Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
This question does not appear to be about programming within the scope defined in the help center.
Closed 5 years ago.
Improve this question
Summary
We are start up and we provide software development services. We develop windows, web, services and mobile applications. We were aware of agile and we are scrum certified developers . We do user story based estimation and task planning. No issues.
Issue
We are approached by many small customers. Customers says very high level features or few words about the concept of their dream project. They asks for Effort Estimation and Cost Estimation. Mostly they are interested in Cost.
For each customer we did create the user stories and estimated the user stories and based on story points, we estimated the effort in days and we convert the days to the cost based on hourly rate. We involve the team of 3 or 4 people and get the estimation done. We spend at least 20 to 30 hours of team total time for estimation. (Team of 4 discussing for 5-6 hours)
The problem is that many customers would never turn back. We do not want to spend 20-30 hours of team effort. We don't want to use the exact user story estimation that we follow for contract signed project.
Question
What could be done in order to provide approximate estimate for small customers with small business?
I don't know there is a solution, other than to find 'better' customers. It sounds like you're doing it right to me. Non-technical customers often want you to spend 30min on the phone with them and then give them a price for the whole thing, so it's good you take the time over it properly. However then you often waste your time.
Maybe you need to say 'no' to customers who you don't think are serious. Or charge for the time spent doing highly skilled estimation work.
By 'better' customers I mean bigger companies, who are more experienced with software (and also probably have bigger budgets). The downside is more paperwork - you are much more 'free' dealing with small firms but also more at risk.
You don't have to stick to a fixed-price contract, for requirements that are vague you should look at doing time and materials.
Basically you need to spread the risk of the cost of overrun between you and the client.
A hybrid option might be to do some T&M proof of concept work then fixed price for the rest when you understand it better.
Alternatively, if you client has a pot of money then use your agile strengths to work with the customer to incrementally deliver functionality until they run out of money.
"We do not want to spend 20-30 hours of team effort."
Then don't.
If your estimating method is too costly, stop doing it.
"Customers ... are interested in Cost."
Then get them a cost more quickly. Do less work. Don't use a team of 3-4 for 20-30 hours. Have one person do it quickly.
One person can create a spreadsheet with stories, story points, priority, hours and cost. That's the project backlog for Scrum. That's the estimate. That's enough to start a conversation. It's not a fixed-price estimate.
A simple spreadsheet with stories, story points, price and priority is all you need. Then you can work with the customer to adjust priorities to determine how many of those stories they can actually afford to buy.
If they want a fixed price, you simply need to review each story summary to see if the points are right. You already have the spreadsheet, and the priorities, and the formula to compute price.
The unfortunate short answer is you are going to have to significantly reduce your estimation costs. The only way to do that is to reduce the number of people to one and use a formula approach.
Take a best guess. Put reasonable assumptions in the contract. If you do a decent job, some will be a little high, some will be a little low and it will average out in the end. If you are off by too much, track changes within the project and charge for them. The key will be in the assumptions placed in the contract, usually in the form of a statement of work.
This is normal business, from small projects to enterprise projects. Just a matter of scale. It doesn't have to be done this way, but it often is.
This reminds me of the challenge between time, money, people and overall quality. Some people can see this easily and others may struggle with the idea. Part of the key point here is to understand is what kind of expectations do you want to set and what kind of leeway do you have with the customer. For example, how are bugs in the software or overall support factored into the project.
You may want to consider how much work do you want to do upfront and on what kind of scale are you spending the 20-30 hours estimating a project. Comparing the cost of that much time spent generating an estimate, which # $40/hour is $800-1,200 by the way, to what the revenue from the project would be is something to consider. If the entire project is $400, then was it worth spending twice that coming up with an estimate? On the flip side, for million dollar projects, it may well make sense to spend that kind of money.
My suggestion would be to see if there is a cookie-cutter approach that could be taken for their projects so that there isn't as much variability to the projects if that is possible.
Theres two sides of estimation, first creating the initial 'ball park' figure this should be done relatively quickly, it should be emphasised its a ball park figure and its the start of your conversation with your customer, its not a contract. Second you do your more detailed team based estimation.
This is how this consulting company does it - Ball Park Estimating
Create a template spreadsheet with times and costs for typical pieces of work, look at the initial requirement and update your template. Start the conversation with this is the ball park but we will need to work together to confirm a final price and get a more accurate estimate. This more accurate estimation process will take x hours and cost x dollars.
Related
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
This question does not appear to be about programming within the scope defined in the help center.
Closed 5 years ago.
Improve this question
I understand that on the first part of your sprint planning, you would size the stories for example using story points and poker cards and then you estimate how many stories you can tackle on the next sprint. So let´s say you agree on working on 3 stories with a total combined of 18 story points.
Then you go ahead with the second part of the sprint planning and you start breaking those stories down into tasks. These tasks are now estimated using actual hours.
I have two questions:
1) How do you estimate the tasks in hours? do you use again poker estimating but with number of hours this time? how does the team agree on the number of hours for each task.
2) Once you estimate all tasks for the 3 stories you agreed for the sprint, you find out that the number of total hours combined for all tasks of all 3 stories is 90 hours for example. If your actual team capacity in hours is 75, how do you adjust your initial commitment of delivering 3 stories now that you realize that you don´t actually have the time to do it? Do you go back to your Product Owner (if he´s not there anymore) and tell them that you will be delivering 2 stories instead or how do you go about this?
Thanks a lot in advance for your help!
The tasks are not really estimated in hours but ideal hours. It's really hard to predict how many ideal hours will be available in a week, and it's generally not a good idea at all to infer the capacity on a sprint based on hour estimates. See for example this Scrum Alliance blog article
Story points and task hour comparison can be thought of as the comparison of the weight and height of an elephant. In general, a taller elephant may be heavier than a shorter one, but this is not always the case. There is no biological proof of a weight-versus-height formula, despite the common perception that more height means more weight. The same explanation applies to story points versus task hours: In general, a more complex user story (higher points) should take more hours to complete, but there are always exceptions.
Normally, tasks are tackled by a single team member (or a pair), and thus they are estimated by them, and not by the whole team.
Furthermore, tasks are re-estimated daily: the number we look for is the number of ideal hours left to complete the task and not the total amount. So, it is a number that can go up or down, or remain constant.
Also, tasks can be added or removed during the sprint. It's actually pretty common to discover that initial plans change. It doesn't matter, as long as the total number of hours planned represents the best current estimate of what's left to do -- it's needed for burndown charts.
In conclusion, don't mind the hours. Use them to monitor progress during the sprint but not to determine capacity. This is what points are for and the two parameters are not interchangeable as it would seem.
In my experience, breaking into and estimating tasks is very useful and often underated. The idea is that the team become more and more accurate as time goes on through their sprints and they inspect and adapt. I have observed teams getting to the point where they end up completing everything in the sprint except for unforseen blockers (eg. Server down time). The idea is that the team calculate their capacity in (hours) and then calculate the amount of tasks they need to complete ( in hours). They can use this as a guide to make a commitment. There is a more detailed explanation here: http://www.pashunconsulting.co.uk/team_commitment_blog.html
You can also use story points to make a commitment but the general principle is that Story points are better for long term estimation (ie. for estimating when the project will deliver). Agile Estimating and Planning is a good book for this kind of stuff, as is the Scrum Mega Pack.
I guess an issue come from the assumption that you are able to deliver 18 points within a sprint. What looks not right when you do an estimation in hours later. So, commit to fewer number of story points initially and after several sprints you would be able to know your actual velocity in story points per sprint.
While breaking stories into tasks is helpful, it is not very useful to estimate those tasks.
The reason is that you will spend a lot of time doing these micro estimates and they will generally be inaccurate.
Besides, adding concrete hourly estimates forces work to fit in those estimates.
If you underestimated you might procrastinate, if you overestimated you might skip on reviews, unit tests or other quality practices just to fit the timeline.
So in general I like to stay with one estimate level -> the one that gives the velocity.
Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 5 years ago.
Improve this question
Im the context of pair-programming ...how do you estimate? A 5 point story ...split in to 3 tasks...each task being swarmed by 2 members. Does it mean it mean it is finishable in half the time?
You estimate using points. The number of points achievable by a pair is called velocity. Check this: http://en.wikipedia.org/Planning_poker
I always stress story sizing over estimating. If you can minimize the variation in size between stories, then estimates become much less interesting. That is, estimation activities lose their value when the difference in size between any two stories is small, which helps fast teams recoup the cost of estimation and put it toward something productive (like building product).
With that in mind, I would suggest proactively pruning the backlog and splitting five point stories into smaller ones (still thin vertical slices) whenever possible. Until your team is experienced, I'd suggest you continue to have estimation parties, but prompt a default estimate of 1 point to each card, with a quick consensus check or discussion as to why any justify a bump to a 2 or 3. For anything that's clearly bigger than a 3, I'd suggest challenging that one of two problems is present: the baseline value of "1" is too small or the story should be split (either made more specific or tracked as an epic).
As the team establishes a decent velocity, the activity can hopefully shift its mindset from estimation to mere story vetting. That is, the question during planning of "how big is this story?" becomes "is this story abnormal?" The latter question takes significantly less time to answer.
Most Agile methodologies suggest that you should estimate in points. However, there are many successful teams out there - including several advanced and highly productive Kanban teams - who estimate in hours. Points come with their own games, perverse incentives and problems. YMMV. Anyway...
I've heard figures of 25% more dev-hours for a pair completing a task. So, the task would be finished in 62.5% of the time, using two developers instead of one. However, the quality of and knowledge-sharing often increases too. Since bugs = rework and rework takes longer than doing it the right way the first time, pair-programming usually pays for itself. This differs for different tasks and levels of skill, eg: simple bug fixes, novice programmers, etc.
In my experience 2/3 of the time is a pretty good ballpark figure. It's longer than 1/2 but less than it would be with just 1 person.
Sallyann Freudenberg is a good name to look up regarding pair programming research. You could also check out the refs on the Wikipedia page:
http://en.wikipedia.org/wiki/Pair_programming
The figure is mostly borne out by the data in this report by Alistair Cockburn and Laurie Williams: http://collaboration.csc.ncsu.edu/laurie/Papers/XPSardinia.PDF
Pair estimates how long it is going to take them. Any estimate is based on experience - the longer experience team has with the project, with technology and with working together in pairs the better the estimates will be. Any arbitrary percentages like add 25% for pairs etc. are of any use only at the very beginning - with new project and new team on new technology - where you have nothing else to base estimation on yet. As soon as experience starts building the estimates will improve.
Remember though, that they are just this - estimates, that is our best guess of the future derived with the help of experience and knowledge from our understanding of the present. It's like weather forecast - the more data we have, the more experience forecasters have the better it is, but it is just a prediction, not reality.
Which is why points are so great, because they help you estimate one parameter you can - how "big" the tasks at hand are.
1) A simple, concrete way to think about estimating units is the number of check-ins it would take to complete the story.
If you're doing TDD, continuous integration and refactoring you'll be working in small, chunks of work, keeping the build green and checking in regularly so under those conditions single check-ins can be a meaningful unit of estimation.
2) Alternatively, think about blocks of uninterrupted pairing time in a day e.g. after stand-up to coffee break, after coffee break to lunch, after lunch to mid-afternoon break, mid-afternoon to going home time - say 4 periods a day.... so say 4 units is a single day. That gives you a bound on what you could expect to do in an interation...
Personally I go for the number of check-ins, because I can sketch out roughly the tasks involved and get an idea of check-in numbers.
The great thing about number of check-ins is that it doesn't matter if you're pairing or not - you're just tracking what you can get done.
As it currently stands, this question is not a good fit for our Q&A format. We expect answers to be supported by facts, references, or expertise, but this question will likely solicit debate, arguments, polling, or extended discussion. If you feel that this question can be improved and possibly reopened, visit the help center for guidance.
Closed 10 years ago.
When working on fixed price software development projects, I frequently find myself having to estimate the total number of hours a project will take after the price is set, but before the work is started (or VERY early on in the development). Unfortunately, these types of projects are best developed using an iterative/agile method, which means that we don’t (and really can’t) do a complete up-front design.
In a typical scenario, we would have a contract that has X features and Y dollars. After contracting, the engineering department would then need to estimate the number of hours required to complete the X features. There are several possible reasons to need this information up front, including:
• The Y dollars translates to Z hours available, so we have to make sure that time(X)<=Z, perhaps by reducing the scope of X.
• The delivery date is set, and so we have to assign the appropriate resources to meet that date.
Kelly Waters has an interesting take on estimating agile here: http://www.agile-software-development.com/2009/04/agile-estimating.html Unfortunately, these are estimations of difficulty, using a points system, and do not translate to hours.
It seems to me that we need to be able to do one of two things:
• Obtain contracts that have a huge amount of flexibility in them to accommodate an agile development process.
• Figure out how to provide reasonably accurate up-front estimates for features that have not yet been designed.
The first option is of course not an option in most cases.
Does anyone have any advice/guidance on how to generate up-front estimates in an agile development scenario?
Alternatively, does anyone see another option for solving our problem through some other process change?
I think every client wants to have at least an estimate of how much the implementation of a given number of feature will cost him. I don't agree with people that say that if your using agile than you can't do this. Agile can be adapted to the real world where clients want to know how much money they're gonna spend on a project, or at least have a rough idea.
So, there are at least two documented ways you can do this and both are described in the "Agile Estimating and Planning" book by Mike Cohn that i strongly advise everyone to read.
Before your project even starts do the exercise of breaking down your stories in tasks and estimate each home in hours. Do the budget math with those estimates. Keep in mind that these estimates will only be used to reach the estimate time/budget. When the project starts the team should be responsible for estimating and creating the tasks like normal.
Use historical data. If the same team has worked before on a project with similar technology then you can use the past team velocity to estimate the project cost.
Again, for more details on how to do this read the referenced book.
Big Upfront Estimation is just Big Upfront Design with $$$ attached
You don't, that would go completely against the entire paradigm of agile.
Agile can be about Fixed Costs
What you can do is set a date and then work torwards it in iterations/sprints and let the product owner(s) decided what is important to get done by that date. In this way a Sprint of 1 or 2 weeks is a Fixed Cost the same way it would be in an Big Up Front Design would be, it is just a smaller fixed cost.
The entire premise behind agile methodologies is to do away with arbitrary deadlines and the death march that they become because change is not accounted for in the contracts and deadlines. SCRUM works, is a great starting point to build a methodology on, read about it and do what it says at least as a starting point.
Short Sprints with customer feedback and approvals
Give you a starting point to refine your future estimations upon, and helps you gain trust from the product owner and customer very quickly, because they see their most important concerns delivered in a very short amount of time.
Fixed price in Agile gives you a number of iterations you can run for a given team size. The point of Agile is then to maximize the value you can get during these iterations. Agile is about scope management.
So, you actually shouldn't do upfront estimations. This would mean fixed scope which is the just the opposite of Agile or anti-Agile if you prefer.
Agile doesn't work like this, you need a new kind of contract, as pointed out in another answer. See for example 10 Agile Contracts and/or google.
Instead of aiming for a contract for the "entire" project, you should create a contract that just runs for a month or two.
Set some very low goals, perhaps only two or three features.
Explain to the client this is a discovery phase of the project. Without this phase you cannot give them an estimate for the entire scope. It would be in no ones interest to give them an inaccurate estimate.
The client benefits in the following ways:
Lower upfront cost (only a fraction of the total price). If things go tits up there is a limit on their financial exposure (sensible given that most software projects fail).
Have working software within a couple of months that may immediately solve some of their problems
Have working software to provoke thought/conversation about requirements
Client will discover more about what they actually want
Client gets to see how well you work. If they don't like it they can look elsewhere.
"you don't, that would go completely against the entire paradigm of agile": I think this is quite incorrect. Agile is common sense based and nobody would invest in project if they had no idea how much it would roughly cost: they understand they may have to cut scope to meet deadlines or increase budget and/or extend deadlines to deliver the whole scope. On my company projects, we estimate projects by comparing their size and are also estimating epics using poker planning. Using the cone of uncertainty, and without trading off quality, we are about maximum 50% off on our estimates prior to starting the first sprint compared to the reality. And we will improve as we build up our Agile expertise (on accuracy and time to get first estimates).
You can't expect marketing to sponsor projects for 3 sprints (so up to 3 months) to have an idea of how much it would cost.
If you estimate the backlog items in story points (which is discussed in the Kelly Waters article you link to) then the question becomes how many story points do you expect your team to deliver in a sprint. If your team has been working together before, then you should be able to take a guess at this (perhaps with upper and lower bounds to indicate a confidence range).
If the team haven't worked together before, then you can take a few well-understood stories and break them down into detailed tasks. This will give you a man-hours number and you can compare this against your story point estimates to try and predict your velocity. Again, a confidence range of upper and lower values is probably appropriate.
Let's suppose your user stories add up to 150 points, and you predict that your team can deliver 15-20 story points per month, then the work will take 7.5 - 10 months (through simple division).
The advantage of this approach is that you can measure the team's actual velocity, and compare it to the plan. Re-estimating you timelines should also be pretty easy, e.g. if you discover after a couple of sprints that the team's velocity is actually only 10 story points per month, then you can easily predict what your revised timelines will be.
Do bear in mind that team velocity does fluctuate for a number of reasons. It is typically lower at the start of a project when the team is still forming. Also the team is most likely concentrating on infrastructure concerns at this point rather than delivering functionality.
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
This question does not appear to be about programming within the scope defined in the help center.
Closed 5 years ago.
Improve this question
I've been learning Scrum and trying a tool called Acunote for use with it. My question is about two fields I have there, for each task. They're "estimate" and "remaining". What unit should I use for those? Do I use Story Points? What about the remaining? For example I have a task that will take 10 units, let's say. I fill the remaining at the end of day with how many "units" I believe it will take me to complete?
Thanks!
I have a few suggestions for you:
If you are new to scrum; use a whiteboard and don't get bogged down in a particular tool's semantics; it will hamper your learning and adoption.
Break your stories small enough so that you don't have to create and estimate Tasks.
Don't do anything with hours, it is a waste of time to estimate at that level.
Burn down story points.
It is all too easy and common for teams to "think" they are on track because Tasks are being completed and burned down. Then they get to the end of the sprint and find that 5 stories are all 90% done and nothing is completed. If you burn down stories you are actually tracking deliverable business value and not just an arbitrary amount of developer-junk.
As always, my first advice would be to not use a tool when adopting/learning Scrum (I start to be tired to repeat the same thing over and over :). Instead, start with the simplest thing that could possibly work (a spreadsheet for the Product Backlog, a white-board and post-it notes for the Sprint Backlog). The rationale behind this is that you want to learn and master Scrum, not a tool. So don't let a tool tell you how to do Scrum and drive the process.
Then, regarding the question, there are 2 schools of thought: 1. what Scrum says in theory, 2. what some people do in practice.
In theory, Scrum has two levels of estimation: one for work (Tasks) to be completed within the current Sprint and one for more distant Product Backlog Items (PBIs). At the Product Backlog level, items (the "what" is being built) should be estimated in Story/T-Shirt/Unit-less points which have a low degree of precision. This approach avoids "analysis paralysis" pitfalls and accurately reflects the general uncertainty surrounding the work in question. At the Sprint Backlog level, items are beaked down into tasks (the "how" a PBI will be achieved) that are estimated in hours. A separate estimation scheme is appropriate because Tasks describe granular work (usually on the order of a few hours, never more than 16h). In fact, Scrum recommends using "ideal engineering hours" for Task-level estimates.
In practice, some people don't estimate in hours because burning down hours doesn't show "real" progress, which isn't false, and they prefer to burn down Story Points (which really means an item is done or not, it's more binary).
While I understand the "spirit" of the later approach, I don't apply it and stick with the theory. Actually, for the reasons previously mentioned, estimating in hours does make sense to me and I actually find that it gives better "control" of the Scrum empirical process during a Sprint (at the end of each day, you should update the estimated remaining work regardless of the actual time spent and this is easier with hours).
Moreover, I don't like the drawback of having only small stories (which can be seen as waste too) but like when a team identifies clearly what has to be done within a Sprint (this is good for transparency and helps the Product Owner to understand the real amount of work too, especially "quality oriented" tasks).
Finally, I think that you can avoid the pitfalls mentioned by DancesWithBamboo with hours too. Just stay vigilant and:
Always focus on the most important PBIs (and related tasks) first.
Pay a special attention to non-finished tasks, they should keep moving on the white board (if you are using columns to represent steps like for example "todo", "in progress", "to be verified", "done"); a non moving task is a smell.
Don't start a new item before the previous is done.
So, in my opinion, it is possible to use hours and to avoid the "nothing done" at the end of the Sprint syndrome. Just use your brain (Scrum and/or any tool won't replace it, luckily for us).
Having that said, and if you don't throw your tool away, the questions to answer are: what do you want to show on the burndown (points or hours depending on if you breakdown the work into tasks or not, I gave you my point of view) and what field does Acunote use to draw the burndown (i.e. where should I update the estimation of the remaining work). If you choose points and don't use tasks, it wouldn't make sense to update remaining work unless it's totally done IMO (a PBI is done, or not).
IMHO you shouldn't use remaining for the SCRUM points, because the points should be really subjective and you probably can't say how far you have gone.
I would recommend that you break the task into smaller ones (these ones would be the steps you need to implement the features) and then estimate them into hours. This way you can easily track the progress of the feature
Use hours for both the initial estimate and the remaining time. Tasks are usually estimated in hours.
You can use Scrum point - or any other unit - to estimate the backlog items.
I would not bother with remaining. A story or a task is boolean (either done or not done).
In our team we started to accept only tasks that are expected to have less than a day. That way no team member should be working on the same task on two consecutive daily scrums. The third day for sure rings alarm bells!
Also breaking up in task is easy and fast, because it doesn't take much estimating. Is it less than a day: OK, otherwise break it down.
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
This question does not appear to be about programming within the scope defined in the help center.
Closed 8 years ago.
Improve this question
How do you charge your customer in a project using agile methodology?
Per hour? Then a great deal of trust has to been established before the project.
Per iteration? There's gonna be a lot of budget decisions, which can take time.
Per project? How can you do that when you don't know the scope? The very essence of agile is to not write a big upfront design/specification.
You charge your customer on the base of the terms defined by your contract that will be slightly different from a traditional fixed bid contract. Let's call that an Agile contract.
Some options are discussed by Alistair Cockburn in Agile contracts.
Another great resource is 10 Contracts for your next Agile Software Project by Peter Stevens.
Mary Poppendieck also has great material on this topic. See agilecontracts, agilecontractsworkshop, Contracts Excerpt From Lean Software Development, Lean Contracts. More here.
Short answer is, you won't. There are a few services companies that are making headway doing it, but it is a difficult path. Your ability to sell the methodology and convince the customer you can deliver will be high.
Customers don't want to risk paying for a solution that will never be delivered.
Typical approaches to this problem are to put "will not exceed" cost. However, if you can't control scope, you are the one taking all of the risk.
In short, you are looking for customers that would have signed up for T&M (time and materials) contracts before Agile became the latest fad (I'm part of that fad, but it is just one in a long line of development processes. Aspects of it will continue to grow and some permutation of it will have a different name in a few years).
If your customer has already bought into the use of an agile methodology, then you have a reasonable framework for negotiating price per iteration. For example, you know:
How long the iteration will be.
How many people will be committed to work on the iteration (and their rates).
An approximate scope of work.
A process for delivery and acceptance.
That's a lot more evidence on which to base pricing decisions than is available for most fixed price contracts.
If the agile methodology is purely an internal development process that doesn't involve the customer, then it's unlikely to have much effect on the pricing negotiation between the supplier and the customer. There is an argument that says that a process that doesn't involve the customer in setting scope and expectations at least once per iteration isn't agile at all.
Regarding Mark's comment, there is a very common pricing model based around fixed price contracts with loosely defined scope and optimistic schedules. A common outcome is that both supplier and customer find that their initial optimism was misplaced. Both end up negotiating from weak positions over the things that really matter to them, and both end up unhappy.
Some of the techniques that work well in managing this type of contract are very similar to those used in managing agile contracts (frequent delivery, horse-trading on scope, priorities & price, frank communication, ...) the main difference is that these aren't built into the original agreement, and the contract may not be flexible enough to accommodate all of them.
My 2c as a non-agile practitioner...in a quest to know more...
If you are doing a specific project for a customer, you will need to know the scope of the project to provide a cost and a timescale. The cost of producing this scope of work is more often than not part of the discovery of the project, you either take a hit on this to get the work or charge for this (explicitly or implicitly) In this case, a project cost can be worked out and agreed. Im this case the project is usually broken up into various stages.
Although agile may be iterative and not require a full specification; a goal, at least is certainly required. There must be some form of basic specification/requirement. It may be that you need to break the project up into smaller goals and apply costs accordingly.
The iterations I suspect are more to do with the development methodoology, ie to achieve the goals?
If there is not enough specification to produce a definitively accurate cost, I would say that a "estimate" should be given but work should be charged at an hourly rate as I would assume that there would be greater changes in the decisions made on the project over each iteration.
I've seen it work well when approached in 2 phases:
Phase 1) Inception (timeboxed)
A timeboxed inception period with the client to scope the project. (A one month intense inception for a project estimated to last a year is about right.)
Outputs to this phase are a full backlog of sized user stories, an estimation of flow rate per dev pair, and parameters to estimate project costs based on the number of developers and overheads of having larger teams.
The inception provides a useful budget estimation that can be tracked throughout phase 2, a clear shared vision for both client and supplier, plus the option for either side to walk away. This isn't upfront design, the stories have just enough detail for a lead dev / tester to assign relative sizes.
Phase 2) Delivery (time and materials)
A delivery contract based on time and materials, with budget estimates based on the outputs from the inception phase. The trust built up in phase 1 is vital to making this work. Because phase 1 delivered relative sizing of the entire backlog, by regularly measuring actual flow it is possible to easily and frequently report projected flow rate for the rest of the backlog with increasingly accurate estimates of budget and delivery date. The supplier should be contracted to report these stats at least every 2 weeks, with the option for the client to walk away at any point.
By paying for time and materials, the client is free to change the scope and proirity of the backlog, and is therefore in control of the budget. They are encoraged to prioritise their highest priority / highest risk stories first, and by allowing them to walk away whenever they like they should experience a positive return on investment at all times.