Comparisons

ChatGPT Enterprise or custom systems: what a 20-person firm should buy

Daniel Kerambrum Bicourt · Founder, CO-ORDI · 20 October 2026 · 5 min read

A founder asks us this most weeks: should we just get everyone onto ChatGPT Enterprise? It is the wrong question before it is even answered, because Enterprise is not built for a firm your size, and a seat was never going to be the thing that fixes the work. Here is what Enterprise actually buys, where a 20-person firm sits against it, and what to buy instead.

What ChatGPT Enterprise actually is

OpenAI does not publish a list price for Enterprise. Sales teams quote it directly, and the figures that leak out through procurement and reseller reports cluster between roughly $45 and $75 a seat a month, with most deals reported nearer $60. The terms that keep showing up alongside it are a 150-seat minimum and an annual contract, which puts the entry ticket at somewhere around $100,000 a year before you have automated a single workflow. In exchange you get single sign-on, admin controls, higher usage limits and a promise that your data is not used to train the model. All real. None of it is a system that touches your CRM, your invoicing or your customer records on its own.

Why a 20-person firm hits the floor, not the ceiling

A seat-based minimum built for firms with hundreds of desks does not bend for a firm with twenty. OpenAI's self-serve tier, now called ChatGPT Business, has a two-seat minimum and runs closer to $20 to $25 a seat a month, which is the realistic entry point for a company your size. There is reportedly a newer mid-tier plan sitting between the two, aimed at firms not ready for the 150-seat floor, but the broader point holds: Enterprise is priced and gated for an organisation several times larger than yours. Buying it anyway, to get the admin features, means paying for headroom you will never use.

What a seat actually fixes

A ChatGPT seat makes one person faster at drafting, summarising and thinking out loud. That is genuinely useful, and worth having across the business at Business-tier prices. It does not touch the things that actually cost a 20-person firm money: a lead that sits in an inbox for three days before anyone replies, an invoice chased by hand every month, a report rebuilt from scratch because three systems do not talk to each other. Those are workflow problems, and a chat window open in a browser tab does not solve a workflow problem. It solves a drafting problem. We have written before about what happens when firms mistake the first for the second: see the real cost of DIY automation, where the sprawl comes from the same instinct, reaching for a generic tool and hoping it adds up to a system.

The evidence on buying versus building

A widely cited MIT study on generative AI adoption in business found that the large majority of AI pilots were delivering no measurable return, and that only a small fraction of custom, in-house-built AI tools ever reached production use. The same research found that firms buying from specialist partners and wiring the result into their own workflows succeeded roughly twice as often as firms building everything internally from scratch. The lesson is not "buy more seats". It is that the win sits with external expertise applied to a specific workflow, not with either a stack of raw tools or a bedroom-built prototype. That is the gap between isolated tools and a coordinated system: the tools rarely fail on their own merits, they fail because nobody wired them to the business.

What a 20-person firm should actually buy

Two separate purchases, not one big one. First, a ChatGPT Business seat for anyone who drafts, researches or summarises for a living, because at that price it earns its keep within a week and needs no procurement process. Second, and this is the purchase that actually moves the numbers: one or two custom-built automations wired into the systems you already run, scoped to the workflows that cost you the most hours or the most missed revenue. That is a flat-fee build, not a seat subscription, typically somewhere in the GBP 750 to 2,000 range for a single workflow or GBP 3,000 to 8,000 for a short sprint across several, built once and owned by you afterwards. It is the difference between giving twenty people a faster pen and fixing the process the pen was never going to touch. If you have already tried the seat route and the leak is still there, that is usually the tell that it is time for the second purchase, not a bigger version of the first. See how CO-ORDI scopes that work if you want the detail on how a build like this is priced and delivered.