Directly put the customer on the payroll

There is a peculiar hiring problem hidden inside customer support. The person who can explain the product may already be somewhere else: using it, discussing it, or helping a stranger get it working. Directly’s proposition was to bring that person into the support operation without first bringing them onto the employee roster.

Consider the distinction between knowing a troubleshooting script and knowing why a product behaves oddly. A frequent user accumulates the second kind of knowledge almost accidentally. Directly organized a business around finding those users, checking their abilities, and paying them to help. The customer became a supplier of expertise. A company’s fans acquired something rather more practical than a badge.

The expert was already in the room

Directly was co-founded by Antony Brydon, Jeff Patterson and Jean Tessier in 2011. These were builders with experience beyond support software. Patterson’s company biography credits him with publishing music on the web in 1993. Tessier created Dependency Finder, an open-source tool for analyzing Java code. The team’s interests suggest a useful recurring question: how do you make specialized knowledge travel?

By 2015, investor Costanoa was describing an enterprise service that routed questions to expert users on their smartphones. Its investment essay named Pinterest, Airbnb, Udemy, Lyft and Republic Wireless, and put the number of resolved inquiries at nearly 500,000. That was the early commercial evidence: companies were willing to have people outside the ordinary support workforce answer their customers.

The appeal becomes clearer in technical products. Directly has named Microsoft, Samsung and Autodesk among the companies using its experts. Autodesk’s users include people who work with its software professionally. A person who has encountered a problem in actual practice can offer something a generic service script struggles to supply: an explanation with experience behind it.

Recruitment, however, is more demanding than collecting enthusiasts. OnDemand describes skills assessments, writing tests, language requirements and checks for empathy. Approved experts complete customized training on the platform and customer service. Knowledge must survive the journey into another person’s inbox. An excellent troubleshooter who cannot explain a fix is an expensive way to produce a second question.

One answer, more than one payday

The reward system is where Directly becomes especially interesting. Its older explanation described compensation for resolving questions, writing automated answers, reviewing other experts’ responses, completing special projects and working during demand surges. Task rewards typically ranged from $2 to $60 in that account. Those were payments to experts, not a price list for enterprise customers.

There was also a reason to write beyond the immediate conversation. If an expert authored an automated answer that successfully resolved later questions, the author could receive further rewards. Peer review had rewards too. In principle, that gives the network a financial interest in producing explanations worth keeping. A carefully written answer can continue doing useful work after its author has logged off.

OnDemand’s reward description also connects reputation to future work. Customer and peer ratings help determine priority routing, alongside expertise, language and response speed. The service sells enterprises a managed combination of people and technology. It gives experts flexibility over their participation, while promising buyers that compensation follows outcomes. The attraction is easy to understand; the operational challenge is defining a resolved issue honestly.

Directly describes its experts as unscripted. That is an appealing promise, but it increases the importance of selection and review. A natural voice can make an explanation easier to follow. It can also deliver a confidently wrong instruction. The network needs more than charm: it needs a way to spot poor answers and direct work toward people who reliably help.

A good answer cannot issue a refund

For a support leader, Directly fits into the existing service stack. Its web and email offering connects to helpdesks, answers repetitive inquiries automatically, and routes selected tickets to experts. This is a practical use for a business with recurring product questions and uneven demand. The customer asks for help through an established channel; the company decides which work can be handled elsewhere.

The boundary matters. Directly’s web-support description leaves complex account and billing issues with company agents. Knowing how a product works does not confer permission to change a customer’s account. An expert may explain a setting perfectly and still be unable to authorize the action the customer actually needs.

That distinction helps explain the November 2019 acquisition of Kylie.ai. Directly said the acquired technology would extend its capabilities into refunds, returns, cancellations and order tracking, linking conversations with back-end execution. The announced ambition was to connect an answer to a transaction. An interface that sympathizes with a missing order is only halfway to being useful.

Directly also resisted treating every automation supplier as an enemy. Its 2020 partner ecosystem named Meya, Percept.ai and SmartAction as participants. The implication for buyers was that expert knowledge could sit behind different conversational tools. The commercial competition therefore included outsourced contact centers, in-house teams and self-service systems, while helpdesk platforms could also be integration partners. Purchasing support is rarely a tidy contest between two identical products.

“The AI makes the experts more effective, the experts make the AI more effective, and the result is better answers.”Antony Brydon / April 2018

The business took two roads

Investment followed that combination of software and expertise. In April 2018, Directly announced a $20 million Series B led by Northgate Capital, with Microsoft Ventures, Costanoa and True Ventures participating. In May 2020, it announced another $11 million, extending the January financing to $31 million. Microsoft and Samsung appeared in the story as customers and strategic backers - a useful overlap between the people buying the service and the people financing it.

Then came a change that any present-day account needs to preserve. On March 15, 2022, a CSS Corp company, OnDemand Answers, Inc., purchased the OnDemand business. The announcement described an expert network across more than 60 countries and integrations including Microsoft Dynamics, Zendesk, Oracle and Salesforce. CSS Corp, now Movate, could combine that freelance capacity with its full-time workforce.

The original Directly company announced a separate focus on Directly Darwin, its automation effort. The transaction was a business-unit sale. Consequently, the venture-backed company’s history and the OnDemand service’s present ownership cannot simply be folded into one undifferentiated corporate biography. Directly’s website now presents OnDemand as part of Movate.

Even an elastic workforce has a server bill

Movate’s June 2025 account of modernizing OnDemand supplies a less glamorous sequel. Rising costs and fragmented deployment tools prompted a move from AWS to Azure. The company reported 30% lower operating costs and faster release cycles. These are Movate’s reported outcomes, not independent measurements.

The account includes setbacks. An initial Memcached deployment showed very high memory use even under low traffic. Concurrent database transfers hit VPN limits; another sync was interrupted by a source-database restart. The remedies included a different cache deployment, sequenced transfers, more bandwidth and tighter change control. The transferable lesson is procedural: map dependencies and control the cutover before treating a cloud move as a cost-saving exercise.

Borrow the incentive, mind the boundary

The most useful thing to copy from Directly is a sequence. Find the product knowledge that already exists. Test whether its holders can communicate. Choose questions they can answer without privileged account access. Reward successful resolutions, then review the answers that deserve reuse. Measure whether the customer’s problem stayed solved, rather than congratulating yourself because a ticket disappeared from a queue.

The approach needs enough suitable inquiries and enough qualified people to make coordination worthwhile. If most cases require account authority, if product knowledge goes stale faster than it can be reviewed, or if a company cannot maintain dependable escalation, a freelance expert network will leave much of the job unfinished. Those are implications of the model’s published boundaries, not a verdict on every implementation.

Directly’s enduring idea is disarmingly social. Businesses spend considerable effort teaching employees what their customers sometimes know already. Giving those customers a route into paid support can make that knowledge useful elsewhere. The bot, the helpdesk and the contract all matter. So does the person who has seen the problem before - and can finally be paid to explain it.