An AI product management consultant does two jobs from one seat: deciding what a product should do next, and deciding which of the work around it can be handed to AI agents under human approval, which is why Janaka’s first call is free and his mapping sprint is US$2,300 (LKR 756,125). specshop.dev is an AI consultancy and AI agency in Colombo, Sri Lanka, led by Janaka Ediriweera, Principal AI and Product Management Consultant.
Every US dollar figure below converts at LKR 328.75, the Central Bank of Sri Lanka’s indicative spot rate for 10 September 2026. Every engagement described here is hypothetical and labelled as such, because the real ones are not ours to publish. This is commercial information, not legal, tax or financial advice.
The morning two decisions turned out to be one
Take a hypothetical firm, eighteen staff, selling maintenance contracts to offices around Colombo. The board wants a customer portal, because the competitors have one and the word sounds like progress.
Two weeks of counting hours says something less flattering. What customers actually ring about is where the technician is and when he will arrive. What staff lose their mornings to is retyping job sheets into invoices, twice, because the second typing corrects the first.
So the product decision is a status page rather than a portal. And the automation decision is job sheet to invoice, drafted by an agent, approved by a named person in finance before anything is sent. Those look like two decisions belonging to two different suppliers. They are one decision, taken on one set of evidence, in one afternoon. Split them and you get the predictable Colombo outcome: a portal nobody logs into, and a clerk still retyping at nine in the morning.
That is the moment Janaka’s title got its second half.
The two halves, and why neither works alone
The product half is old and well defined. Silicon Valley Product Group, Marty Cagan’s firm, states the job plainly: as product manager “you’re responsible for ensuring that what gets built is both valuable and viable”. It adds the distinction most teams get wrong, that “feature teams deliver output, but product teams deliver outcomes”. Valuable, viable, and an outcome rather than a shipped list. None of that has changed.
The AI half is newer and narrower than the market pretends. It is not model selection. It is deciding which repeatable work can be handed over safely, in what order, and with which approvals, which is the whole of what Janaka does when firms hire an AI consultant in Colombo. The order matters more than the tooling. Automate the invoicing before you have agreed what an invoice must contain and you have industrialised an argument.
Neither half works alone. A product manager with no view of what agents can absorb writes a roadmap costed against a team that no longer needs to exist at that size. An automation specialist with no product judgement makes the slowest workflow faster and leaves the wrong workflow in place. Janaka wrote about the first failure at length in I’m a Dead PM. AI Killed Me., where the point was that the coordination layer, the Jira grooming and the handoff management, is what actually got automated, not the thinking.
RMIT Activator, and what 150 startups taught Janaka
RMIT Activator describes itself as “RMIT’s centre for entrepreneurship”, working with “aspiring founders, startups and organisations who are driven to create meaningful change”, and states its mission as “to deliver impact through entrepreneurship”. Janaka trained more than 150 startups there in Melbourne. He is not going to name any of them, and the useful residue is patterns rather than stories.
Three patterns held almost every time.
Founders describe solutions, not problems. Ask what the product does and you get a feature list within ten seconds. Ask who is currently doing this by hand, how often, and what it costs them when it goes wrong, and you get a pause. The pause is where the work is.
The scarce resource is never engineering. It is the decision about what not to build, and that decision has an owner in perhaps one team in five. Everywhere else it is taken by whoever spoke last in the meeting.
Written beats verbal, by a distance that surprises people. Nine years of spec-first delivery have made him boring on this point. A one-page specification, agreed before anyone builds, is not bureaucracy, it is the cheapest place to have the fight. It is also the input an agent needs, a coincidence he did not plan and has been living off since.
Twelve years, 120+ products shipped across five countries, and the biography behind that is on the Janaka Ediriweera page rather than in this paragraph.
Product decisions an agent cannot make
Agents are extremely good at the middle of the job and useless at the ends. Here is what does not delegate.
- Which customer to disappoint. Every roadmap disappoints somebody. Choosing who is a commercial judgement with a name attached to it.
- What not to build. A model asked to expand a backlog will expand it. It has no incentive to say the quiet thing, which is that six of these nine items are not worth the maintenance.
- What the price is. Pricing is a position in a market, taken with incomplete information and defended for a year.
- What “done” means. Somebody has to define the acceptance test before the work starts, and be wrong about it in public when it does not hold.
- Where the approval line sits. Which drafts a human must read before they leave the building is a risk decision, not a settings toggle. It is the difference between an agent and an accident.
The mechanism Janaka uses to keep that honest is approval rate: the share of drafts a named reviewer at your firm approves untouched. Your people produce that number, which is exactly why he cannot polish it. If it has not improved on its day-one baseline by day 30 when you hire an AI agent, you do not confirm the hire and the balance is never invoiced.
A week on an engagement, hypothetical
Composed for illustration, not reported from a client. Monday is hour counting: Janaka sits with the three people who do the work and count, in minutes, what the busiest workflows actually consume. Tuesday is the disagreement, because the partners or the founders rarely agree on which workflow is the expensive one, and an hour count settles it faster than a workshop. Wednesday he writes the specification, one page per workflow, including what the agent must never do. Thursday is the approval design: who reads what, at which volume, and what happens at midnight when that person is asleep. Friday is the product half, the part clients did not think they were buying: given all of the above, what should the product do next, and what falls off the list.
The deliverable is written and it is short. If it needs a deck, he has not decided anything.
What the engagement looks like, by shape
| Engagement | The product half | The AI half | Deliverable | Price shape |
|---|---|---|---|---|
| Pre-launch startup, two founders | What the first version must do, and the six things it must not | Which founder tasks an agent can draft so two people ship like five | One-page spec, a cut list, an approval line | Free thirty-minute call, then US$2,300 (LKR 756,125) mapping |
| SME automating back office, 10 to 50 staff | Which workflow is worth fixing, ranked by counted hours | The agent, its reviewer, and the 30-day probation | Workflow map with staff hours, then a live agent | US$2,300 sprint, agents from US$1,225 a month (LKR 402,700) plus one-off US$9,500 (LKR 3,123,125) |
| 30-engineer product team | The specification nobody has written, and the release it belongs to | Where agents sit inside the product, and where they must not | Written spec, sequencing, an engineer inside the team | Embedded engineer from US$15,000 for three months (LKR 4,931,250) |
| Founder-led firm needing a product owner in name | Roadmap ownership, pricing, what gets killed | Which of that roadmap agents now absorb | Standing decisions, monthly, in writing | Retained around the sprint and agent prices above, never a day rate |
How to hire one: fractional, embedded, or a sprint
There is no published rate card for fractional product management in Sri Lanka. We looked, and the fractional marketplaces that publish benchmarks price the American market and would not open to us for a citation, so we are not going to quote a number we cannot show you. The nearest local anchor is what a salaried software product manager earns here, and it is thin: PayScale’s Sri Lankan page for software product managers reports an average of LKR 2,441,676 a year, about US$7,427, from a sample of just 10 profiles last updated in January 2024. Ten profiles is a hint, not a benchmark, and we would not price anything off it.
So here is Janaka’s own shape instead, published before you call. The first thirty-minute call is free, and the honest answer is sometimes that you should not buy anything yet. The smallest real unit is the US$2,300 mapping sprint: two video sessions, your three busiest workflows, staff hours counted against each, a written report inside 48 hours, credited in full against the recruitment fee if you hire within 30 days. Beyond that you are either hiring an agent at US$1,225 a month with a measured probation, or you hire a builder, an embedded engineer at a fixed engagement price rather than an hourly one. He does not sell day rates, and he leads every engagement the team takes rather than handing you to somebody you met once.
If you are still deciding whether a consultant is the right purchase at all, the categories are set against each other in AI agency vs AI consultancy vs freelancer.
What to do this week
Write down the next product decision your firm has to take, in one sentence, and put a name next to it. If the name is blank, that is your finding, and no tool fixes it.
Then take the three tasks your team repeats most this week and time them honestly. Multiply by the loaded cost of the person doing them, in rupees. Now read the two lists side by side and ask whether the product decision and the automation decision are actually the same decision, as they were for the eighteen-person firm above. If they are, buy them together, from someone who does both halves. If they are not, buy the cheaper one first.
Questions people actually ask.
What does an AI product management consultant do?
An AI product management consultant decides what a product should do next and which of the surrounding work can safely be handed to AI agents, then writes both decisions down as a specification a team can build against. In practice that means counting the hours a workflow really costs, ranking what to build and what to kill, designing who approves an agent’s output before it leaves the building, and staying on the engagement while it ships.
What is the difference between a product manager and an AI consultant?
A product manager owns what gets built and is judged on the outcome, which Silicon Valley Product Group sums up as ensuring what gets built is both valuable and viable. An AI consultant owns which repeatable work is handed to software, in what order and under whose approval, so the two roles answer different questions and only overlap when the same evidence, usually a count of staff hours, settles both.
Can I hire a fractional product manager in Sri Lanka?
Yes, though nobody in the Sri Lankan market publishes a fractional product management rate card, so you will be quoted after a call in almost every case. Janaka’s own version is priced rather than hourly: a free thirty-minute call, a US$2,300 mapping sprint (LKR 756,125), and after that either an agent at US$1,225 a month or an embedded engineer on a fixed three, six or twelve-month engagement.
Who is Janaka Ediriweera?
Janaka Ediriweera is the Principal AI and Product Management Consultant at specshop.dev and the founder of both specshop.dev and Humanised Technology Solutions (Pvt) Ltd, based in Colombo 5. Twelve years, 120+ products shipped across five countries, nine years of spec-first delivery, and more than 150 startups trained at RMIT Activator in Melbourne.
How much does a product management consultant cost in Colombo?
There is no published category rate in Colombo, so the only honest figures are individual published prices: Janaka’s start at a free thirty-minute call and a US$2,300 workflow mapping sprint, about LKR 756,125, with agents at US$1,225 a month and embedded engineers from US$15,000 for three months. For context on the employed side, PayScale’s Sri Lankan software product manager page reports LKR 2,441,676 a year from a sample of only 10 profiles, which is too thin to price a consultant against.
Sources
- RMIT Activator, RMIT University, 2026 — “RMIT’s centre for entrepreneurship”, working with “aspiring founders, startups and organisations who are driven to create meaningful change”, with the stated mission “to deliver impact through entrepreneurship”. Fetched 10 September 2026.
- Product Management: Start Here, Silicon Valley Product Group — “you’re responsible for ensuring that what gets built is both valuable and viable”, and “feature teams deliver output, but product teams deliver outcomes”. Fetched 10 September 2026.
- Product Manager, Software Salary in Sri Lanka, PayScale — average LKR 2,441,676 a year, from 10 salary profiles, last updated 28 January 2024. A sample this small is indicative only. Fetched 10 September 2026.
- Daily Indicative USD/LKR Spot Exchange Rates, Central Bank of Sri Lanka, 2026 — the indicative rate of LKR 328.75 per US$1 on 10 September 2026, used for every rupee conversion in this post. Fetched 10 September 2026.