The cost of a knowledge gap: how to put a number on missing knowledge

What does a knowledge gap actually cost?

A knowledge gap costs money every time a customer asks the question it leaves unanswered. The agent searches, puts the customer on hold, asks a colleague, transfers the call, promises a callback or gives an answer that brings the customer back next week. Each of those steps is time that wouldn't have been spent if a good article existed.

The cost per conversation is usually small, a few minutes and a few dollars. It becomes significant because it repeats on every affected conversation, every month, until someone fixes the article. That is also why a gap backlog is so easy to ignore: no single contact looks expensive.

Putting a number on each gap does two jobs. It orders the backlog, so the knowledge team works on the gaps that cost most. And it turns a content problem into a business case that operations and finance leaders recognise. The pillar knowledge gap analysis guide introduces a one-line cost model. This guide breaks it into parts so you can build a figure you can defend.

Which costs should you count?

Count the extra work a gap causes on the conversations it affects. Four components cover most of it, and each can be measured from conversation data:

Two further components apply in some operations. If customers tried self-service first and then contacted you because it couldn't answer, the difference between an assisted and a self-service contact belongs to the gap. Gartner puts those at $13.50 and $1.84 on average (Gartner, 2024). And on service desks, where tickets are the unit of work, the equivalent of handle time is the extra effort and elapsed time before a ticket resolves; finding knowledge gaps in Freshservice shows how escalations and SLA risk change the picture.

How do you set the cost per minute?

Use a fully loaded cost per productive agent minute. That means agent wages plus on-costs such as superannuation and leave, a share of team leader and support salaries, technology, facilities and overheads, divided by the minutes agents actually spend handling contacts. Finance teams often hold this figure already for workforce planning or outsourcing comparisons.

Two cautions. First, a cost per minute based on wages alone understates the true cost, often by a wide margin, and makes every gap look cheaper than it is. Second, extra handle time only becomes a cash saving when it changes staffing, overtime or outsourced volume. Below that point it shows up as capacity: shorter queues and more time for other work. Both are real value, and it helps to say which one you are claiming.

Agree the cost basis with finance once, document it, and use the same figure for every gap. The ranking matters more than the precision of any single number, and a consistent basis keeps the ranking fair.

What does a worked example look like?

Take a fictional energy retailer, with invented numbers throughout. Its biggest gap last month was customers asking whether their concession carries over when they move house. No article answered the combined question. Here is what the conversations showed:

• 800 conversations affected in the month.
• Average handle time of 10.5 minutes on those conversations, against 7 minutes for similar billing contacts. That is 3.5 extra minutes.
• 25% transferred to a specialist team, with a second leg averaging 6 minutes.
• 18% of customers called back on the same topic within seven days.
• 10% ended with a promised callback, averaging 5 minutes to research and make.

The retailer's fully loaded cost is $1.20 per productive minute, so a typical 7-minute contact costs $8.40.

That is about $78,000 a year for one missing article. Writing and approving it might take a day. Even if the fix only recovers half of the cost, because some customers will still need help, the article pays for itself many times over.

Notice which component dominates. Extra handle time is half the total, which is typical: the hold and the search happen on every affected conversation, while transfers and repeats happen on a share of them.

Why isn't the most-asked gap the most expensive?

Because friction varies far more than volume. The same retailer has a second gap: customers asking how to read the new bill layout. It affects 2,000 conversations a month, two and a half times as many. But the existing billing article covers most of it, agents explain it quickly, and almost nobody calls back.

• 2,000 conversations × 0.5 extra minutes × $1.20 = $1,200.
• No transfers and no callbacks.
• 2% repeat contacts: 40 × $8.40 = $336.

Total: $1,536 a month, less than a quarter of the concession gap. Ranked by volume, the bill layout question goes first. Ranked by cost, it goes well down the list, and the first day of writing goes to the article that recovers the most money.

This is the main reason to cost gaps at all. Volume tells you what customers ask about. Friction tells you where the knowledge base is failing them. The cost combines the two.

Which costs are hard to put a number on?

Some of the most serious effects of a knowledge gap don't fit a per-minute model. Record them beside the cost as flags, so they can lift a gap's priority without inflating the headline number:

• Compliance and conduct risk. A wrong answer about hardship, cooling-off rights, eligibility or disclosure can matter far more than its handle time. Flag regulated topics so a low-volume gap there is reviewed early.
• Complaints. Inconsistent answers produce complaints, and complaints carry their own handling cost and, in some industries, external dispute costs.
• Customer effort and churn. Customers who need several contacts to resolve something are less satisfied and more likely to leave. The link is real, but attributing churn to one gap is guesswork.
• Agent experience. New agents take longer to become productive when the knowledge base can't be trusted, and experienced agents stop using it, which hides further gaps.

What mistakes distort the number?

• Counting the whole topic. Only count conversations where the knowledge base failed, and don't use failed searches as a proxy for them; most affected conversations never produce one. If agents handle most of a topic well, those contacts carry no gap cost.
• Using the centre-wide handle time as the baseline. Compare with similar contacts on the same queue. A complex product queue is always slower than the average.
• Double counting. A transferred contact that also generates a repeat contact should count both, because both happened. A repeat contact should not be counted again as a new affected conversation in the same month.
• Assuming full recovery. A fix rarely removes all the friction. State the recovery you are assuming and check it after the article is published.
• Leaving the estimate static. Gaps grow and shrink with seasons, launches and policy changes. Recalculate monthly, and watch for new gaps that are growing quickly while they are still cheap to fix.

How does Knowledge Quest rank gaps, and where does your cost model fit?

Knowledge Quest reads every conversation from your contact centre platform or service desk, works out what each customer wanted and checks it against your knowledge base. It ranks gaps by impact: how many conversations hit each gap, weighted by how hard those conversations were, using signals such as transfers, callbacks, unresolved contacts and negative sentiment. That priority is calculated by rules from the conversation data, so it is consistent and auditable, and it tracks the same volume-and-friction logic as the dollar model above.

The dollar figure stays yours. Each knowledge area in Knowledge Quest can hold its own cost per interaction and cost per callback, so the numbers you agreed with finance sit beside the conversations they apply to. Use them with the model in this guide to turn the top of the impact-ranked list into a business case, and every count you use links back to the conversations it came from. Knowledge Copilot then drafts the article or update from those conversations, and your team approves it before it goes to your knowledge base. After it is live, compare the same measures for the topic, like for like, to see whether the friction fell. The guide to measuring article impact sets out a fair before-and-after.