What it used to cost to share knowledge, and what it costs now

What sharing knowledge already costs you, before you buy anything

Thursday, half three. A customer asks whether the extended warranty covers a refurbished unit bought through a reseller. Your rep knows the answer exists. They search the help centre for "warranty", get three articles about registering a new product, none about resellers. They open the shared drive. They check a thread from March. "I'm sure Priya answered this one, was it in the ticket or on Slack?" Priya is on a call until four.

Nine minutes gone. The customer has been told "let me just check that for you" twice. The answer was in a PDF that already existed.

That is what sharing knowledge has always cost, and it has always been paid in salaried hours rather than licence fees, which is why nobody ever approved it. You've got good reps, decent help docs and training that works. The problem isn't competence, it's that knowledge sits in five places and only arrives when a human goes and fetches it. What's changed recently is the price of fixing that. An AI knowledge base for support teams used to be an enterprise line item. It now starts at roughly what your team spends on coffee, and it's worth doing the arithmetic before you take anyone's word for it.

Do the sums on your own team first

Use your numbers, not ours. Here's the shape of it for a support team of eight.

Take a loaded cost of about £22 an hour per rep (salary plus employer NI, pension, tooling). Assume each rep loses 20 minutes a day to hunting for answers they know exist — not learning anything new, just retrieval. That's 2 hours 40 minutes of team time a day, about £59, or roughly £1,240 a month across 21 working days.

Then the interruptions. Your product lead gets asked six times a week, ten minutes a go, and each one breaks whatever they were doing. Call it an hour a week at £35 an hour, plus the recovery time you don't bill. Around £150 a month, and it is the same four questions.

Then the reworked things. The proposal rewritten because nobody found the one from March. The reply drafted from scratch because the rep couldn't locate the version legal already approved.

You're somewhere north of £1,400 a month, and none of it appears on a purchase order. It appears as slower first-response times, as a new starter waiting two days for an answer that was already written down, and as your best-informed person being the bottleneck for everyone else.

The three routes out, and what each one actually cost

Every one of these was a sensible decision at the time. That's worth saying, because the reason they didn't work is structural, not stupid.

The wiki. Cheap to start, and it works for about five months. Then it needs an owner, and the owner is your most knowledgeable person, so the cost of maintaining it is the cost of that person not doing support. Pages go stale quietly. The rep who gets burned once by an out-of-date page stops trusting the whole thing, and goes back to asking Priya.

The intranet or enterprise search project. Scoped in months, priced per seat, and it returns keyword hits. A customer asks about batteries for solar panels, so your rep searches "battery", then "solar panels", then "best battery". The words match. The intention doesn't.

The bespoke build. Genuinely good outcome, if you had a developer to spare and a budget with a comma in it. For a ten-to-fifty-person business that meant it never got built.

Meanwhile the category itself was still informal. Gartner only published its first-ever Magic Quadrant for Knowledge Management Systems for Customer Service on 16 July 2026 [1]. Buyers have been solving this for a decade with no map.

“In the age of AI, trusted knowledge matters more than ever for customer service automation. Our customers confirm that better knowledge, not just better models, is what makes AI dependable in customer service.”

eGain, on being named a Leader in the first-ever Gartner Magic Quadrant for Customer Service Knowledge Management Systems

GlobeNewswire, 20 July 2026

LESSON ONE: If the fix needs a maintainer, price the maintainer, because that is the real cost.

LESSON TWO: Search that matches words instead of intent will always send your reps back to a colleague.

LESSON THREE: Any answer a rep can't verify in one click gets treated as a rumour.

What an AI knowledge base for support teams costs in 2026

£75 a month, unlimited users, no contract, no per-seat maths. That's the floor now for a branded assistant your reps query instead of searching, with every answer cited back to the document it came from.

Compare that against the £1,400-ish above, or against one afternoon of a contractor's time, or against the seat licences you're already paying for a tool three people log into.

The build is four steps, and it's worth knowing what those ten minutes contain rather than taking "live in minutes" on faith:

One, connect a source. Google Drive, Dropbox, Notion, direct file uploads (PDF, Word, Excel, plain text) or a website URL. Point it at your help centre and your product docs folder. Content is imported and counted in page imports, so you decide what goes in rather than dumping the whole drive.

Two, brand it. Logo, colours, theme, so it looks like your company and not a demo.

Three, set who gets in. Public link, anyone with an @yourcompany.com email, or an invite-only list of specific addresses. For a support knowledge base you almost certainly want one of the last two — this is an internal tool your reps type into, not a bot facing your customers.

Four, publish. It's live on a hosted URL, and your team joins with their work email. No SSO project, no seat admin, no IT ticket.

The payback window is short, which is the part most people get wrong when they budget this as a transformation programme.

“Customer service teams are seeing results quickly: 70% of organizations with AI service agents say they observe measurable value within 60 days of deployment.”

Salesforce, State of Service: AI Agents Edition

Survey of 3,075 service professionals, 9 March to 4 April 2026

Where this goes wrong, and the half-hour that prevents it

Cheap deployment does not make your documents good. A knowledge base over bad documents is simply a faster route to a wrong answer, delivered with more confidence than a rep would have used. The old cost doesn't vanish, it moves: somebody still has to decide what's worth reading.

The people closest to the data already know this. In Salesforce's 2026 service research, 72% of service operations professionals said data readiness was a major blocker to AI, against 59% of customer service leaders [2]. The gap between those two numbers is the interesting bit.

So before you connect anything, spend half an hour on this, whether or not you ever buy a tool. Export a CSV of your last 200 tickets. Paste it into ChatGPT or Claude and ask it to group them into the twenty most common questions. Now go through that list and mark which ones have a current document behind them. The unmarked rows are your actual problem, and they're also the shortlist for what to write next.

Two more watch-outs. Content is imported rather than continuously synced, so when you rewrite the returns policy, re-import it — treat it like publishing, not plumbing. And insist your reps click the citation on anything they're pasting to a customer. A cited answer is checkable in one click, which is the whole reason it's usable at all.

Is it worth it if your documentation is genuinely thin? Probably not yet. Fix the top five questions first, then deploy.

The gap between big and small is closing faster than the pricing suggests

In June 2026, 49% of UK businesses with 250 or more employees reported using at least one AI technology, against 28% of those with fewer than ten [3]. That gap was never about capability. It was about who could afford a six-month project and a per-seat contract.

That constraint has gone. Meanwhile customer expectations keep creeping up: the UK Customer Satisfaction Index hit 78.3 out of 100 in July 2026, a fourth successive rise, and very close to its all-time high [4]. Your customers are comparing your response time to whoever answered them fastest this week.

Your next hour is the cheapest one you'll get. Connect your help articles and product docs, gate an AI knowledge base to your support team, and see what your reps stop asking each other by Friday.

Sources

[1] eGain Corporation press release via GlobeNewswire, 'eGain Named a Leader in the First-Ever Gartner® Magic Quadrant™ for Customer Service Knowledge Management Systems', 20 July 2026 — https://www.globenewswire.com/news-release/2026/07/20/3329813/0/en/eGain-Named-a-Leader-in-the-First-Ever-Gartner-Magic-Quadrant-for-Customer-Service-Knowledge-Management-Systems.html

[2] Salesforce, State of Service: AI Agents Edition, 20 May 2026 — https://www.salesforce.com/news/stories/ai-service-agents-improve-customer-satisfaction/

[3] Office for National Statistics, Artificial intelligence in UK businesses: 2023 to 2026, released 20 July 2026 — https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026

[4] Institute of Customer Service, 'Welcome upturn in customer satisfaction – but vigilance needed in a fragile economy', 28 July 2026 — https://www.instituteofcustomerservice.com/backslide-in-customer-satisfaction/