model: we build, run and maintain the automation for you, so it keeps working long after launch.
What to take away
There is no single best AI model, only the best model for the task in front of you. Gemini brings the visuals, ChatGPT brings the breadth, Claude brings the words and the reasoning, Copilot brings your own Microsoft 365 data, and in tool agents bring context inside the platforms you already use.
The firms getting the most from AI are the ones that map their tasks first and pick tools second. They give their teams clear guidance on what to use and when, they keep client data inside approved business tools, and they are honest about the jobs that never needed AI in the first place. For those, automation is quicker, cheaper and more reliable.
Not sure where AI fits in your firm?
We help accounting firms work out which tasks need AI, which need automation and which need neither. Get in touch with the bots for that team and we will help you map your processes, cut the licences you don’t need and put the right tool on every job.
